panfrost: Move pandecode into lib/
authorAlyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Wed, 5 Aug 2020 20:25:52 +0000 (16:25 -0400)
committerMarge Bot <eric+marge@anholt.net>
Thu, 6 Aug 2020 23:54:24 +0000 (23:54 +0000)
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6196>

13 files changed:
src/panfrost/bifrost/test/bi_submit.c
src/panfrost/bifrost/test/bit.h
src/panfrost/lib/decode.c [new file with mode: 0644]
src/panfrost/lib/decode.h [new file with mode: 0644]
src/panfrost/lib/decode_common.c [new file with mode: 0644]
src/panfrost/lib/meson.build
src/panfrost/lib/wrap.h [new file with mode: 0644]
src/panfrost/meson.build
src/panfrost/pandecode/common.c [deleted file]
src/panfrost/pandecode/decode.c [deleted file]
src/panfrost/pandecode/decode.h [deleted file]
src/panfrost/pandecode/meson.build [deleted file]
src/panfrost/pandecode/public.h [deleted file]

index 38539c09cbc827ef7b41d0682919525dfb861762..f11fea3bd2c7de58fcd1c64b840fda6f57f6025b 100644 (file)
@@ -27,7 +27,7 @@
 #include "bit.h"
 #include "panfrost/lib/decode.h"
 #include "drm-uapi/panfrost_drm.h"
-#include "panfrost/encoder/pan_encoder.h"
+#include "panfrost/lib/pan_encoder.h"
 
 /* Standalone compiler tests submitting jobs directly to the hardware. Uses the
  * `bit` prefix for `BIfrost Tests` and because bit sounds wicked cool. */
index 50bbf00493511dea0320811527d96dfa1685717b..74d988a90eedae79d107afbf83162eac4f5dbfda 100644 (file)
@@ -28,8 +28,8 @@
 #define __BIFROST_TEST_H
 
 #include "panfrost/include/panfrost-job.h"
-#include "panfrost/encoder/pan_device.h"
-#include "panfrost/encoder/pan_bo.h"
+#include "panfrost/lib/pan_device.h"
+#include "panfrost/lib/pan_bo.h"
 #include "bifrost_compile.h"
 #include "bifrost/compiler.h"
 
diff --git a/src/panfrost/lib/decode.c b/src/panfrost/lib/decode.c
new file mode 100644 (file)
index 0000000..f3ce25e
--- /dev/null
@@ -0,0 +1,3421 @@
+/*
+ * Copyright (C) 2017-2019 Alyssa Rosenzweig
+ * Copyright (C) 2017-2019 Connor Abbott
+ * Copyright (C) 2019 Collabora, Ltd.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <panfrost-job.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <memory.h>
+#include <stdbool.h>
+#include <stdarg.h>
+#include <ctype.h>
+#include "decode.h"
+#include "util/macros.h"
+#include "util/u_math.h"
+
+#include "midgard/disassemble.h"
+#include "bifrost/disassemble.h"
+
+#include "pan_encoder.h"
+
+static void pandecode_swizzle(unsigned swizzle, enum mali_format format);
+
+#define MEMORY_PROP(obj, p) {\
+        if (obj->p) { \
+                char *a = pointer_as_memory_reference(obj->p); \
+                pandecode_prop("%s = %s", #p, a); \
+                free(a); \
+        } \
+}
+
+#define MEMORY_PROP_DIR(obj, p) {\
+        if (obj.p) { \
+                char *a = pointer_as_memory_reference(obj.p); \
+                pandecode_prop("%s = %s", #p, a); \
+                free(a); \
+        } \
+}
+
+FILE *pandecode_dump_stream;
+
+/* Semantic logging type.
+ *
+ * Raw: for raw messages to be printed as is.
+ * Message: for helpful information to be commented out in replays.
+ * Property: for properties of a struct
+ *
+ * Use one of pandecode_log, pandecode_msg, or pandecode_prop as syntax sugar.
+ */
+
+enum pandecode_log_type {
+        PANDECODE_RAW,
+        PANDECODE_MESSAGE,
+        PANDECODE_PROPERTY
+};
+
+#define pandecode_log(...)  pandecode_log_typed(PANDECODE_RAW,      __VA_ARGS__)
+#define pandecode_msg(...)  pandecode_log_typed(PANDECODE_MESSAGE,  __VA_ARGS__)
+#define pandecode_prop(...) pandecode_log_typed(PANDECODE_PROPERTY, __VA_ARGS__)
+
+unsigned pandecode_indent = 0;
+
+static void
+pandecode_make_indent(void)
+{
+        for (unsigned i = 0; i < pandecode_indent; ++i)
+                fprintf(pandecode_dump_stream, "    ");
+}
+
+static void
+pandecode_log_typed(enum pandecode_log_type type, const char *format, ...)
+{
+        va_list ap;
+
+        pandecode_make_indent();
+
+        if (type == PANDECODE_MESSAGE)
+                fprintf(pandecode_dump_stream, "// ");
+        else if (type == PANDECODE_PROPERTY)
+                fprintf(pandecode_dump_stream, ".");
+
+        va_start(ap, format);
+        vfprintf(pandecode_dump_stream, format, ap);
+        va_end(ap);
+
+        if (type == PANDECODE_PROPERTY)
+                fprintf(pandecode_dump_stream, ",\n");
+}
+
+static void
+pandecode_log_cont(const char *format, ...)
+{
+        va_list ap;
+
+        va_start(ap, format);
+        vfprintf(pandecode_dump_stream, format, ap);
+        va_end(ap);
+}
+
+/* To check for memory safety issues, validates that the given pointer in GPU
+ * memory is valid, containing at least sz bytes. The goal is to eliminate
+ * GPU-side memory bugs (NULL pointer dereferences, buffer overflows, or buffer
+ * overruns) by statically validating pointers.
+ */
+
+static void
+pandecode_validate_buffer(mali_ptr addr, size_t sz)
+{
+        if (!addr) {
+                pandecode_msg("XXX: null pointer deref");
+                return;
+        }
+
+        /* Find a BO */
+
+        struct pandecode_mapped_memory *bo =
+                pandecode_find_mapped_gpu_mem_containing(addr);
+
+        if (!bo) {
+                pandecode_msg("XXX: invalid memory dereference\n");
+                return;
+        }
+
+        /* Bounds check */
+
+        unsigned offset = addr - bo->gpu_va;
+        unsigned total = offset + sz;
+
+        if (total > bo->length) {
+                pandecode_msg("XXX: buffer overrun. "
+                                "Chunk of size %zu at offset %d in buffer of size %zu. "
+                                "Overrun by %zu bytes. \n",
+                                sz, offset, bo->length, total - bo->length);
+                return;
+        }
+}
+
+struct pandecode_flag_info {
+        u64 flag;
+        const char *name;
+};
+
+static void
+pandecode_log_decoded_flags(const struct pandecode_flag_info *flag_info,
+                            u64 flags)
+{
+        bool decodable_flags_found = false;
+
+        for (int i = 0; flag_info[i].name; i++) {
+                if ((flags & flag_info[i].flag) != flag_info[i].flag)
+                        continue;
+
+                if (!decodable_flags_found) {
+                        decodable_flags_found = true;
+                } else {
+                        pandecode_log_cont(" | ");
+                }
+
+                pandecode_log_cont("%s", flag_info[i].name);
+
+                flags &= ~flag_info[i].flag;
+        }
+
+        if (decodable_flags_found) {
+                if (flags)
+                        pandecode_log_cont(" | 0x%" PRIx64, flags);
+        } else {
+                pandecode_log_cont("0x%" PRIx64, flags);
+        }
+}
+
+#define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
+static const struct pandecode_flag_info gl_enable_flag_info[] = {
+        FLAG_INFO(OCCLUSION_QUERY),
+        FLAG_INFO(OCCLUSION_PRECISE),
+        FLAG_INFO(FRONT_CCW_TOP),
+        FLAG_INFO(CULL_FACE_FRONT),
+        FLAG_INFO(CULL_FACE_BACK),
+        {}
+};
+#undef FLAG_INFO
+
+#define FLAG_INFO(flag) { MALI_CLEAR_##flag, "MALI_CLEAR_" #flag }
+static const struct pandecode_flag_info clear_flag_info[] = {
+        FLAG_INFO(FAST),
+        FLAG_INFO(SLOW),
+        FLAG_INFO(SLOW_STENCIL),
+        {}
+};
+#undef FLAG_INFO
+
+#define FLAG_INFO(flag) { MALI_MASK_##flag, "MALI_MASK_" #flag }
+static const struct pandecode_flag_info mask_flag_info[] = {
+        FLAG_INFO(R),
+        FLAG_INFO(G),
+        FLAG_INFO(B),
+        FLAG_INFO(A),
+        {}
+};
+#undef FLAG_INFO
+
+#define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
+static const struct pandecode_flag_info u3_flag_info[] = {
+        FLAG_INFO(HAS_MSAA),
+        FLAG_INFO(PER_SAMPLE),
+        FLAG_INFO(CAN_DISCARD),
+        FLAG_INFO(HAS_BLEND_SHADER),
+        FLAG_INFO(DEPTH_WRITEMASK),
+        FLAG_INFO(DEPTH_CLIP_NEAR),
+        FLAG_INFO(DEPTH_CLIP_FAR),
+        {}
+};
+
+static const struct pandecode_flag_info u4_flag_info[] = {
+        FLAG_INFO(NO_MSAA),
+        FLAG_INFO(NO_DITHER),
+        FLAG_INFO(DEPTH_RANGE_A),
+        FLAG_INFO(DEPTH_RANGE_B),
+        FLAG_INFO(STENCIL_TEST),
+        FLAG_INFO(ALPHA_TO_COVERAGE),
+        {}
+};
+#undef FLAG_INFO
+
+#define FLAG_INFO(flag) { MALI_MFBD_FORMAT_##flag, "MALI_MFBD_FORMAT_" #flag }
+static const struct pandecode_flag_info mfbd_fmt_flag_info[] = {
+        FLAG_INFO(SRGB),
+        {}
+};
+#undef FLAG_INFO
+
+#define FLAG_INFO(flag) { MALI_AFBC_##flag, "MALI_AFBC_" #flag }
+static const struct pandecode_flag_info afbc_fmt_flag_info[] = {
+        FLAG_INFO(YTR),
+        {}
+};
+#undef FLAG_INFO
+
+#define FLAG_INFO(flag) { MALI_EXTRA_##flag, "MALI_EXTRA_" #flag }
+static const struct pandecode_flag_info mfbd_extra_flag_hi_info[] = {
+        FLAG_INFO(PRESENT),
+        {}
+};
+#undef FLAG_INFO
+
+#define FLAG_INFO(flag) { MALI_EXTRA_##flag, "MALI_EXTRA_" #flag }
+static const struct pandecode_flag_info mfbd_extra_flag_lo_info[] = {
+        FLAG_INFO(ZS),
+        {}
+};
+#undef FLAG_INFO
+
+#define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
+static const struct pandecode_flag_info shader_midgard1_flag_lo_info [] = {
+        FLAG_INFO(WRITES_Z),
+        FLAG_INFO(EARLY_Z),
+        FLAG_INFO(READS_TILEBUFFER),
+        FLAG_INFO(WRITES_GLOBAL),
+        FLAG_INFO(READS_ZS),
+        {}
+};
+
+static const struct pandecode_flag_info shader_midgard1_flag_hi_info [] = {
+        FLAG_INFO(WRITES_S),
+        FLAG_INFO(SUPPRESS_INF_NAN),
+        {}
+};
+#undef FLAG_INFO
+
+#define FLAG_INFO(flag) { MALI_BIFROST_##flag, "MALI_BIFROST_" #flag }
+static const struct pandecode_flag_info shader_bifrost_info [] = {
+        FLAG_INFO(FULL_THREAD),
+        FLAG_INFO(EARLY_Z),
+        FLAG_INFO(FIRST_ATEST),
+        {}
+};
+
+#undef FLAG_INFO
+
+#define FLAG_INFO(flag) { MALI_MFBD_##flag, "MALI_MFBD_" #flag }
+static const struct pandecode_flag_info mfbd_flag_info [] = {
+        FLAG_INFO(DEPTH_WRITE),
+        FLAG_INFO(EXTRA),
+        {}
+};
+#undef FLAG_INFO
+
+#define FLAG_INFO(flag) { MALI_SAMP_##flag, "MALI_SAMP_" #flag }
+static const struct pandecode_flag_info sampler_flag_info [] = {
+        FLAG_INFO(MAG_NEAREST),
+        FLAG_INFO(MIN_NEAREST),
+        FLAG_INFO(MIP_LINEAR_1),
+        FLAG_INFO(MIP_LINEAR_2),
+        FLAG_INFO(NORM_COORDS),
+        {}
+};
+#undef FLAG_INFO
+
+#define FLAG_INFO(flag) { MALI_SFBD_FORMAT_##flag, "MALI_SFBD_FORMAT_" #flag }
+static const struct pandecode_flag_info sfbd_unk1_info [] = {
+        FLAG_INFO(MSAA_8),
+        FLAG_INFO(MSAA_A),
+        {}
+};
+#undef FLAG_INFO
+
+#define FLAG_INFO(flag) { MALI_SFBD_FORMAT_##flag, "MALI_SFBD_FORMAT_" #flag }
+static const struct pandecode_flag_info sfbd_unk2_info [] = {
+        FLAG_INFO(MSAA_B),
+        FLAG_INFO(SRGB),
+        {}
+};
+#undef FLAG_INFO
+
+#define DEFINE_CASE(name) case MALI_## name: return "MALI_" #name
+static char *pandecode_format(enum mali_format format)
+{
+        static char unk_format_str[10];
+
+        switch (format) {
+                DEFINE_CASE(ETC2_RGB8);
+                DEFINE_CASE(ETC2_R11_UNORM);
+                DEFINE_CASE(ETC2_RGBA8);
+                DEFINE_CASE(ETC2_RG11_UNORM);
+                DEFINE_CASE(ETC2_R11_SNORM);
+                DEFINE_CASE(ETC2_RG11_SNORM);
+                DEFINE_CASE(ETC2_RGB8A1);
+                DEFINE_CASE(NXR);
+                DEFINE_CASE(BC1_UNORM);
+                DEFINE_CASE(BC2_UNORM);
+                DEFINE_CASE(BC3_UNORM);
+                DEFINE_CASE(BC4_UNORM);
+                DEFINE_CASE(BC4_SNORM);
+                DEFINE_CASE(BC5_UNORM);
+                DEFINE_CASE(BC5_SNORM);
+                DEFINE_CASE(BC6H_UF16);
+                DEFINE_CASE(BC6H_SF16);
+                DEFINE_CASE(BC7_UNORM);
+                DEFINE_CASE(ASTC_3D_LDR);
+                DEFINE_CASE(ASTC_3D_HDR);
+                DEFINE_CASE(ASTC_2D_LDR);
+                DEFINE_CASE(ASTC_2D_HDR);
+                DEFINE_CASE(RGB565);
+                DEFINE_CASE(RGB5_X1_UNORM);
+                DEFINE_CASE(RGB5_A1_UNORM);
+                DEFINE_CASE(RGB10_A2_UNORM);
+                DEFINE_CASE(RGB10_A2_SNORM);
+                DEFINE_CASE(RGB10_A2UI);
+                DEFINE_CASE(RGB10_A2I);
+                DEFINE_CASE(RGB332_UNORM);
+                DEFINE_CASE(RGB233_UNORM);
+                DEFINE_CASE(Z24X8_UNORM);
+                DEFINE_CASE(R32_FIXED);
+                DEFINE_CASE(RG32_FIXED);
+                DEFINE_CASE(RGB32_FIXED);
+                DEFINE_CASE(RGBA32_FIXED);
+                DEFINE_CASE(R11F_G11F_B10F);
+                DEFINE_CASE(R9F_G9F_B9F_E5F);
+                DEFINE_CASE(VARYING_POS);
+                DEFINE_CASE(VARYING_DISCARD);
+
+                DEFINE_CASE(R8_SNORM);
+                DEFINE_CASE(R16_SNORM);
+                DEFINE_CASE(R32_SNORM);
+                DEFINE_CASE(RG8_SNORM);
+                DEFINE_CASE(RG16_SNORM);
+                DEFINE_CASE(RG32_SNORM);
+                DEFINE_CASE(RGB8_SNORM);
+                DEFINE_CASE(RGB16_SNORM);
+                DEFINE_CASE(RGB32_SNORM);
+                DEFINE_CASE(RGBA8_SNORM);
+                DEFINE_CASE(RGBA16_SNORM);
+                DEFINE_CASE(RGBA32_SNORM);
+
+                DEFINE_CASE(R8UI);
+                DEFINE_CASE(R16UI);
+                DEFINE_CASE(R32UI);
+                DEFINE_CASE(RG8UI);
+                DEFINE_CASE(RG16UI);
+                DEFINE_CASE(RG32UI);
+                DEFINE_CASE(RGB8UI);
+                DEFINE_CASE(RGB16UI);
+                DEFINE_CASE(RGB32UI);
+                DEFINE_CASE(RGBA8UI);
+                DEFINE_CASE(RGBA16UI);
+                DEFINE_CASE(RGBA32UI);
+
+                DEFINE_CASE(R8_UNORM);
+                DEFINE_CASE(R16_UNORM);
+                DEFINE_CASE(R32_UNORM);
+                DEFINE_CASE(R32F);
+                DEFINE_CASE(RG8_UNORM);
+                DEFINE_CASE(RG16_UNORM);
+                DEFINE_CASE(RG32_UNORM);
+                DEFINE_CASE(RG32F);
+                DEFINE_CASE(RGB8_UNORM);
+                DEFINE_CASE(RGB16_UNORM);
+                DEFINE_CASE(RGB32_UNORM);
+                DEFINE_CASE(RGB32F);
+                DEFINE_CASE(RGBA4_UNORM);
+                DEFINE_CASE(RGBA8_UNORM);
+                DEFINE_CASE(RGBA16_UNORM);
+                DEFINE_CASE(RGBA32_UNORM);
+                DEFINE_CASE(RGBA32F);
+
+                DEFINE_CASE(R8I);
+                DEFINE_CASE(R16I);
+                DEFINE_CASE(R32I);
+                DEFINE_CASE(RG8I);
+                DEFINE_CASE(R16F);
+                DEFINE_CASE(RG16I);
+                DEFINE_CASE(RG32I);
+                DEFINE_CASE(RG16F);
+                DEFINE_CASE(RGB8I);
+                DEFINE_CASE(RGB16I);
+                DEFINE_CASE(RGB32I);
+                DEFINE_CASE(RGB16F);
+                DEFINE_CASE(RGBA8I);
+                DEFINE_CASE(RGBA16I);
+                DEFINE_CASE(RGBA32I);
+                DEFINE_CASE(RGBA16F);
+
+                DEFINE_CASE(RGBA4);
+                DEFINE_CASE(RGBA8_2);
+                DEFINE_CASE(RGB10_A2_2);
+        default:
+                snprintf(unk_format_str, sizeof(unk_format_str), "MALI_0x%02x", format);
+                return unk_format_str;
+        }
+}
+
+#undef DEFINE_CASE
+
+static char *
+pandecode_job_type(enum mali_job_type type)
+{
+#define DEFINE_CASE(name) case JOB_TYPE_ ## name: return "JOB_TYPE_" #name
+
+        switch (type) {
+                DEFINE_CASE(NULL);
+                DEFINE_CASE(WRITE_VALUE);
+                DEFINE_CASE(CACHE_FLUSH);
+                DEFINE_CASE(COMPUTE);
+                DEFINE_CASE(VERTEX);
+                DEFINE_CASE(TILER);
+                DEFINE_CASE(FUSED);
+                DEFINE_CASE(FRAGMENT);
+
+        case JOB_NOT_STARTED:
+                return "NOT_STARTED";
+
+        default:
+                pandecode_log("Warning! Unknown job type %x\n", type);
+                return "!?!?!?";
+        }
+
+#undef DEFINE_CASE
+}
+
+static char *
+pandecode_draw_mode(enum mali_draw_mode mode)
+{
+#define DEFINE_CASE(name) case MALI_ ## name: return "MALI_" #name
+
+        switch (mode) {
+                DEFINE_CASE(DRAW_NONE);
+                DEFINE_CASE(POINTS);
+                DEFINE_CASE(LINES);
+                DEFINE_CASE(TRIANGLES);
+                DEFINE_CASE(TRIANGLE_STRIP);
+                DEFINE_CASE(TRIANGLE_FAN);
+                DEFINE_CASE(LINE_STRIP);
+                DEFINE_CASE(LINE_LOOP);
+                DEFINE_CASE(POLYGON);
+                DEFINE_CASE(QUADS);
+                DEFINE_CASE(QUAD_STRIP);
+
+        default:
+                pandecode_msg("XXX: invalid draw mode %X\n", mode);
+                return "";
+        }
+
+#undef DEFINE_CASE
+}
+
+#define DEFINE_CASE(name) case MALI_FUNC_ ## name: return "MALI_FUNC_" #name
+static char *
+pandecode_func(enum mali_func mode)
+{
+        switch (mode) {
+                DEFINE_CASE(NEVER);
+                DEFINE_CASE(LESS);
+                DEFINE_CASE(EQUAL);
+                DEFINE_CASE(LEQUAL);
+                DEFINE_CASE(GREATER);
+                DEFINE_CASE(NOTEQUAL);
+                DEFINE_CASE(GEQUAL);
+                DEFINE_CASE(ALWAYS);
+
+        default:
+                pandecode_msg("XXX: invalid func %X\n", mode);
+                return "";
+        }
+}
+#undef DEFINE_CASE
+
+#define DEFINE_CASE(name) case MALI_MSAA_ ## name: return "MALI_MSAA_" #name
+static char *
+pandecode_msaa_mode(enum mali_msaa_mode mode)
+{
+        switch (mode) {
+                DEFINE_CASE(SINGLE);
+                DEFINE_CASE(AVERAGE);
+                DEFINE_CASE(MULTIPLE);
+                DEFINE_CASE(LAYERED);
+        default:
+                unreachable("Impossible");
+                return "";
+        }
+}
+#undef DEFINE_CASE
+
+#define DEFINE_CASE(name) case MALI_STENCIL_ ## name: return "MALI_STENCIL_" #name
+static char *
+pandecode_stencil_op(enum mali_stencil_op op)
+{
+        switch (op) {
+                DEFINE_CASE(KEEP);
+                DEFINE_CASE(REPLACE);
+                DEFINE_CASE(ZERO);
+                DEFINE_CASE(INVERT);
+                DEFINE_CASE(INCR_WRAP);
+                DEFINE_CASE(DECR_WRAP);
+                DEFINE_CASE(INCR);
+                DEFINE_CASE(DECR);
+
+        default:
+                pandecode_msg("XXX: invalid stencil op %X\n", op);
+                return "";
+        }
+}
+
+#undef DEFINE_CASE
+
+static char *pandecode_attr_mode_short(enum mali_attr_mode mode)
+{
+        switch(mode) {
+                /* TODO: Combine to just "instanced" once this can be done
+                 * unambiguously in all known cases */
+        case MALI_ATTR_POT_DIVIDE:
+                return "instanced_pot";
+        case MALI_ATTR_MODULO:
+                return "instanced_mod";
+        case MALI_ATTR_NPOT_DIVIDE:
+                return "instanced_npot";
+        case MALI_ATTR_IMAGE:
+                return "image";
+        default:
+                pandecode_msg("XXX: invalid attribute mode %X\n", mode);
+                return "";
+        }
+}
+
+static const char *
+pandecode_special_record(uint64_t v, bool* attribute)
+{
+        switch(v) {
+        case MALI_ATTR_VERTEXID:
+                *attribute = true;
+                return "gl_VertexID";
+        case MALI_ATTR_INSTANCEID:
+                *attribute = true;
+                return "gl_InstanceID";
+        case MALI_VARYING_FRAG_COORD:
+                return "gl_FragCoord";
+        case MALI_VARYING_FRONT_FACING:
+                return "gl_FrontFacing";
+        case MALI_VARYING_POINT_COORD:
+                return "gl_PointCoord";
+        default:
+                pandecode_msg("XXX: invalid special record %" PRIx64 "\n", v);
+                return "";
+        }
+}
+
+#define DEFINE_CASE(name) case MALI_WRAP_## name: return "MALI_WRAP_" #name
+static char *
+pandecode_wrap_mode(enum mali_wrap_mode op)
+{
+        switch (op) {
+                DEFINE_CASE(REPEAT);
+                DEFINE_CASE(CLAMP_TO_EDGE);
+                DEFINE_CASE(CLAMP);
+                DEFINE_CASE(CLAMP_TO_BORDER);
+                DEFINE_CASE(MIRRORED_REPEAT);
+                DEFINE_CASE(MIRRORED_CLAMP_TO_EDGE);
+                DEFINE_CASE(MIRRORED_CLAMP);
+                DEFINE_CASE(MIRRORED_CLAMP_TO_BORDER);
+
+        default:
+                pandecode_msg("XXX: invalid wrap mode %X\n", op);
+                return "";
+        }
+}
+#undef DEFINE_CASE
+
+#define DEFINE_CASE(name) case MALI_BLOCK_## name: return "MALI_BLOCK_" #name
+static char *
+pandecode_block_format(enum mali_block_format fmt)
+{
+        switch (fmt) {
+                DEFINE_CASE(TILED);
+                DEFINE_CASE(UNKNOWN);
+                DEFINE_CASE(LINEAR);
+                DEFINE_CASE(AFBC);
+
+        default:
+                unreachable("Invalid case");
+        }
+}
+#undef DEFINE_CASE
+
+#define DEFINE_CASE(name) case MALI_EXCEPTION_ACCESS_## name: return ""#name
+static char *
+pandecode_exception_access(unsigned access)
+{
+        switch (access) {
+                DEFINE_CASE(NONE);
+                DEFINE_CASE(EXECUTE);
+                DEFINE_CASE(READ);
+                DEFINE_CASE(WRITE);
+
+        default:
+                unreachable("Invalid case");
+        }
+}
+#undef DEFINE_CASE
+
+/* Midgard's tiler descriptor is embedded within the
+ * larger FBD */
+
+static void
+pandecode_midgard_tiler_descriptor(
+                const struct midgard_tiler_descriptor *t,
+                unsigned width,
+                unsigned height,
+                bool is_fragment,
+                bool has_hierarchy)
+{
+        pandecode_log(".tiler = {\n");
+        pandecode_indent++;
+
+        if (t->hierarchy_mask == MALI_TILER_DISABLED)
+                pandecode_prop("hierarchy_mask = MALI_TILER_DISABLED");
+        else
+                pandecode_prop("hierarchy_mask = 0x%" PRIx16, t->hierarchy_mask);
+
+        /* We know this name from the kernel, but we never see it nonzero */
+
+        if (t->flags)
+                pandecode_msg("XXX: unexpected tiler flags 0x%" PRIx16, t->flags);
+
+        MEMORY_PROP(t, polygon_list);
+
+        /* The body is offset from the base of the polygon list */
+        //assert(t->polygon_list_body > t->polygon_list);
+        unsigned body_offset = t->polygon_list_body - t->polygon_list;
+
+        /* It needs to fit inside the reported size */
+        //assert(t->polygon_list_size >= body_offset);
+
+        /* Now that we've sanity checked, we'll try to calculate the sizes
+         * ourselves for comparison */
+
+        unsigned ref_header = panfrost_tiler_header_size(width, height, t->hierarchy_mask, has_hierarchy);
+        unsigned ref_size = panfrost_tiler_full_size(width, height, t->hierarchy_mask, has_hierarchy);
+
+        if (!((ref_header == body_offset) && (ref_size == t->polygon_list_size))) {
+                pandecode_msg("XXX: bad polygon list size (expected %d / 0x%x)\n",
+                                ref_header, ref_size);
+                pandecode_prop("polygon_list_size = 0x%x", t->polygon_list_size);
+                pandecode_msg("body offset %d\n", body_offset);
+        }
+
+        /* The tiler heap has a start and end specified -- it should be
+         * identical to what we have in the BO. The exception is if tiling is
+         * disabled. */
+
+        MEMORY_PROP(t, heap_start);
+        assert(t->heap_end >= t->heap_start);
+
+        struct pandecode_mapped_memory *heap =
+                pandecode_find_mapped_gpu_mem_containing(t->heap_start);
+
+        unsigned heap_size = t->heap_end - t->heap_start;
+
+        /* Tiling is enabled with a special flag */
+        unsigned hierarchy_mask = t->hierarchy_mask & MALI_HIERARCHY_MASK;
+        unsigned tiler_flags = t->hierarchy_mask ^ hierarchy_mask;
+
+        bool tiling_enabled = hierarchy_mask;
+
+        if (tiling_enabled) {
+                /* When tiling is enabled, the heap should be a tight fit */
+                unsigned heap_offset = t->heap_start - heap->gpu_va;
+                if ((heap_offset + heap_size) != heap->length) {
+                        pandecode_msg("XXX: heap size %u (expected %zu)\n",
+                                        heap_size, heap->length - heap_offset);
+                }
+
+                /* We should also have no other flags */
+                if (tiler_flags)
+                        pandecode_msg("XXX: unexpected tiler %X\n", tiler_flags);
+        } else {
+                /* When tiling is disabled, we should have that flag and no others */
+
+                if (tiler_flags != MALI_TILER_DISABLED) {
+                        pandecode_msg("XXX: unexpected tiler flag %X, expected MALI_TILER_DISABLED\n",
+                                        tiler_flags);
+                }
+
+                /* We should also have an empty heap */
+                if (heap_size) {
+                        pandecode_msg("XXX: tiler heap size %d given, expected empty\n",
+                                        heap_size);
+                }
+
+                /* Disabled tiling is used only for clear-only jobs, which are
+                 * purely FRAGMENT, so we should never see this for
+                 * non-FRAGMENT descriptors. */
+
+                if (!is_fragment)
+                        pandecode_msg("XXX: tiler disabled for non-FRAGMENT job\n");
+        }
+
+        /* We've never seen weights used in practice, but we know from the
+         * kernel these fields is there */
+
+        bool nonzero_weights = false;
+
+        for (unsigned w = 0; w < ARRAY_SIZE(t->weights); ++w) {
+                nonzero_weights |= t->weights[w] != 0x0;
+        }
+
+        if (nonzero_weights) {
+                pandecode_log(".weights = { ");
+
+                for (unsigned w = 0; w < ARRAY_SIZE(t->weights); ++w) {
+                        pandecode_log_cont("%d, ", t->weights[w]);
+                }
+
+                pandecode_log("},");
+        }
+
+        pandecode_indent--;
+        pandecode_log("}\n");
+}
+
+/* TODO: The Bifrost tiler is not understood at all yet */
+
+static void
+pandecode_bifrost_tiler_descriptor(const struct mali_framebuffer *fb)
+{
+        pandecode_log(".tiler = {\n");
+        pandecode_indent++;
+
+        MEMORY_PROP(fb, tiler_meta);
+
+        for (int i = 0; i < 16; i++) {
+                if (fb->zeros[i] != 0) {
+                        pandecode_msg("XXX: tiler descriptor zero %d tripped, value %x\n",
+                                      i, fb->zeros[i]);
+                }
+        }
+
+        pandecode_log("},\n");
+
+        pandecode_indent--;
+        pandecode_log("}\n");
+
+}
+
+/* Information about the framebuffer passed back for
+ * additional analysis */
+
+struct pandecode_fbd {
+        unsigned width;
+        unsigned height;
+        unsigned rt_count;
+        bool has_extra;
+};
+
+static void
+pandecode_sfbd_format(struct mali_sfbd_format format)
+{
+        pandecode_log(".format = {\n");
+        pandecode_indent++;
+
+        pandecode_log(".unk1 = ");
+        pandecode_log_decoded_flags(sfbd_unk1_info, format.unk1);
+        pandecode_log_cont(",\n");
+
+        /* TODO: Map formats so we can check swizzles and print nicely */
+        pandecode_log("swizzle");
+        pandecode_swizzle(format.swizzle, MALI_RGBA8_UNORM);
+        pandecode_log_cont(",\n");
+
+        pandecode_prop("nr_channels = MALI_POSITIVE(%d)",
+                       (format.nr_channels + 1));
+
+        pandecode_log(".unk2 = ");
+        pandecode_log_decoded_flags(sfbd_unk2_info, format.unk2);
+        pandecode_log_cont(",\n");
+
+        pandecode_prop("block = %s", pandecode_block_format(format.block));
+
+        pandecode_prop("unk3 = 0x%" PRIx32, format.unk3);
+
+        pandecode_indent--;
+        pandecode_log("},\n");
+}
+
+static void
+pandecode_shared_memory(const struct mali_shared_memory *desc, bool is_compute)
+{
+        pandecode_prop("stack_shift = 0x%x", desc->stack_shift);
+
+        if (desc->unk0)
+                pandecode_prop("unk0 = 0x%x", desc->unk0);
+
+        if (desc->shared_workgroup_count != 0x1F) {
+                pandecode_prop("shared_workgroup_count = %d", desc->shared_workgroup_count);
+                if (!is_compute)
+                        pandecode_msg("XXX: wrong workgroup count for noncompute\n");
+        }
+
+        if (desc->shared_unk1 || desc->shared_shift) {
+                pandecode_prop("shared_unk1 = %X", desc->shared_unk1);
+                pandecode_prop("shared_shift = %X", desc->shared_shift);
+
+                if (!is_compute)
+                        pandecode_msg("XXX: shared memory configured in noncompute shader");
+        }
+
+        if (desc->shared_zero) {
+                pandecode_msg("XXX: shared memory zero tripped\n");
+                pandecode_prop("shared_zero = 0x%" PRIx32, desc->shared_zero);
+        }
+
+        if (desc->shared_memory && !is_compute)
+                pandecode_msg("XXX: shared memory used in noncompute shader\n");
+
+        MEMORY_PROP(desc, scratchpad);
+        MEMORY_PROP(desc, shared_memory);
+        MEMORY_PROP(desc, unknown1);
+
+        if (desc->scratchpad) {
+                struct pandecode_mapped_memory *smem =
+                        pandecode_find_mapped_gpu_mem_containing(desc->scratchpad);
+
+                pandecode_msg("scratchpad size %u\n", smem->length);
+        }
+
+}
+
+static struct pandecode_fbd
+pandecode_sfbd(uint64_t gpu_va, int job_no, bool is_fragment, unsigned gpu_id)
+{
+        struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
+        const struct mali_single_framebuffer *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va);
+
+        struct pandecode_fbd info = {
+                .has_extra = false,
+                .rt_count = 1
+        };
+
+        pandecode_log("struct mali_single_framebuffer framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
+        pandecode_indent++;
+
+        pandecode_log(".shared_memory = {\n");
+        pandecode_indent++;
+        pandecode_shared_memory(&s->shared_memory, false);
+        pandecode_indent--;
+        pandecode_log("},\n");
+
+        pandecode_sfbd_format(s->format);
+
+        info.width = s->width + 1;
+        info.height = s->height + 1;
+
+        pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", info.width);
+        pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", info.height);
+
+        MEMORY_PROP(s, checksum);
+
+        if (s->checksum_stride)
+                pandecode_prop("checksum_stride = %d", s->checksum_stride);
+
+        MEMORY_PROP(s, framebuffer);
+        pandecode_prop("stride = %d", s->stride);
+
+        /* Earlier in the actual commandstream -- right before width -- but we
+         * delay to flow nicer */
+
+        pandecode_log(".clear_flags = ");
+        pandecode_log_decoded_flags(clear_flag_info, s->clear_flags);
+        pandecode_log_cont(",\n");
+
+        if (s->depth_buffer) {
+                MEMORY_PROP(s, depth_buffer);
+                pandecode_prop("depth_stride = %d", s->depth_stride);
+        }
+
+        if (s->stencil_buffer) {
+                MEMORY_PROP(s, stencil_buffer);
+                pandecode_prop("stencil_stride = %d", s->stencil_stride);
+        }
+
+        if (s->depth_stride_zero ||
+            s->stencil_stride_zero ||
+            s->zero7 || s->zero8) {
+                pandecode_msg("XXX: Depth/stencil zeros tripped\n");
+                pandecode_prop("depth_stride_zero = 0x%x",
+                               s->depth_stride_zero);
+                pandecode_prop("stencil_stride_zero = 0x%x",
+                               s->stencil_stride_zero);
+                pandecode_prop("zero7 = 0x%" PRIx32,
+                               s->zero7);
+                pandecode_prop("zero8 = 0x%" PRIx32,
+                               s->zero8);
+        }
+
+        if (s->clear_color_1 | s->clear_color_2 | s->clear_color_3 | s->clear_color_4) {
+                pandecode_prop("clear_color_1 = 0x%" PRIx32, s->clear_color_1);
+                pandecode_prop("clear_color_2 = 0x%" PRIx32, s->clear_color_2);
+                pandecode_prop("clear_color_3 = 0x%" PRIx32, s->clear_color_3);
+                pandecode_prop("clear_color_4 = 0x%" PRIx32, s->clear_color_4);
+        }
+
+        if (s->clear_depth_1 != 0 || s->clear_depth_2 != 0 || s->clear_depth_3 != 0 || s->clear_depth_4 != 0) {
+                pandecode_prop("clear_depth_1 = %f", s->clear_depth_1);
+                pandecode_prop("clear_depth_2 = %f", s->clear_depth_2);
+                pandecode_prop("clear_depth_3 = %f", s->clear_depth_3);
+                pandecode_prop("clear_depth_4 = %f", s->clear_depth_4);
+        }
+
+        if (s->clear_stencil) {
+                pandecode_prop("clear_stencil = 0x%x", s->clear_stencil);
+        }
+
+        const struct midgard_tiler_descriptor t = s->tiler;
+
+        bool has_hierarchy = !(gpu_id == 0x0720 || gpu_id == 0x0820 || gpu_id == 0x0830);
+        pandecode_midgard_tiler_descriptor(&t, s->width + 1, s->height + 1, is_fragment, has_hierarchy);
+
+        pandecode_indent--;
+        pandecode_log("};\n");
+
+        pandecode_prop("zero2 = 0x%" PRIx32, s->zero2);
+        pandecode_prop("zero4 = 0x%" PRIx32, s->zero4);
+        pandecode_prop("zero5 = 0x%" PRIx32, s->zero5);
+
+        pandecode_log_cont(".zero3 = {");
+
+        for (int i = 0; i < sizeof(s->zero3) / sizeof(s->zero3[0]); ++i)
+                pandecode_log_cont("%X, ", s->zero3[i]);
+
+        pandecode_log_cont("},\n");
+
+        pandecode_log_cont(".zero6 = {");
+
+        for (int i = 0; i < sizeof(s->zero6) / sizeof(s->zero6[0]); ++i)
+                pandecode_log_cont("%X, ", s->zero6[i]);
+
+        pandecode_log_cont("},\n");
+
+        return info;
+}
+
+static void
+pandecode_compute_fbd(uint64_t gpu_va, int job_no)
+{
+        struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
+        const struct mali_shared_memory *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va);
+
+        pandecode_log("struct mali_shared_memory shared_%"PRIx64"_%d = {\n", gpu_va, job_no);
+        pandecode_indent++;
+        pandecode_shared_memory(s, true);
+        pandecode_indent--;
+        pandecode_log("},\n");
+}
+
+/* Extracts the number of components associated with a Mali format */
+
+static unsigned
+pandecode_format_component_count(enum mali_format fmt)
+{
+        /* Mask out the format class */
+        unsigned top = fmt & 0b11100000;
+
+        switch (top) {
+        case MALI_FORMAT_SNORM:
+        case MALI_FORMAT_UINT:
+        case MALI_FORMAT_UNORM:
+        case MALI_FORMAT_SINT:
+                return ((fmt >> 3) & 3) + 1;
+        default:
+                /* TODO: Validate */
+                return 4;
+        }
+}
+
+/* Extracts a mask of accessed components from a 12-bit Mali swizzle */
+
+static unsigned
+pandecode_access_mask_from_channel_swizzle(unsigned swizzle)
+{
+        unsigned mask = 0;
+        assert(MALI_CHANNEL_RED == 0);
+
+        for (unsigned c = 0; c < 4; ++c) {
+                enum mali_channel chan = (swizzle >> (3*c)) & 0x7;
+
+                if (chan <= MALI_CHANNEL_ALPHA)
+                        mask |= (1 << chan);
+        }
+
+        return mask;
+}
+
+/* Validates that a (format, swizzle) pair is valid, in the sense that the
+ * swizzle doesn't access any components that are undefined in the format.
+ * Returns whether the swizzle is trivial (doesn't do any swizzling) and can be
+ * omitted */
+
+static bool
+pandecode_validate_format_swizzle(enum mali_format fmt, unsigned swizzle)
+{
+        unsigned nr_comp = pandecode_format_component_count(fmt);
+        unsigned access_mask = pandecode_access_mask_from_channel_swizzle(swizzle);
+        unsigned valid_mask = (1 << nr_comp) - 1;
+        unsigned invalid_mask = ~valid_mask;
+
+        if (access_mask & invalid_mask) {
+                pandecode_msg("XXX: invalid components accessed\n");
+                return false;
+        }
+
+        /* Check for the default non-swizzling swizzle so we can suppress
+         * useless printing for the defaults */
+
+        unsigned default_swizzles[4] = {
+                MALI_CHANNEL_RED | (MALI_CHANNEL_ZERO  << 3) | (MALI_CHANNEL_ZERO << 6) | (MALI_CHANNEL_ONE   << 9),
+                MALI_CHANNEL_RED | (MALI_CHANNEL_GREEN << 3) | (MALI_CHANNEL_ZERO << 6) | (MALI_CHANNEL_ONE   << 9),
+                MALI_CHANNEL_RED | (MALI_CHANNEL_GREEN << 3) | (MALI_CHANNEL_BLUE << 6) | (MALI_CHANNEL_ONE   << 9),
+                MALI_CHANNEL_RED | (MALI_CHANNEL_GREEN << 3) | (MALI_CHANNEL_BLUE << 6) | (MALI_CHANNEL_ALPHA << 9)
+        };
+
+        return (swizzle == default_swizzles[nr_comp - 1]);
+}
+
+/* Maps MALI_RGBA32F to rgba32f, etc */
+
+static void
+pandecode_format_short(enum mali_format fmt, bool srgb)
+{
+        /* We want a type-like format, so cut off the initial MALI_ */
+        char *format = pandecode_format(fmt);
+        format += strlen("MALI_");
+
+        unsigned len = strlen(format);
+        char *lower_format = calloc(1, len + 1);
+
+        for (unsigned i = 0; i < len; ++i)
+                lower_format[i] = tolower(format[i]);
+
+        /* Sanity check sRGB flag is applied to RGB, per the name */
+        if (srgb && lower_format[0] != 'r')
+                pandecode_msg("XXX: sRGB applied to non-colour format\n");
+
+        /* Just prefix with an s, so you get formats like srgba8_unorm */
+        if (srgb)
+                pandecode_log_cont("s");
+
+        pandecode_log_cont("%s", lower_format);
+        free(lower_format);
+}
+
+static void
+pandecode_swizzle(unsigned swizzle, enum mali_format format)
+{
+        /* First, do some validation */
+        bool trivial_swizzle = pandecode_validate_format_swizzle(
+                        format, swizzle);
+
+        if (trivial_swizzle)
+                return;
+
+        /* Next, print the swizzle */
+        pandecode_log_cont(".");
+
+        static const char components[] = "rgba01";
+
+        for (unsigned c = 0; c < 4; ++c) {
+                enum mali_channel chan = (swizzle >> (3 * c)) & 0x7;
+
+                if (chan >= MALI_CHANNEL_RESERVED_0) {
+                        pandecode_log("XXX: invalid swizzle channel %d\n", chan);
+                        continue;
+                }
+                pandecode_log_cont("%c", components[chan]);
+        }
+}
+
+static void
+pandecode_rt_format(struct mali_rt_format format)
+{
+        pandecode_log(".format = {\n");
+        pandecode_indent++;
+
+        pandecode_prop("unk1 = 0x%" PRIx32, format.unk1);
+        pandecode_prop("unk2 = 0x%" PRIx32, format.unk2);
+        pandecode_prop("unk3 = 0x%" PRIx32, format.unk3);
+        pandecode_prop("unk4 = 0x%" PRIx32, format.unk4);
+
+        pandecode_prop("block = %s", pandecode_block_format(format.block));
+
+        /* TODO: Map formats so we can check swizzles and print nicely */
+        pandecode_log("swizzle");
+        pandecode_swizzle(format.swizzle, MALI_RGBA8_UNORM);
+        pandecode_log_cont(",\n");
+
+        pandecode_prop("nr_channels = MALI_POSITIVE(%d)",
+                       (format.nr_channels + 1));
+
+        pandecode_log(".flags = ");
+        pandecode_log_decoded_flags(mfbd_fmt_flag_info, format.flags);
+        pandecode_log_cont(",\n");
+
+        pandecode_prop("msaa = %s", pandecode_msaa_mode(format.msaa));
+
+        /* In theory, the no_preload bit can be cleared to enable MFBD preload,
+         * which is a faster hardware-based alternative to the wallpaper method
+         * to preserve framebuffer contents across frames. In practice, MFBD
+         * preload is buggy on Midgard, and so this is a chicken bit. If this
+         * bit isn't set, most likely something broke unrelated to preload */
+
+        if (!format.no_preload) {
+                pandecode_msg("XXX: buggy MFBD preload enabled - chicken bit should be clear\n");
+                pandecode_prop("no_preload = 0x%" PRIx32, format.no_preload);
+        }
+
+        if (format.zero)
+                pandecode_prop("zero = 0x%" PRIx32, format.zero);
+
+        pandecode_indent--;
+        pandecode_log("},\n");
+}
+
+static void
+pandecode_render_target(uint64_t gpu_va, unsigned job_no, const struct mali_framebuffer *fb)
+{
+        pandecode_log("struct mali_render_target rts_list_%"PRIx64"_%d[] = {\n", gpu_va, job_no);
+        pandecode_indent++;
+
+        for (int i = 0; i < (fb->rt_count_1 + 1); i++) {
+                mali_ptr rt_va = gpu_va + i * sizeof(struct mali_render_target);
+                struct pandecode_mapped_memory *mem =
+                        pandecode_find_mapped_gpu_mem_containing(rt_va);
+                const struct mali_render_target *PANDECODE_PTR_VAR(rt, mem, (mali_ptr) rt_va);
+
+                pandecode_log("{\n");
+                pandecode_indent++;
+
+                pandecode_rt_format(rt->format);
+
+                if (rt->format.block == MALI_BLOCK_AFBC) {
+                        pandecode_log(".afbc = {\n");
+                        pandecode_indent++;
+
+                        char *a = pointer_as_memory_reference(rt->afbc.metadata);
+                        pandecode_prop("metadata = %s", a);
+                        free(a);
+
+                        pandecode_prop("stride = %d", rt->afbc.stride);
+
+                        pandecode_log(".flags = ");
+                        pandecode_log_decoded_flags(afbc_fmt_flag_info, rt->afbc.flags);
+                        pandecode_log_cont(",\n");
+
+                        pandecode_indent--;
+                        pandecode_log("},\n");
+                } else if (rt->afbc.metadata || rt->afbc.stride || rt->afbc.flags) {
+                        pandecode_msg("XXX: AFBC disabled but AFBC field set (0x%lX, 0x%x, 0x%x)\n",
+                                        rt->afbc.metadata,
+                                        rt->afbc.stride,
+                                        rt->afbc.flags);
+                }
+
+                MEMORY_PROP(rt, framebuffer);
+                pandecode_prop("framebuffer_stride = %d", rt->framebuffer_stride);
+
+                if (rt->layer_stride)
+                        pandecode_prop("layer_stride = %d", rt->layer_stride);
+
+                if (rt->clear_color_1 | rt->clear_color_2 | rt->clear_color_3 | rt->clear_color_4) {
+                        pandecode_prop("clear_color_1 = 0x%" PRIx32, rt->clear_color_1);
+                        pandecode_prop("clear_color_2 = 0x%" PRIx32, rt->clear_color_2);
+                        pandecode_prop("clear_color_3 = 0x%" PRIx32, rt->clear_color_3);
+                        pandecode_prop("clear_color_4 = 0x%" PRIx32, rt->clear_color_4);
+                }
+
+                if (rt->zero1 || rt->zero2) {
+                        pandecode_msg("XXX: render target zeros tripped\n");
+                        pandecode_prop("zero1 = 0x%" PRIx64, rt->zero1);
+                        pandecode_prop("zero2 = 0x%" PRIx32, rt->zero2);
+                }
+
+                pandecode_indent--;
+                pandecode_log("},\n");
+        }
+
+        pandecode_indent--;
+        pandecode_log("};\n");
+}
+
+static struct pandecode_fbd
+pandecode_mfbd_bfr(uint64_t gpu_va, int job_no, bool is_fragment, bool is_compute, bool is_bifrost)
+{
+        struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
+        const struct mali_framebuffer *PANDECODE_PTR_VAR(fb, mem, (mali_ptr) gpu_va);
+
+        struct pandecode_fbd info;
+
+        if (is_bifrost && fb->msaa.sample_locations) {
+                /* The blob stores all possible sample locations in a single buffer
+                 * allocated on startup, and just switches the pointer when switching
+                 * MSAA state. For now, we just put the data into the cmdstream, but we
+                 * should do something like what the blob does with a real driver.
+                 *
+                 * There seem to be 32 slots for sample locations, followed by another
+                 * 16. The second 16 is just the center location followed by 15 zeros
+                 * in all the cases I've identified (maybe shader vs. depth/color
+                 * samples?).
+                 */
+
+                struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(fb->msaa.sample_locations);
+
+                const u16 *PANDECODE_PTR_VAR(samples, smem, fb->msaa.sample_locations);
+
+                pandecode_log("uint16_t sample_locations_%d[] = {\n", job_no);
+                pandecode_indent++;
+
+                for (int i = 0; i < 32 + 16; i++) {
+                        pandecode_log("%d, %d,\n", samples[2 * i], samples[2 * i + 1]);
+                }
+
+                pandecode_indent--;
+                pandecode_log("};\n");
+        }
+        pandecode_log("struct mali_framebuffer framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
+        pandecode_indent++;
+
+        if (is_bifrost) {
+                pandecode_log(".msaa = {\n");
+                pandecode_indent++;
+
+                if (fb->msaa.sample_locations)
+                        pandecode_prop("sample_locations = sample_locations_%d", job_no);
+                else
+                        pandecode_msg("XXX: sample_locations missing\n");
+
+                if (fb->msaa.zero1 || fb->msaa.zero2 || fb->msaa.zero4) {
+                        pandecode_msg("XXX: multisampling zero tripped\n");
+                        pandecode_prop("zero1 = %" PRIx64, fb->msaa.zero1);
+                        pandecode_prop("zero2 = %" PRIx64, fb->msaa.zero2);
+                        pandecode_prop("zero4 = %" PRIx64, fb->msaa.zero4);
+                }
+
+                pandecode_indent--;
+                pandecode_log("},\n");
+        } else {
+                pandecode_log(".shared_memory = {\n");
+                pandecode_indent++;
+                pandecode_shared_memory(&fb->shared_memory, is_compute);
+                pandecode_indent--;
+                pandecode_log("},\n");
+        }
+
+        info.width = fb->width1 + 1;
+        info.height = fb->height1 + 1;
+        info.rt_count = fb->rt_count_1 + 1;
+
+        pandecode_prop("width1 = MALI_POSITIVE(%d)", fb->width1 + 1);
+        pandecode_prop("height1 = MALI_POSITIVE(%d)", fb->height1 + 1);
+        pandecode_prop("width2 = MALI_POSITIVE(%d)", fb->width2 + 1);
+        pandecode_prop("height2 = MALI_POSITIVE(%d)", fb->height2 + 1);
+
+        pandecode_prop("unk1 = 0x%x", fb->unk1);
+        pandecode_prop("unk2 = 0x%x", fb->unk2);
+        pandecode_prop("rt_count_1 = MALI_POSITIVE(%d)", fb->rt_count_1 + 1);
+        pandecode_prop("rt_count_2 = %d", fb->rt_count_2);
+
+        pandecode_log(".mfbd_flags = ");
+        pandecode_log_decoded_flags(mfbd_flag_info, fb->mfbd_flags);
+        pandecode_log_cont(",\n");
+
+        if (fb->clear_stencil)
+                pandecode_prop("clear_stencil = 0x%x", fb->clear_stencil);
+
+        if (fb->clear_depth)
+                pandecode_prop("clear_depth = %f", fb->clear_depth);
+
+        if (!is_compute)
+                if (is_bifrost)
+                        pandecode_bifrost_tiler_descriptor(fb);
+                else {
+                        const struct midgard_tiler_descriptor t = fb->tiler;
+                        pandecode_midgard_tiler_descriptor(&t, fb->width1 + 1, fb->height1 + 1, is_fragment, true);
+                }
+        else
+                pandecode_msg("XXX: skipping compute MFBD, fixme\n");
+
+        if (fb->zero3 || fb->zero4) {
+                pandecode_msg("XXX: framebuffer zeros tripped\n");
+                pandecode_prop("zero3 = 0x%" PRIx32, fb->zero3);
+                pandecode_prop("zero4 = 0x%" PRIx32, fb->zero4);
+        }
+
+        pandecode_indent--;
+        pandecode_log("};\n");
+
+        gpu_va += sizeof(struct mali_framebuffer);
+
+        info.has_extra = (fb->mfbd_flags & MALI_MFBD_EXTRA) && is_fragment;
+
+        if (info.has_extra) {
+                mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
+                const struct mali_framebuffer_extra *PANDECODE_PTR_VAR(fbx, mem, (mali_ptr) gpu_va);
+
+                pandecode_log("struct mali_framebuffer_extra fb_extra_%"PRIx64"_%d = {\n", gpu_va, job_no);
+                pandecode_indent++;
+
+                MEMORY_PROP(fbx, checksum);
+
+                if (fbx->checksum_stride)
+                        pandecode_prop("checksum_stride = %d", fbx->checksum_stride);
+
+                pandecode_log(".flags_hi = ");
+                pandecode_log_decoded_flags(mfbd_extra_flag_hi_info, fbx->flags_hi);
+                pandecode_log_cont(",\n");
+
+                pandecode_log(".flags_lo = ");
+                pandecode_log_decoded_flags(mfbd_extra_flag_lo_info, fbx->flags_lo);
+                pandecode_log_cont(",\n");
+
+                pandecode_prop("zs_block = %s", pandecode_block_format(fbx->zs_block));
+                pandecode_prop("zs_samples = MALI_POSITIVE(%u)", fbx->zs_samples + 1);
+
+                if (fbx->zs_block == MALI_BLOCK_AFBC) {
+                        pandecode_log(".ds_afbc = {\n");
+                        pandecode_indent++;
+
+                        MEMORY_PROP_DIR(fbx->ds_afbc, depth_stencil_afbc_metadata);
+                        pandecode_prop("depth_stencil_afbc_stride = %d",
+                                       fbx->ds_afbc.depth_stencil_afbc_stride);
+                        MEMORY_PROP_DIR(fbx->ds_afbc, depth_stencil);
+
+                        pandecode_log(".flags = ");
+                        pandecode_log_decoded_flags(afbc_fmt_flag_info, fbx->ds_afbc.flags);
+                        pandecode_log_cont(",\n");
+
+                        if (fbx->ds_afbc.padding) {
+                                pandecode_msg("XXX: Depth/stencil AFBC zeros tripped\n");
+                                pandecode_prop("padding = 0x%" PRIx64, fbx->ds_afbc.padding);
+                        }
+
+                        pandecode_indent--;
+                        pandecode_log("},\n");
+                } else {
+                        pandecode_log(".ds_linear = {\n");
+                        pandecode_indent++;
+
+                        if (fbx->ds_linear.depth) {
+                                MEMORY_PROP_DIR(fbx->ds_linear, depth);
+                                pandecode_prop("depth_stride = %d",
+                                               fbx->ds_linear.depth_stride);
+                                pandecode_prop("depth_layer_stride = %d",
+                                               fbx->ds_linear.depth_layer_stride);
+                        } else if (fbx->ds_linear.depth_stride || fbx->ds_linear.depth_layer_stride) {
+                                pandecode_msg("XXX: depth stride zero tripped %d %d\n", fbx->ds_linear.depth_stride, fbx->ds_linear.depth_layer_stride);
+                        }
+
+                        if (fbx->ds_linear.stencil) {
+                                MEMORY_PROP_DIR(fbx->ds_linear, stencil);
+                                pandecode_prop("stencil_stride = %d",
+                                               fbx->ds_linear.stencil_stride);
+                                pandecode_prop("stencil_layer_stride = %d",
+                                               fbx->ds_linear.stencil_layer_stride);
+                        } else if (fbx->ds_linear.stencil_stride || fbx->ds_linear.stencil_layer_stride) {
+                                pandecode_msg("XXX: stencil stride zero tripped %d %d\n", fbx->ds_linear.stencil_stride, fbx->ds_linear.stencil_layer_stride);
+                        }
+
+                        if (fbx->ds_linear.depth_stride_zero ||
+                            fbx->ds_linear.stencil_stride_zero) {
+                                pandecode_msg("XXX: Depth/stencil zeros tripped\n");
+                                pandecode_prop("depth_stride_zero = 0x%x",
+                                               fbx->ds_linear.depth_stride_zero);
+                                pandecode_prop("stencil_stride_zero = 0x%x",
+                                               fbx->ds_linear.stencil_stride_zero);
+                        }
+
+                        pandecode_indent--;
+                        pandecode_log("},\n");
+                }
+
+                if (fbx->clear_color_1 | fbx->clear_color_2) {
+                        pandecode_prop("clear_color_1 = 0x%" PRIx32, fbx->clear_color_1);
+                        pandecode_prop("clear_color_2 = 0x%" PRIx32, fbx->clear_color_2);
+                }
+
+                if (fbx->zero3) {
+                        pandecode_msg("XXX: fb_extra zeros tripped\n");
+                        pandecode_prop("zero3 = 0x%" PRIx64, fbx->zero3);
+                }
+
+                pandecode_indent--;
+                pandecode_log("};\n");
+
+                gpu_va += sizeof(struct mali_framebuffer_extra);
+        }
+
+        if (is_fragment)
+                pandecode_render_target(gpu_va, job_no, fb);
+
+        return info;
+}
+
+/* Just add a comment decoding the shift/odd fields forming the padded vertices
+ * count */
+
+static void
+pandecode_padded_vertices(unsigned shift, unsigned k)
+{
+        unsigned odd = 2*k + 1;
+        unsigned pot = 1 << shift;
+        pandecode_msg("padded_num_vertices = %d\n", odd * pot);
+}
+
+/* Given a magic divisor, recover what we were trying to divide by.
+ *
+ * Let m represent the magic divisor. By definition, m is an element on Z, whre
+ * 0 <= m < 2^N, for N bits in m.
+ *
+ * Let q represent the number we would like to divide by.
+ *
+ * By definition of a magic divisor for N-bit unsigned integers (a number you
+ * multiply by to magically get division), m is a number such that:
+ *
+ *      (m * x) & (2^N - 1) = floor(x/q).
+ *      for all x on Z where 0 <= x < 2^N
+ *
+ * Ignore the case where any of the above values equals zero; it is irrelevant
+ * for our purposes (instanced arrays).
+ *
+ * Choose x = q. Then:
+ *
+ *      (m * x) & (2^N - 1) = floor(x/q).
+ *      (m * q) & (2^N - 1) = floor(q/q).
+ *
+ *      floor(q/q) = floor(1) = 1, therefore:
+ *
+ *      (m * q) & (2^N - 1) = 1
+ *
+ * Recall the identity that the bitwise AND of one less than a power-of-two
+ * equals the modulo with that power of two, i.e. for all x:
+ *
+ *      x & (2^N - 1) = x % N
+ *
+ * Therefore:
+ *
+ *      mq % (2^N) = 1
+ *
+ * By definition, a modular multiplicative inverse of a number m is the number
+ * q such that with respect to a modulos M:
+ *
+ *      mq % M = 1
+ *
+ * Therefore, q is the modular multiplicative inverse of m with modulus 2^N.
+ *
+ */
+
+static void
+pandecode_magic_divisor(uint32_t magic, unsigned shift, unsigned orig_divisor, unsigned extra)
+{
+#if 0
+        /* Compute the modular inverse of `magic` with respect to 2^(32 -
+         * shift) the most lame way possible... just repeatedly add.
+         * Asymptoptically slow but nobody cares in practice, unless you have
+         * massive numbers of vertices or high divisors. */
+
+        unsigned inverse = 0;
+
+        /* Magic implicitly has the highest bit set */
+        magic |= (1 << 31);
+
+        /* Depending on rounding direction */
+        if (extra)
+                magic++;
+
+        for (;;) {
+                uint32_t product = magic * inverse;
+
+                if (shift) {
+                        product >>= shift;
+                }
+
+                if (product == 1)
+                        break;
+
+                ++inverse;
+        }
+
+        pandecode_msg("dividing by %d (maybe off by two)\n", inverse);
+
+        /* Recall we're supposed to divide by (gl_level_divisor *
+         * padded_num_vertices) */
+
+        unsigned padded_num_vertices = inverse / orig_divisor;
+
+        pandecode_msg("padded_num_vertices = %d\n", padded_num_vertices);
+#endif
+}
+
+static void
+pandecode_attributes(const struct pandecode_mapped_memory *mem,
+                            mali_ptr addr, int job_no, char *suffix,
+                            int count, bool varying, enum mali_job_type job_type)
+{
+        char *prefix = varying ? "varying" : "attribute";
+        assert(addr);
+
+        if (!count) {
+                pandecode_msg("warn: No %s records\n", prefix);
+                return;
+        }
+
+        union mali_attr *attr = pandecode_fetch_gpu_mem(mem, addr, sizeof(union mali_attr) * count);
+
+        for (int i = 0; i < count; ++i) {
+                /* First, check for special records */
+                if (attr[i].elements < MALI_RECORD_SPECIAL) {
+                        if (attr[i].size)
+                                pandecode_msg("XXX: tripped size=%d\n", attr[i].size);
+
+                        if (attr[i].stride) {
+                                /* gl_InstanceID passes a magic divisor in the
+                                 * stride field to divide by the padded vertex
+                                 * count. No other records should do so, so
+                                 * stride should otherwise be zero. Note that
+                                 * stride in the usual attribute sense doesn't
+                                 * apply to special records. */
+
+                                bool has_divisor = attr[i].elements == MALI_ATTR_INSTANCEID;
+
+                                pandecode_log_cont("/* %smagic divisor = %X */ ",
+                                                has_divisor ? "" : "XXX: ", attr[i].stride);
+                        }
+
+                        if (attr[i].shift || attr[i].extra_flags) {
+                                /* Attributes use these fields for
+                                 * instancing/padding/etc type issues, but
+                                 * varyings don't */
+
+                                pandecode_log_cont("/* %sshift=%d, extra=%d */ ",
+                                                varying ? "XXX: " : "",
+                                                attr[i].shift, attr[i].extra_flags);
+                        }
+
+                        /* Print the special record name */
+                        bool attribute = false;
+                        pandecode_log("%s_%d = %s;\n", prefix, i, pandecode_special_record(attr[i].elements, &attribute));
+
+                        /* Sanity check */
+                        if (attribute == varying)
+                                pandecode_msg("XXX: mismatched special record\n");
+
+                        continue;
+        }
+
+                enum mali_attr_mode mode = attr[i].elements & 7;
+
+                if (mode == MALI_ATTR_UNUSED)
+                        pandecode_msg("XXX: unused attribute record\n");
+
+                /* For non-linear records, we need to print the type of record */
+                if (mode != MALI_ATTR_LINEAR)
+                        pandecode_log_cont("%s ", pandecode_attr_mode_short(mode));
+
+                /* Print the name to link with attr_meta */
+                pandecode_log_cont("%s_%d", prefix, i);
+
+                /* Print the stride and size */
+                pandecode_log_cont("<%u>[%u]", attr[i].stride, attr[i].size);
+
+                /* TODO: Sanity check the quotient itself. It must be equal to
+                 * (or be greater than, if the driver added padding) the padded
+                 * vertex count. */
+
+                /* Finally, print the pointer */
+                mali_ptr raw_elements = attr[i].elements & ~7;
+                char *a = pointer_as_memory_reference(raw_elements);
+                pandecode_log_cont(" = (%s);\n", a);
+                free(a);
+
+                /* Check the pointer */
+                pandecode_validate_buffer(raw_elements, attr[i].size);
+
+                /* shift/extra_flags exist only for instanced */
+                if (attr[i].shift | attr[i].extra_flags) {
+                        /* These are set to random values by the blob for
+                         * varyings, most likely a symptom of uninitialized
+                         * memory where the hardware masked the bug. As such we
+                         * put this at a warning, not an error. */
+
+                        if (mode == MALI_ATTR_LINEAR)
+                                pandecode_msg("warn: instancing fields set for linear\n");
+
+                        pandecode_prop("shift = %d", attr[i].shift);
+                        pandecode_prop("extra_flags = %d", attr[i].extra_flags);
+                }
+
+                /* Decode further where possible */
+
+                if (mode == MALI_ATTR_MODULO) {
+                        pandecode_padded_vertices(
+                                attr[i].shift,
+                                attr[i].extra_flags);
+                }
+
+                if (mode == MALI_ATTR_NPOT_DIVIDE) {
+                        i++;
+                        pandecode_log("{\n");
+                        pandecode_indent++;
+                        pandecode_prop("unk = 0x%x", attr[i].unk);
+                        pandecode_prop("magic_divisor = 0x%08x", attr[i].magic_divisor);
+                        if (attr[i].zero != 0)
+                                pandecode_prop("XXX: zero tripped (0x%x)\n", attr[i].zero);
+                        pandecode_prop("divisor = %d", attr[i].divisor);
+                        pandecode_magic_divisor(attr[i].magic_divisor, attr[i - 1].shift, attr[i].divisor, attr[i - 1].extra_flags);
+                        pandecode_indent--;
+                        pandecode_log("}, \n");
+                }
+
+        }
+
+        pandecode_log("\n");
+}
+
+static mali_ptr
+pandecode_shader_address(const char *name, mali_ptr ptr)
+{
+        /* TODO: Decode flags */
+        mali_ptr shader_ptr = ptr & ~15;
+
+        char *a = pointer_as_memory_reference(shader_ptr);
+        pandecode_prop("%s = (%s) | %d", name, a, (int) (ptr & 15));
+        free(a);
+
+        return shader_ptr;
+}
+
+static void
+pandecode_stencil(const char *name, const struct mali_stencil_test *stencil)
+{
+        unsigned any_nonzero =
+                stencil->ref | stencil->mask | stencil->func |
+                stencil->sfail | stencil->dpfail | stencil->dppass;
+
+        if (any_nonzero == 0)
+                return;
+
+        const char *func = pandecode_func(stencil->func);
+        const char *sfail = pandecode_stencil_op(stencil->sfail);
+        const char *dpfail = pandecode_stencil_op(stencil->dpfail);
+        const char *dppass = pandecode_stencil_op(stencil->dppass);
+
+        if (stencil->zero)
+                pandecode_msg("XXX: stencil zero tripped: %X\n", stencil->zero);
+
+        pandecode_log(".stencil_%s = {\n", name);
+        pandecode_indent++;
+        pandecode_prop("ref = %d", stencil->ref);
+        pandecode_prop("mask = 0x%02X", stencil->mask);
+        pandecode_prop("func = %s", func);
+        pandecode_prop("sfail = %s", sfail);
+        pandecode_prop("dpfail = %s", dpfail);
+        pandecode_prop("dppass = %s", dppass);
+        pandecode_indent--;
+        pandecode_log("},\n");
+}
+
+static void
+pandecode_blend_equation(const struct mali_blend_equation *blend)
+{
+        if (blend->zero1)
+                pandecode_msg("XXX: blend zero tripped: %X\n", blend->zero1);
+
+        pandecode_log(".equation = {\n");
+        pandecode_indent++;
+
+        pandecode_prop("rgb_mode = 0x%X", blend->rgb_mode);
+        pandecode_prop("alpha_mode = 0x%X", blend->alpha_mode);
+
+        pandecode_log(".color_mask = ");
+        pandecode_log_decoded_flags(mask_flag_info, blend->color_mask);
+        pandecode_log_cont(",\n");
+
+        pandecode_indent--;
+        pandecode_log("},\n");
+}
+
+/* Decodes a Bifrost blend constant. See the notes in bifrost_blend_rt */
+
+static unsigned
+decode_bifrost_constant(u16 constant)
+{
+        float lo = (float) (constant & 0xFF);
+        float hi = (float) (constant >> 8);
+
+        return (hi / 255.0) + (lo / 65535.0);
+}
+
+static mali_ptr
+pandecode_bifrost_blend(void *descs, int job_no, int rt_no)
+{
+        struct bifrost_blend_rt *b =
+                ((struct bifrost_blend_rt *) descs) + rt_no;
+
+        pandecode_log("struct bifrost_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
+        pandecode_indent++;
+
+        pandecode_prop("flags = 0x%" PRIx16, b->flags);
+        pandecode_prop("constant = 0x%" PRIx8 " /* %f */",
+                       b->constant, decode_bifrost_constant(b->constant));
+
+        /* TODO figure out blend shader enable bit */
+        pandecode_blend_equation(&b->equation);
+
+        pandecode_prop("unk2 = 0x%" PRIx16, b->unk2);
+        pandecode_prop("index = 0x%" PRIx16, b->index);
+
+        pandecode_log(".format = ");
+        pandecode_format_short(b->format, false);
+        pandecode_swizzle(b->swizzle, b->format);
+        pandecode_log_cont(",\n");
+
+        pandecode_prop("swizzle = 0x%" PRIx32, b->swizzle);
+        pandecode_prop("format = 0x%" PRIx32, b->format);
+
+        if (b->zero1) {
+                pandecode_msg("XXX: pandecode_bifrost_blend zero1 tripped\n");
+                pandecode_prop("zero1 = 0x%" PRIx32, b->zero1);
+        }
+
+        pandecode_log(".shader_type = ");
+        switch(b->shader_type) {
+        case BIFROST_BLEND_F16:
+                pandecode_log_cont("BIFROST_BLEND_F16");
+                break;
+        case BIFROST_BLEND_F32:
+                pandecode_log_cont("BIFROST_BLEND_F32");
+                break;
+        case BIFROST_BLEND_I32:
+                pandecode_log_cont("BIFROST_BLEND_I32");
+                break;
+        case BIFROST_BLEND_U32:
+                pandecode_log_cont("BIFROST_BLEND_U32");
+                break;
+        case BIFROST_BLEND_I16:
+                pandecode_log_cont("BIFROST_BLEND_I16");
+                break;
+        case BIFROST_BLEND_U16:
+                pandecode_log_cont("BIFROST_BLEND_U16");
+                break;
+        }
+        pandecode_log_cont(",\n");
+
+        if (b->zero2) {
+                pandecode_msg("XXX: pandecode_bifrost_blend zero2 tripped\n");
+                pandecode_prop("zero2 = 0x%" PRIx32, b->zero2);
+        }
+
+        pandecode_prop("shader = 0x%" PRIx32, b->shader);
+
+        pandecode_indent--;
+        pandecode_log("},\n");
+
+        return 0;
+}
+
+static mali_ptr
+pandecode_midgard_blend(union midgard_blend *blend, bool is_shader)
+{
+        /* constant/equation is in a union */
+        if (!blend->shader)
+                return 0;
+
+        pandecode_log(".blend = {\n");
+        pandecode_indent++;
+
+        if (is_shader) {
+                pandecode_shader_address("shader", blend->shader);
+        } else {
+                pandecode_blend_equation(&blend->equation);
+                pandecode_prop("constant = %f", blend->constant);
+        }
+
+        pandecode_indent--;
+        pandecode_log("},\n");
+
+        /* Return blend shader to disassemble if present */
+        return is_shader ? (blend->shader & ~0xF) : 0;
+}
+
+static mali_ptr
+pandecode_midgard_blend_mrt(void *descs, int job_no, int rt_no)
+{
+        struct midgard_blend_rt *b =
+                ((struct midgard_blend_rt *) descs) + rt_no;
+
+        /* Flags determine presence of blend shader */
+        bool is_shader = (b->flags & 0xF) >= 0x2;
+
+        pandecode_log("struct midgard_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
+        pandecode_indent++;
+
+        pandecode_prop("flags = 0x%" PRIx64, b->flags);
+
+        union midgard_blend blend = b->blend;
+        mali_ptr shader = pandecode_midgard_blend(&blend, is_shader);
+
+        pandecode_indent--;
+        pandecode_log("};\n");
+
+        return shader;
+}
+
+/* Attributes and varyings have descriptor records, which contain information
+ * about their format and ordering with the attribute/varying buffers. We'll
+ * want to validate that the combinations specified are self-consistent.
+ */
+
+static int
+pandecode_attribute_meta(int job_no, int count, const struct mali_vertex_tiler_postfix *v, bool varying, char *suffix)
+{
+        char base[128];
+        char *prefix = varying ? "varying" : "attribute";
+        unsigned max_index = 0;
+        snprintf(base, sizeof(base), "%s_meta", prefix);
+
+        struct mali_attr_meta *attr_meta;
+        mali_ptr p = varying ? v->varying_meta : v->attribute_meta;
+
+        struct pandecode_mapped_memory *attr_mem = pandecode_find_mapped_gpu_mem_containing(p);
+
+        for (int i = 0; i < count; ++i, p += sizeof(struct mali_attr_meta)) {
+                attr_meta = pandecode_fetch_gpu_mem(attr_mem, p,
+                                                    sizeof(*attr_mem));
+
+                /* If the record is discard, it should be zero for everything else */
+
+                if (attr_meta->format == MALI_VARYING_DISCARD) {
+                        uint64_t zero =
+                                attr_meta->index |
+                                attr_meta->unknown1 |
+                                attr_meta->unknown3 |
+                                attr_meta->src_offset;
+
+                        if (zero)
+                                pandecode_msg("XXX: expected empty record for varying discard\n");
+
+                        /* We want to look for a literal 0000 swizzle -- this
+                         * is not encoded with all zeroes, however */
+
+                        enum mali_channel z = MALI_CHANNEL_ZERO;
+                        unsigned zero_swizzle = z | (z << 3) | (z << 6) | (z << 9);
+                        bool good_swizzle = attr_meta->swizzle == zero_swizzle;
+
+                        if (!good_swizzle)
+                                pandecode_msg("XXX: expected zero swizzle for discard\n");
+
+                        if (!varying)
+                                pandecode_msg("XXX: cannot discard attribute\n");
+
+                        /* If we're all good, omit the record */
+                        if (!zero && varying && good_swizzle) {
+                                pandecode_log("/* discarded varying */\n");
+                                continue;
+                        }
+                }
+
+                if (attr_meta->index > max_index)
+                        max_index = attr_meta->index;
+
+                if (attr_meta->unknown1 != 0x2) {
+                        pandecode_msg("XXX: expected unknown1 = 0x2\n");
+                        pandecode_prop("unknown1 = 0x%" PRIx64, (u64) attr_meta->unknown1);
+                }
+
+                if (attr_meta->unknown3) {
+                        pandecode_msg("XXX: unexpected unknown3 set\n");
+                        pandecode_prop("unknown3 = 0x%" PRIx64, (u64) attr_meta->unknown3);
+                }
+
+                pandecode_format_short(attr_meta->format, false);
+                pandecode_log_cont(" %s_%u", prefix, attr_meta->index);
+
+                if (attr_meta->src_offset)
+                        pandecode_log_cont("[%u]", attr_meta->src_offset);
+
+                pandecode_swizzle(attr_meta->swizzle, attr_meta->format);
+
+                pandecode_log_cont(";\n");
+        }
+
+        pandecode_log("\n");
+
+        return count ? (max_index + 1) : 0;
+}
+
+/* return bits [lo, hi) of word */
+static u32
+bits(u32 word, u32 lo, u32 hi)
+{
+        if (hi - lo >= 32)
+                return word; // avoid undefined behavior with the shift
+
+        return (word >> lo) & ((1 << (hi - lo)) - 1);
+}
+
+static void
+pandecode_vertex_tiler_prefix(struct mali_vertex_tiler_prefix *p, int job_no, bool graphics)
+{
+        pandecode_log(".prefix = {\n");
+        pandecode_indent++;
+
+        /* Decode invocation_count. See the comment before the definition of
+         * invocation_count for an explanation.
+         */
+
+        unsigned size_y_shift = bits(p->invocation_shifts, 0, 5);
+        unsigned size_z_shift = bits(p->invocation_shifts, 5, 10);
+        unsigned workgroups_x_shift = bits(p->invocation_shifts, 10, 16);
+        unsigned workgroups_y_shift = bits(p->invocation_shifts, 16, 22);
+        unsigned workgroups_z_shift = bits(p->invocation_shifts, 22, 28);
+        unsigned workgroups_x_shift_2 = bits(p->invocation_shifts, 28, 32);
+
+        unsigned size_x = bits(p->invocation_count, 0, size_y_shift) + 1;
+        unsigned size_y = bits(p->invocation_count, size_y_shift, size_z_shift) + 1;
+        unsigned size_z = bits(p->invocation_count, size_z_shift, workgroups_x_shift) + 1;
+
+        unsigned groups_x = bits(p->invocation_count, workgroups_x_shift, workgroups_y_shift) + 1;
+        unsigned groups_y = bits(p->invocation_count, workgroups_y_shift, workgroups_z_shift) + 1;
+        unsigned groups_z = bits(p->invocation_count, workgroups_z_shift, 32) + 1;
+
+        /* Even though we have this decoded, we want to ensure that the
+         * representation is "unique" so we don't lose anything by printing only
+         * the final result. More specifically, we need to check that we were
+         * passed something in canonical form, since the definition per the
+         * hardware is inherently not unique. How? Well, take the resulting
+         * decode and pack it ourselves! If it is bit exact with what we
+         * decoded, we're good to go. */
+
+        struct mali_vertex_tiler_prefix ref;
+        panfrost_pack_work_groups_compute(&ref, groups_x, groups_y, groups_z, size_x, size_y, size_z, graphics);
+
+        bool canonical =
+                (p->invocation_count == ref.invocation_count) &&
+                (p->invocation_shifts == ref.invocation_shifts);
+
+        if (!canonical) {
+                pandecode_msg("XXX: non-canonical workgroups packing\n");
+                pandecode_msg("expected: %X, %X",
+                                ref.invocation_count,
+                                ref.invocation_shifts);
+
+                pandecode_prop("invocation_count = 0x%" PRIx32, p->invocation_count);
+                pandecode_prop("size_y_shift = %d", size_y_shift);
+                pandecode_prop("size_z_shift = %d", size_z_shift);
+                pandecode_prop("workgroups_x_shift = %d", workgroups_x_shift);
+                pandecode_prop("workgroups_y_shift = %d", workgroups_y_shift);
+                pandecode_prop("workgroups_z_shift = %d", workgroups_z_shift);
+                pandecode_prop("workgroups_x_shift_2 = %d", workgroups_x_shift_2);
+        }
+
+        /* Regardless, print the decode */
+        pandecode_msg("size (%d, %d, %d), count (%d, %d, %d)\n",
+                        size_x, size_y, size_z,
+                        groups_x, groups_y, groups_z);
+
+        /* TODO: Decode */
+        if (p->unknown_draw)
+                pandecode_prop("unknown_draw = 0x%" PRIx32, p->unknown_draw);
+
+        pandecode_prop("workgroups_x_shift_3 = 0x%" PRIx32, p->workgroups_x_shift_3);
+
+        if (p->draw_mode != MALI_DRAW_NONE)
+                pandecode_prop("draw_mode = %s", pandecode_draw_mode(p->draw_mode));
+
+        /* Index count only exists for tiler jobs anyway */
+
+        if (p->index_count)
+                pandecode_prop("index_count = MALI_POSITIVE(%" PRId32 ")", p->index_count + 1);
+
+
+        unsigned index_raw_size = (p->unknown_draw & MALI_DRAW_INDEXED_SIZE);
+        index_raw_size >>= MALI_DRAW_INDEXED_SHIFT;
+
+        /* Validate an index buffer is present if we need one. TODO: verify
+         * relationship between invocation_count and index_count */
+
+        if (p->indices) {
+                unsigned count = p->index_count;
+
+                /* Grab the size */
+                unsigned size = (index_raw_size == 0x3) ? 4 : index_raw_size;
+
+                /* Ensure we got a size, and if so, validate the index buffer
+                 * is large enough to hold a full set of indices of the given
+                 * size */
+
+                if (!index_raw_size)
+                        pandecode_msg("XXX: index size missing\n");
+                else
+                        pandecode_validate_buffer(p->indices, count * size);
+        } else if (index_raw_size)
+                pandecode_msg("XXX: unexpected index size %u\n", index_raw_size);
+
+        if (p->offset_bias_correction)
+                pandecode_prop("offset_bias_correction = %d", p->offset_bias_correction);
+
+        /* TODO: Figure out what this is. It's not zero */
+        pandecode_prop("zero1 = 0x%" PRIx32, p->zero1);
+
+        pandecode_indent--;
+        pandecode_log("},\n");
+}
+
+static void
+pandecode_uniform_buffers(mali_ptr pubufs, int ubufs_count, int job_no)
+{
+        struct pandecode_mapped_memory *umem = pandecode_find_mapped_gpu_mem_containing(pubufs);
+        uint64_t *PANDECODE_PTR_VAR(ubufs, umem, pubufs);
+
+        for (int i = 0; i < ubufs_count; i++) {
+                unsigned size = (ubufs[i] & ((1 << 10) - 1)) * 16;
+                mali_ptr addr = (ubufs[i] >> 10) << 2;
+
+                pandecode_validate_buffer(addr, size);
+
+                char *ptr = pointer_as_memory_reference(addr);
+                pandecode_log("ubuf_%d[%u] = %s;\n", i, size, ptr);
+                free(ptr);
+        }
+
+        pandecode_log("\n");
+}
+
+static void
+pandecode_uniforms(mali_ptr uniforms, unsigned uniform_count)
+{
+        pandecode_validate_buffer(uniforms, uniform_count * 16);
+
+        char *ptr = pointer_as_memory_reference(uniforms);
+        pandecode_log("vec4 uniforms[%u] = %s;\n", uniform_count, ptr);
+        free(ptr);
+}
+
+static const char *
+shader_type_for_job(unsigned type)
+{
+        switch (type) {
+        case JOB_TYPE_VERTEX:  return "VERTEX";
+        case JOB_TYPE_TILER:   return "FRAGMENT";
+        case JOB_TYPE_COMPUTE: return "COMPUTE";
+        default:
+                               return "UNKNOWN";
+        }
+}
+
+static unsigned shader_id = 0;
+
+static struct midgard_disasm_stats
+pandecode_shader_disassemble(mali_ptr shader_ptr, int shader_no, int type,
+                             bool is_bifrost, unsigned gpu_id)
+{
+        struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(shader_ptr);
+        uint8_t *PANDECODE_PTR_VAR(code, mem, shader_ptr);
+
+        /* Compute maximum possible size */
+        size_t sz = mem->length - (shader_ptr - mem->gpu_va);
+
+        /* Print some boilerplate to clearly denote the assembly (which doesn't
+         * obey indentation rules), and actually do the disassembly! */
+
+        pandecode_log_cont("\n\n");
+
+        struct midgard_disasm_stats stats;
+
+        if (is_bifrost) {
+                disassemble_bifrost(pandecode_dump_stream, code, sz, true);
+
+                /* TODO: Extend stats to Bifrost */
+                stats.texture_count = -128;
+                stats.sampler_count = -128;
+                stats.attribute_count = -128;
+                stats.varying_count = -128;
+                stats.uniform_count = -128;
+                stats.uniform_buffer_count = -128;
+                stats.work_count = -128;
+
+                stats.instruction_count = 0;
+                stats.bundle_count = 0;
+                stats.quadword_count = 0;
+                stats.helper_invocations = false;
+        } else {
+                stats = disassemble_midgard(pandecode_dump_stream,
+                                code, sz, gpu_id,
+                                type == JOB_TYPE_TILER ?
+                                MESA_SHADER_FRAGMENT : MESA_SHADER_VERTEX);
+        }
+
+        /* Print shader-db stats. Skip COMPUTE jobs since they are used for
+         * driver-internal purposes with the blob and interfere */
+
+        bool should_shaderdb = type != JOB_TYPE_COMPUTE;
+
+        if (should_shaderdb) {
+                unsigned nr_threads =
+                        (stats.work_count <= 4) ? 4 :
+                        (stats.work_count <= 8) ? 2 :
+                        1;
+
+                pandecode_log_cont("shader%d - MESA_SHADER_%s shader: "
+                        "%u inst, %u bundles, %u quadwords, "
+                        "%u registers, %u threads, 0 loops, 0:0 spills:fills\n\n\n",
+                        shader_id++,
+                        shader_type_for_job(type),
+                        stats.instruction_count, stats.bundle_count, stats.quadword_count,
+                        stats.work_count, nr_threads);
+        }
+
+
+        return stats;
+}
+
+static void
+pandecode_texture_payload(mali_ptr payload,
+                          enum mali_texture_type type,
+                          enum mali_texture_layout layout,
+                          bool manual_stride,
+                          uint8_t levels,
+                          uint16_t depth,
+                          uint16_t array_size,
+                          struct pandecode_mapped_memory *tmem)
+{
+        pandecode_log(".payload = {\n");
+        pandecode_indent++;
+
+        /* A bunch of bitmap pointers follow.
+         * We work out the correct number,
+         * based on the mipmap/cubemap
+         * properties, but dump extra
+         * possibilities to futureproof */
+
+        int bitmap_count = levels + 1;
+
+        /* Miptree for each face */
+        if (type == MALI_TEX_CUBE)
+                bitmap_count *= 6;
+
+        /* Array of layers */
+        bitmap_count *= (depth + 1);
+
+        /* Array of textures */
+        bitmap_count *= (array_size + 1);
+
+        /* Stride for each element */
+        if (manual_stride)
+                bitmap_count *= 2;
+
+        mali_ptr *pointers_and_strides = pandecode_fetch_gpu_mem(tmem,
+                payload, sizeof(mali_ptr) * bitmap_count);
+        for (int i = 0; i < bitmap_count; ++i) {
+                /* How we dump depends if this is a stride or a pointer */
+
+                if (manual_stride && (i & 1)) {
+                        /* signed 32-bit snuck in as a 64-bit pointer */
+                        uint64_t stride_set = pointers_and_strides[i];
+                        uint32_t clamped_stride = stride_set;
+                        int32_t stride = clamped_stride;
+                        assert(stride_set == clamped_stride);
+                        pandecode_log("(mali_ptr) %d /* stride */, \n", stride);
+                } else {
+                        char *a = pointer_as_memory_reference(pointers_and_strides[i]);
+                        pandecode_log("%s, \n", a);
+                        free(a);
+                }
+        }
+
+        pandecode_indent--;
+        pandecode_log("},\n");
+}
+
+static void
+pandecode_texture(mali_ptr u,
+                struct pandecode_mapped_memory *tmem,
+                unsigned job_no, unsigned tex)
+{
+        struct mali_texture_descriptor *PANDECODE_PTR_VAR(t, tmem, u);
+
+        pandecode_log("struct mali_texture_descriptor texture_descriptor_%"PRIx64"_%d_%d = {\n", u, job_no, tex);
+        pandecode_indent++;
+
+        pandecode_prop("width = %" PRId32, t->width);
+        pandecode_prop("height = %" PRId32, t->height);
+        pandecode_prop("depth = %" PRId32, t->depth);
+        pandecode_prop("array_size = %" PRId32, t->array_size);
+
+        pandecode_log("\n");
+        pandecode_prop("f.swizzle = 0x%" PRIx32, t->format.swizzle);
+        pandecode_prop("f.format = 0x%" PRIx32, t->format.format);
+        pandecode_prop("f.srgb = 0x%" PRIx32, t->format.srgb);
+        pandecode_prop("f.unknown1 = 0x%" PRIx32, t->format.unknown1);
+        pandecode_prop("f.type = %" PRId32, t->format.type);
+        pandecode_prop("f.layout = %" PRId32, t->format.layout);
+        pandecode_prop("f.unknown2 = 0x%" PRIx32, t->format.unknown2);
+        pandecode_prop("f.manual_stride = %" PRId32, t->format.manual_stride);
+        pandecode_prop("f.zero = 0x%" PRIx32, t->format.zero);
+        pandecode_log("\n");
+
+        pandecode_prop("unknown3 = 0x%" PRIx32, t->unknown3);
+        pandecode_prop("unknown3A = 0x%" PRIx32, t->unknown3A);
+        pandecode_prop("levels = %" PRId32, t->levels);
+        pandecode_prop("swizzle = 0x%" PRIx32, t->swizzle);
+        pandecode_prop("swizzle_zero = 0x%" PRIx32, t->swizzle_zero);
+
+        pandecode_prop("unknown5 = 0x%" PRIx32, t->unknown5);
+        pandecode_prop("unknown6 = 0x%" PRIx32, t->unknown6);
+        pandecode_prop("unknown7 = 0x%" PRIx32, t->unknown7);
+        pandecode_log("\n");
+
+        struct mali_texture_format f = t->format;
+
+        /* See the definiton of enum mali_texture_type */
+
+        bool is_cube = f.type == MALI_TEX_CUBE;
+        unsigned dimension = is_cube ? 2 : f.type;
+
+        pandecode_make_indent();
+
+        /* Print the layout. Default is linear; a modifier can denote AFBC or
+         * u-interleaved/tiled modes */
+
+        if (f.layout == MALI_TEXTURE_AFBC)
+                pandecode_log_cont("afbc");
+        else if (f.layout == MALI_TEXTURE_TILED)
+                pandecode_log_cont("tiled");
+        else if (f.layout == MALI_TEXTURE_LINEAR)
+                pandecode_log_cont("linear");
+        else
+                pandecode_msg("XXX: invalid texture layout 0x%X\n", f.layout);
+
+        pandecode_swizzle(t->swizzle, f.format);
+        pandecode_log_cont(" ");
+
+        /* Distinguish cube/2D with modifier */
+
+        if (is_cube)
+                pandecode_log_cont("cube ");
+
+        pandecode_format_short(f.format, f.srgb);
+        pandecode_swizzle(f.swizzle, f.format);
+
+        /* All four width/height/depth/array_size dimensions are present
+         * regardless of the type of texture, but it is an error to have
+         * non-zero dimensions for unused dimensions. Verify this. array_size
+         * can always be set, as can width. Depth used for MSAA. */
+
+        if (t->height && dimension < 2)
+                pandecode_msg("XXX: nonzero height for <2D texture\n");
+
+        /* Print only the dimensions that are actually there */
+
+        pandecode_log_cont(": %d", t->width + 1);
+
+        if (t->height || t->depth)
+                pandecode_log_cont("x%u", t->height + 1);
+
+        if (t->depth)
+                pandecode_log_cont("x%u", t->depth + 1);
+
+        if (t->array_size)
+                pandecode_log_cont("[%u]", t->array_size + 1);
+
+        if (t->levels)
+                pandecode_log_cont(" mip %u", t->levels);
+
+        pandecode_log_cont("\n");
+
+        if (f.unknown1 | f.zero) {
+                pandecode_msg("XXX: texture format zero tripped\n");
+                pandecode_prop("unknown1 = %" PRId32, f.unknown1);
+                pandecode_prop("zero = %" PRId32, f.zero);
+        }
+
+        if (!f.unknown2) {
+                pandecode_msg("XXX: expected unknown texture bit set\n");
+                pandecode_prop("unknown2 = %" PRId32, f.unknown2);
+        }
+
+        if (t->swizzle_zero) {
+                pandecode_msg("XXX: swizzle zero tripped\n");
+                pandecode_prop("swizzle_zero = %d", t->swizzle_zero);
+        }
+
+        if (t->unknown3 | t->unknown3A | t->unknown5 | t->unknown6 | t->unknown7) {
+                pandecode_msg("XXX: texture zero tripped\n");
+                pandecode_prop("unknown3 = %" PRId16, t->unknown3);
+                pandecode_prop("unknown3A = %" PRId8, t->unknown3A);
+                pandecode_prop("unknown5 = 0x%" PRIx32, t->unknown5);
+                pandecode_prop("unknown6 = 0x%" PRIx32, t->unknown6);
+                pandecode_prop("unknown7 = 0x%" PRIx32, t->unknown7);
+        }
+
+        pandecode_texture_payload(u + sizeof(*t), f.type, f.layout, f.manual_stride, t->levels, t->depth, t->array_size, tmem);
+
+        pandecode_indent--;
+        pandecode_log("};\n");
+}
+
+static void
+pandecode_bifrost_texture(
+                const struct bifrost_texture_descriptor *t,
+                unsigned job_no,
+                unsigned tex)
+{
+        pandecode_log("struct bifrost_texture_descriptor texture_descriptor_%d_%d = {\n", job_no, tex);
+        pandecode_indent++;
+
+        pandecode_prop("format_unk = 0x%" PRIx32, t->format_unk);
+        pandecode_prop("type = %" PRId32, t->type);
+
+        if (t->zero) {
+                pandecode_msg("XXX: zero tripped\n");
+                pandecode_prop("zero = 0x%" PRIx32, t->zero);
+        }
+
+        pandecode_prop("format_swizzle = 0x%" PRIx32, t->format_swizzle);
+        pandecode_prop("format = 0x%" PRIx32, t->format);
+        pandecode_prop("srgb = 0x%" PRIx32, t->srgb);
+        pandecode_prop("format_unk3 = 0x%" PRIx32, t->format_unk3);
+        pandecode_prop("width = %" PRId32, t->width);
+        pandecode_prop("height = %" PRId32, t->height);
+        pandecode_prop("swizzle = 0x%" PRIx32, t->swizzle);
+        pandecode_prop("levels = %" PRId32, t->levels);
+        pandecode_prop("unk1 = 0x%" PRIx32, t->unk1);
+        pandecode_prop("levels_unk = %" PRId32, t->levels_unk);
+        pandecode_prop("level_2 = %" PRId32, t->level_2);
+        pandecode_prop("payload = 0x%" PRIx64, t->payload);
+        pandecode_prop("array_size = %" PRId32, t->array_size);
+        pandecode_prop("unk4 = 0x%" PRIx32, t->unk4);
+        pandecode_prop("depth = %" PRId32, t->depth);
+        pandecode_prop("unk5 = 0x%" PRIx32, t->unk5);
+        pandecode_log("\n");
+
+        /* See the definiton of enum mali_texture_type */
+
+        bool is_cube = t->type == MALI_TEX_CUBE;
+        unsigned dimension = is_cube ? 2 : t->type;
+
+        /* Print the layout. Default is linear; a modifier can denote AFBC or
+         * u-interleaved/tiled modes */
+
+        if (t->layout == MALI_TEXTURE_AFBC)
+                pandecode_log_cont("afbc");
+        else if (t->layout == MALI_TEXTURE_TILED)
+                pandecode_log_cont("tiled");
+        else if (t->layout == MALI_TEXTURE_LINEAR)
+                pandecode_log_cont("linear");
+        else
+                pandecode_msg("XXX: invalid texture layout 0x%X\n", t->layout);
+
+        pandecode_swizzle(t->swizzle, t->format);
+        pandecode_log_cont(" ");
+
+        /* Distinguish cube/2D with modifier */
+
+        if (is_cube)
+                pandecode_log_cont("cube ");
+
+        pandecode_format_short(t->format, t->srgb);
+
+        /* All four width/height/depth/array_size dimensions are present
+         * regardless of the type of texture, but it is an error to have
+         * non-zero dimensions for unused dimensions. Verify this. array_size
+         * can always be set, as can width. */
+
+        if (t->height && dimension < 2)
+                pandecode_msg("XXX: nonzero height for <2D texture\n");
+
+        if (t->depth && dimension < 3)
+                pandecode_msg("XXX: nonzero depth for <2D texture\n");
+
+        /* Print only the dimensions that are actually there */
+
+        pandecode_log_cont(": %d", t->width + 1);
+
+        if (dimension >= 2)
+                pandecode_log_cont("x%u", t->height + 1);
+
+        if (dimension >= 3)
+                pandecode_log_cont("x%u", t->depth + 1);
+
+        if (t->array_size)
+                pandecode_log_cont("[%u]", t->array_size + 1);
+
+        if (t->levels)
+                pandecode_log_cont(" mip %u", t->levels);
+
+        pandecode_log_cont("\n");
+
+        struct pandecode_mapped_memory *tmem = pandecode_find_mapped_gpu_mem_containing(t->payload);
+        if (t->payload) {
+                pandecode_texture_payload(t->payload, t->type, t->layout,
+                                          true, t->levels, t->depth,
+                                          t->array_size, tmem);
+        }
+
+        pandecode_indent--;
+        pandecode_log("};\n");
+}
+
+/* For shader properties like texture_count, we have a claimed property in the shader_meta, and the actual Truth from static analysis (this may just be an upper limit). We validate accordingly */
+
+static void
+pandecode_shader_prop(const char *name, unsigned claim, signed truth, bool fuzzy)
+{
+        /* Nothing to do */
+        if (claim == truth)
+                return;
+
+        if (fuzzy && (truth < 0))
+                pandecode_msg("XXX: fuzzy %s, claimed %d, expected %d\n", name, claim, truth);
+
+        if ((truth >= 0) && !fuzzy) {
+                pandecode_msg("%s: expected %s = %d, claimed %u\n",
+                                (truth < claim) ? "warn" : "XXX",
+                                name, truth, claim);
+        } else if ((claim > -truth) && !fuzzy) {
+                pandecode_msg("XXX: expected %s <= %u, claimed %u\n",
+                                name, -truth, claim);
+        } else if (fuzzy && (claim < truth))
+                pandecode_msg("XXX: expected %s >= %u, claimed %u\n",
+                                name, truth, claim);
+
+        pandecode_log(".%s = %" PRId16, name, claim);
+
+        if (fuzzy)
+                pandecode_log_cont(" /* %u used */", truth);
+
+        pandecode_log_cont(",\n");
+}
+
+static void
+pandecode_blend_shader_disassemble(mali_ptr shader, int job_no, int job_type,
+                                   bool is_bifrost, unsigned gpu_id)
+{
+        struct midgard_disasm_stats stats =
+                pandecode_shader_disassemble(shader, job_no, job_type, is_bifrost, gpu_id);
+
+        bool has_texture = (stats.texture_count > 0);
+        bool has_sampler = (stats.sampler_count > 0);
+        bool has_attribute = (stats.attribute_count > 0);
+        bool has_varying = (stats.varying_count > 0);
+        bool has_uniform = (stats.uniform_count > 0);
+        bool has_ubo = (stats.uniform_buffer_count > 0);
+
+        if (has_texture || has_sampler)
+                pandecode_msg("XXX: blend shader accessing textures\n");
+
+        if (has_attribute || has_varying)
+                pandecode_msg("XXX: blend shader accessing interstage\n");
+
+        if (has_uniform || has_ubo)
+                pandecode_msg("XXX: blend shader accessing uniforms\n");
+}
+
+static void
+pandecode_textures(mali_ptr textures, unsigned texture_count, int job_no, bool is_bifrost)
+{
+        struct pandecode_mapped_memory *mmem = pandecode_find_mapped_gpu_mem_containing(textures);
+
+        if (!mmem)
+                return;
+
+        if (is_bifrost) {
+                const struct bifrost_texture_descriptor *PANDECODE_PTR_VAR(t, mmem, textures);
+
+                pandecode_log("uint64_t textures_%"PRIx64"_%d[] = {\n", textures, job_no);
+                pandecode_indent++;
+
+                for (unsigned tex = 0; tex < texture_count; ++tex)
+                        pandecode_bifrost_texture(&t[tex], job_no, tex);
+
+                pandecode_indent--;
+                pandecode_log("};\n");
+        } else {
+                mali_ptr *PANDECODE_PTR_VAR(u, mmem, textures);
+
+                pandecode_log("uint64_t textures_%"PRIx64"_%d[] = {\n", textures, job_no);
+                pandecode_indent++;
+
+                for (int tex = 0; tex < texture_count; ++tex) {
+                        mali_ptr *PANDECODE_PTR_VAR(u, mmem, textures + tex * sizeof(mali_ptr));
+                        char *a = pointer_as_memory_reference(*u);
+                        pandecode_log("%s,\n", a);
+                        free(a);
+                }
+
+                pandecode_indent--;
+                pandecode_log("};\n");
+
+                /* Now, finally, descend down into the texture descriptor */
+                for (unsigned tex = 0; tex < texture_count; ++tex) {
+                        mali_ptr *PANDECODE_PTR_VAR(u, mmem, textures + tex * sizeof(mali_ptr));
+                        struct pandecode_mapped_memory *tmem = pandecode_find_mapped_gpu_mem_containing(*u);
+                        if (tmem)
+                                pandecode_texture(*u, tmem, job_no, tex);
+                }
+        }
+}
+
+static void
+pandecode_samplers(mali_ptr samplers, unsigned sampler_count, int job_no, bool is_bifrost)
+{
+        struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(samplers);
+
+        if (!smem)
+                return;
+
+        if (is_bifrost) {
+                struct bifrost_sampler_descriptor *s;
+
+                for (int i = 0; i < sampler_count; ++i) {
+                        s = pandecode_fetch_gpu_mem(smem, samplers + sizeof(*s) * i, sizeof(*s));
+
+                        pandecode_log("struct bifrost_sampler_descriptor sampler_descriptor_%"PRIx64"_%d_%d = {\n", samplers + sizeof(*s) * i, job_no, i);
+                        pandecode_indent++;
+
+                        if (s->unk1 != 1) {
+                                pandecode_msg("XXX: unk1 tripped\n");
+                                pandecode_prop("unk1 = 0x%x", s->unk1);
+                        }
+
+                        pandecode_prop("wrap_s = %s", pandecode_wrap_mode(s->wrap_s));
+                        pandecode_prop("wrap_t = %s", pandecode_wrap_mode(s->wrap_t));
+                        pandecode_prop("wrap_r = %s", pandecode_wrap_mode(s->wrap_r));
+
+                        if (s->unk8 != 0x8) {
+                                pandecode_msg("XXX: unk8 tripped\n");
+                                pandecode_prop("unk8 = 0x%x", s->unk8);
+                        }
+
+                        pandecode_prop("unk2 = 0x%x", s->unk2);
+                        pandecode_prop("unk3 = 0x%x", s->unk3);
+                        pandecode_prop("min_filter = %s", s->min_filter ? "nearest" : "linear");
+                        pandecode_prop("norm_coords = 0x%x", s->norm_coords & 0x1);
+                        pandecode_prop("zero1 = 0x%x", s->zero1 & 0x1);
+                        pandecode_prop("mip_filter = %s", s->mip_filter ? "linear" : "nearest");
+                        pandecode_prop("mag_filter = %s", s->mag_filter ? "linear" : "nearest");
+
+                        pandecode_prop("min_lod = FIXED_16(%f)", DECODE_FIXED_16(s->min_lod));
+                        pandecode_prop("max_lod = FIXED_16(%f)", DECODE_FIXED_16(s->max_lod));
+
+                        if (s->zero1 || s->zero2 || s->zero3 || s->zero4) {
+                                pandecode_msg("XXX: sampler zero tripped\n");
+                                pandecode_prop("zero = 0x%" PRIx8 ", 0x%" PRIx64 ", 0x%" PRIx64 ", 0x%" PRIx64 "\n", s->zero1, s->zero2, s->zero3, s->zero4);
+                        }
+
+                        pandecode_indent--;
+                        pandecode_log("};\n");
+                }
+        } else {
+                struct mali_sampler_descriptor *s;
+
+                for (int i = 0; i < sampler_count; ++i) {
+                        s = pandecode_fetch_gpu_mem(smem, samplers + sizeof(*s) * i, sizeof(*s));
+
+                        pandecode_log("struct mali_sampler_descriptor sampler_descriptor_%"PRIx64"_%d_%d = {\n", samplers + sizeof(*s) * i, job_no, i);
+                        pandecode_indent++;
+
+                        pandecode_log(".filter_mode = ");
+                        pandecode_log_decoded_flags(sampler_flag_info, s->filter_mode);
+                        pandecode_log_cont(",\n");
+
+                        pandecode_prop("min_lod = FIXED_16(%f)", DECODE_FIXED_16(s->min_lod));
+                        pandecode_prop("max_lod = FIXED_16(%f)", DECODE_FIXED_16(s->max_lod));
+
+                        if (s->lod_bias)
+                                pandecode_prop("lod_bias = FIXED_16(%f)", DECODE_FIXED_16(s->lod_bias));
+
+                        pandecode_prop("wrap_s = %s", pandecode_wrap_mode(s->wrap_s));
+                        pandecode_prop("wrap_t = %s", pandecode_wrap_mode(s->wrap_t));
+                        pandecode_prop("wrap_r = %s", pandecode_wrap_mode(s->wrap_r));
+
+                        pandecode_prop("compare_func = %s", pandecode_func(s->compare_func));
+
+                        if (s->zero || s->zero2) {
+                                pandecode_msg("XXX: sampler zero tripped\n");
+                                pandecode_prop("zero = 0x%X, 0x%X\n", s->zero, s->zero2);
+                        }
+
+                        pandecode_prop("seamless_cube_map = %d", s->seamless_cube_map);
+
+                        pandecode_prop("border_color = { %f, %f, %f, %f }",
+                                       s->border_color[0],
+                                       s->border_color[1],
+                                       s->border_color[2],
+                                       s->border_color[3]);
+
+                        pandecode_indent--;
+                        pandecode_log("};\n");
+                }
+        }
+}
+
+static void
+pandecode_vertex_tiler_postfix_pre(
+                const struct mali_vertex_tiler_postfix *p,
+                int job_no, enum mali_job_type job_type,
+                char *suffix, bool is_bifrost, unsigned gpu_id)
+{
+        struct pandecode_mapped_memory *attr_mem;
+
+        /* On Bifrost, since the tiler heap (for tiler jobs) and the scratchpad
+         * are the only things actually needed from the FBD, vertex/tiler jobs
+         * no longer reference the FBD -- instead, this field points to some
+         * info about the scratchpad.
+         */
+
+        struct pandecode_fbd fbd_info = {
+                /* Default for Bifrost */
+                .rt_count = 1
+        };
+
+        if (is_bifrost) {
+                pandecode_log_cont("\t/* %X %/\n", p->shared_memory & 1);
+                pandecode_compute_fbd(p->shared_memory & ~1, job_no);
+        } else if (p->shared_memory & MALI_MFBD)
+                fbd_info = pandecode_mfbd_bfr((u64) ((uintptr_t) p->shared_memory) & FBD_MASK, job_no, false, job_type == JOB_TYPE_COMPUTE, false);
+        else if (job_type == JOB_TYPE_COMPUTE)
+                pandecode_compute_fbd((u64) (uintptr_t) p->shared_memory, job_no);
+        else
+                fbd_info = pandecode_sfbd((u64) (uintptr_t) p->shared_memory, job_no, false, gpu_id);
+
+        int varying_count = 0, attribute_count = 0, uniform_count = 0, uniform_buffer_count = 0;
+        int texture_count = 0, sampler_count = 0;
+
+        if (p->shader) {
+                struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(p->shader);
+                struct mali_shader_meta *PANDECODE_PTR_VAR(s, smem, p->shader);
+
+                /* Disassemble ahead-of-time to get stats. Initialize with
+                 * stats for the missing-shader case so we get validation
+                 * there, too */
+
+                struct midgard_disasm_stats info = {
+                        .texture_count = 0,
+                        .sampler_count = 0,
+                        .attribute_count = 0,
+                        .varying_count = 0,
+                        .work_count = 1,
+
+                        .uniform_count = -128,
+                        .uniform_buffer_count = 0
+                };
+
+                if (s->shader & ~0xF)
+                        info = pandecode_shader_disassemble(s->shader & ~0xF, job_no, job_type, is_bifrost, gpu_id);
+
+                pandecode_log("struct mali_shader_meta shader_meta_%"PRIx64"_%d%s = {\n", p->shader, job_no, suffix);
+                pandecode_indent++;
+
+                /* Save for dumps */
+                attribute_count = s->attribute_count;
+                varying_count = s->varying_count;
+                texture_count = s->texture_count;
+                sampler_count = s->sampler_count;
+
+                if (is_bifrost) {
+                        uniform_count = s->bifrost2.uniform_count;
+                        uniform_buffer_count = s->bifrost1.uniform_buffer_count;
+                } else {
+                        uniform_count = s->midgard1.uniform_count;
+                        uniform_buffer_count = s->midgard1.uniform_buffer_count;
+                }
+
+                pandecode_shader_address("shader", s->shader);
+
+                pandecode_shader_prop("texture_count", s->texture_count, info.texture_count, false);
+                pandecode_shader_prop("sampler_count", s->sampler_count, info.sampler_count, false);
+                pandecode_shader_prop("attribute_count", s->attribute_count, info.attribute_count, false);
+                pandecode_shader_prop("varying_count", s->varying_count, info.varying_count, false);
+                pandecode_shader_prop("uniform_buffer_count",
+                                uniform_buffer_count,
+                                info.uniform_buffer_count, true);
+
+                if (!is_bifrost) {
+                        pandecode_shader_prop("uniform_count",
+                                        uniform_count,
+                                        info.uniform_count, false);
+
+                        pandecode_shader_prop("work_count",
+                                        s->midgard1.work_count, info.work_count, false);
+                }
+
+                if (is_bifrost) {
+                        pandecode_log("bifrost1.unk1 = ");
+                        pandecode_log_decoded_flags(shader_bifrost_info, s->bifrost1.unk1);
+                        pandecode_log_cont(",\n");
+                } else {
+                        bool helpers = s->midgard1.flags_lo & MALI_HELPER_INVOCATIONS;
+
+                        if (helpers != info.helper_invocations) {
+                                pandecode_msg("XXX: expected helpers %u but got %u\n",
+                                                info.helper_invocations, helpers);
+                        }
+
+                        pandecode_log(".midgard1.flags_lo = ");
+                        pandecode_log_decoded_flags(shader_midgard1_flag_lo_info,
+                                                    s->midgard1.flags_lo & ~MALI_HELPER_INVOCATIONS);
+                        pandecode_log_cont(",\n");
+
+                        pandecode_log(".midgard1.flags_hi = ");
+                        pandecode_log_decoded_flags(shader_midgard1_flag_hi_info, s->midgard1.flags_hi);
+                        pandecode_log_cont(",\n");
+                }
+
+                if (s->depth_units || s->depth_factor) {
+                        pandecode_prop("depth_factor = %f", s->depth_factor);
+                        pandecode_prop("depth_units = %f", s->depth_units);
+                }
+
+                if (s->coverage_mask)
+                        pandecode_prop("coverage_mask = 0x%X", s->coverage_mask);
+
+                if (s->unknown2_2)
+                        pandecode_prop(".unknown2_2 = %X", s->unknown2_2);
+
+                if (s->unknown2_3 || s->unknown2_4) {
+                        pandecode_log(".unknown2_3 = ");
+
+                        int unknown2_3 = s->unknown2_3;
+                        int unknown2_4 = s->unknown2_4;
+
+                        /* We're not quite sure what these flags mean without the depth test, if anything */
+
+                        if (unknown2_3 & (MALI_DEPTH_WRITEMASK | MALI_DEPTH_FUNC_MASK)) {
+                                const char *func = pandecode_func(MALI_GET_DEPTH_FUNC(unknown2_3));
+                                unknown2_3 &= ~MALI_DEPTH_FUNC_MASK;
+
+                                pandecode_log_cont("MALI_DEPTH_FUNC(%s) | ", func);
+                        }
+
+                        pandecode_log_decoded_flags(u3_flag_info, unknown2_3);
+                        pandecode_log_cont(",\n");
+
+                        pandecode_log(".unknown2_4 = ");
+                        pandecode_log_decoded_flags(u4_flag_info, unknown2_4);
+                        pandecode_log_cont(",\n");
+                }
+
+                if (s->stencil_mask_front || s->stencil_mask_back) {
+                        pandecode_prop("stencil_mask_front = 0x%02X", s->stencil_mask_front);
+                        pandecode_prop("stencil_mask_back = 0x%02X", s->stencil_mask_back);
+                }
+
+                pandecode_stencil("front", &s->stencil_front);
+                pandecode_stencil("back", &s->stencil_back);
+
+                if (is_bifrost) {
+                        pandecode_log(".bifrost2 = {\n");
+                        pandecode_indent++;
+
+                        pandecode_prop("unk3 = 0x%" PRIx32, s->bifrost2.unk3);
+                        pandecode_prop("preload_regs = 0x%" PRIx32, s->bifrost2.preload_regs);
+                        pandecode_prop("uniform_count = %" PRId32, s->bifrost2.uniform_count);
+                        pandecode_prop("unk4 = 0x%" PRIx32, s->bifrost2.unk4);
+
+                        pandecode_indent--;
+                        pandecode_log("},\n");
+                } else if (s->midgard2.unknown2_7) {
+                        pandecode_log(".midgard2 = {\n");
+                        pandecode_indent++;
+
+                        pandecode_prop("unknown2_7 = 0x%" PRIx32, s->midgard2.unknown2_7);
+                        pandecode_indent--;
+                        pandecode_log("},\n");
+                }
+
+                if (s->padding) {
+                        pandecode_msg("XXX: shader padding tripped\n");
+                        pandecode_prop("padding = 0x%" PRIx32, s->padding);
+                }
+
+                if (!is_bifrost) {
+                        /* TODO: Blend shaders routing/disasm */
+                        union midgard_blend blend = s->blend;
+                        mali_ptr shader = pandecode_midgard_blend(&blend, s->unknown2_3 & MALI_HAS_BLEND_SHADER);
+                        if (shader & ~0xF)
+                                pandecode_blend_shader_disassemble(shader, job_no, job_type, false, gpu_id);
+                } else {
+                        pandecode_msg("mdg_blend = %" PRIx64 "\n", s->blend.shader);
+                }
+
+                pandecode_indent--;
+                pandecode_log("};\n");
+
+                /* MRT blend fields are used whenever MFBD is used, with
+                 * per-RT descriptors */
+
+                if (job_type == JOB_TYPE_TILER && (is_bifrost || p->shared_memory & MALI_MFBD)) {
+                        void* blend_base = (void *) (s + 1);
+
+                        for (unsigned i = 0; i < fbd_info.rt_count; i++) {
+                                mali_ptr shader = 0;
+
+                                if (is_bifrost)
+                                        shader = pandecode_bifrost_blend(blend_base, job_no, i);
+                                else
+                                        shader = pandecode_midgard_blend_mrt(blend_base, job_no, i);
+
+                                if (shader & ~0xF)
+                                        pandecode_blend_shader_disassemble(shader, job_no, job_type, false, gpu_id);
+
+                        }
+                }
+        } else
+                pandecode_msg("XXX: missing shader descriptor\n");
+
+        if (p->viewport) {
+                struct pandecode_mapped_memory *fmem = pandecode_find_mapped_gpu_mem_containing(p->viewport);
+                struct mali_viewport *PANDECODE_PTR_VAR(f, fmem, p->viewport);
+
+                pandecode_log("struct mali_viewport viewport_%"PRIx64"_%d%s = {\n", p->viewport, job_no, suffix);
+                pandecode_indent++;
+
+                pandecode_prop("clip_minx = %f", f->clip_minx);
+                pandecode_prop("clip_miny = %f", f->clip_miny);
+                pandecode_prop("clip_minz = %f", f->clip_minz);
+                pandecode_prop("clip_maxx = %f", f->clip_maxx);
+                pandecode_prop("clip_maxy = %f", f->clip_maxy);
+                pandecode_prop("clip_maxz = %f", f->clip_maxz);
+
+                /* Only the higher coordinates are MALI_POSITIVE scaled */
+
+                pandecode_prop("viewport0 = { %d, %d }",
+                               f->viewport0[0], f->viewport0[1]);
+
+                pandecode_prop("viewport1 = { MALI_POSITIVE(%d), MALI_POSITIVE(%d) }",
+                               f->viewport1[0] + 1, f->viewport1[1] + 1);
+
+                pandecode_indent--;
+                pandecode_log("};\n");
+        }
+
+        unsigned max_attr_index = 0;
+
+        if (p->attribute_meta)
+                max_attr_index = pandecode_attribute_meta(job_no, attribute_count, p, false, suffix);
+
+        if (p->attributes) {
+                attr_mem = pandecode_find_mapped_gpu_mem_containing(p->attributes);
+                pandecode_attributes(attr_mem, p->attributes, job_no, suffix, max_attr_index, false, job_type);
+        }
+
+        /* Varyings are encoded like attributes but not actually sent; we just
+         * pass a zero buffer with the right stride/size set, (or whatever)
+         * since the GPU will write to it itself */
+
+        if (p->varying_meta) {
+                varying_count = pandecode_attribute_meta(job_no, varying_count, p, true, suffix);
+        }
+
+        if (p->varyings) {
+                attr_mem = pandecode_find_mapped_gpu_mem_containing(p->varyings);
+
+                /* Number of descriptors depends on whether there are
+                 * non-internal varyings */
+
+                pandecode_attributes(attr_mem, p->varyings, job_no, suffix, varying_count, true, job_type);
+        }
+
+        if (p->uniform_buffers) {
+                if (uniform_buffer_count)
+                        pandecode_uniform_buffers(p->uniform_buffers, uniform_buffer_count, job_no);
+                else
+                        pandecode_msg("warn: UBOs specified but not referenced\n");
+        } else if (uniform_buffer_count)
+                pandecode_msg("XXX: UBOs referenced but not specified\n");
+
+        /* We don't want to actually dump uniforms, but we do need to validate
+         * that the counts we were given are sane */
+
+        if (p->uniforms) {
+                if (uniform_count)
+                        pandecode_uniforms(p->uniforms, uniform_count);
+                else
+                        pandecode_msg("warn: Uniforms specified but not referenced\n");
+        } else if (uniform_count)
+                pandecode_msg("XXX: Uniforms referenced but not specified\n");
+
+        if (p->textures)
+                pandecode_textures(p->textures, texture_count, job_no, is_bifrost);
+
+        if (p->sampler_descriptor)
+                pandecode_samplers(p->sampler_descriptor, sampler_count, job_no, is_bifrost);
+}
+
+static void
+pandecode_gl_enables(uint32_t gl_enables, int job_type)
+{
+        pandecode_log(".gl_enables = ");
+
+        pandecode_log_decoded_flags(gl_enable_flag_info, gl_enables);
+
+        pandecode_log_cont(",\n");
+}
+
+static void
+pandecode_vertex_tiler_postfix(const struct mali_vertex_tiler_postfix *p, int job_no, bool is_bifrost)
+{
+        if (p->shader & 0xF)
+                pandecode_msg("warn: shader tagged %X\n", (unsigned) (p->shader & 0xF));
+
+        pandecode_log(".postfix = {\n");
+        pandecode_indent++;
+
+        pandecode_gl_enables(p->gl_enables, JOB_TYPE_TILER);
+        pandecode_prop("instance_shift = 0x%x", p->instance_shift);
+        pandecode_prop("instance_odd = 0x%x", p->instance_odd);
+
+        if (p->zero4) {
+                pandecode_msg("XXX: vertex only zero tripped");
+                pandecode_prop("zero4 = 0x%" PRIx32, p->zero4);
+        }
+
+        pandecode_prop("offset_start = 0x%x", p->offset_start);
+
+        if (p->zero5) {
+                pandecode_msg("XXX: vertex only zero tripped");
+                pandecode_prop("zero5 = 0x%" PRIx64, p->zero5);
+        }
+
+        MEMORY_PROP(p, position_varying);
+        MEMORY_PROP(p, occlusion_counter);
+
+        pandecode_indent--;
+        pandecode_log("},\n");
+}
+
+static void
+pandecode_tiler_heap_meta(mali_ptr gpu_va, int job_no)
+{
+        struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
+        const struct bifrost_tiler_heap_meta *PANDECODE_PTR_VAR(h, mem, gpu_va);
+
+        pandecode_log("struct bifrost_tiler_heap_meta tiler_heap_meta_%"PRIx64"_%d = {\n", gpu_va, job_no);
+        pandecode_indent++;
+
+        if (h->zero) {
+                pandecode_msg("XXX: tiler heap zero tripped\n");
+                pandecode_prop("zero = 0x%x", h->zero);
+        }
+
+        pandecode_prop("heap_size = 0x%x", h->heap_size);
+        MEMORY_PROP(h, tiler_heap_start);
+        MEMORY_PROP(h, tiler_heap_free);
+
+        /* this might point to the beginning of another buffer, when it's
+         * really the end of the tiler heap buffer, so we have to be careful
+         * here. but for zero length, we need the same pointer.
+         */
+
+        if (h->tiler_heap_end == h->tiler_heap_start) {
+                MEMORY_PROP(h, tiler_heap_start);
+        } else {
+                char *a = pointer_as_memory_reference(h->tiler_heap_end - 1);
+                pandecode_prop("tiler_heap_end = %s + 1", a);
+                free(a);
+        }
+
+        for (int i = 0; i < 10; i++) {
+                if (h->zeros[i] != 0) {
+                        pandecode_msg("XXX: tiler heap zero %d tripped, value %x\n",
+                                      i, h->zeros[i]);
+                }
+        }
+
+        if (h->unk1 != 0x1) {
+                pandecode_msg("XXX: tiler heap unk1 tripped\n");
+                pandecode_prop("unk1 = 0x%x", h->unk1);
+        }
+
+        if (h->unk7e007e != 0x7e007e) {
+                pandecode_msg("XXX: tiler heap unk7e007e tripped\n");
+                pandecode_prop("unk7e007e = 0x%x", h->unk7e007e);
+        }
+
+        pandecode_indent--;
+        pandecode_log("};\n");
+}
+
+static void
+pandecode_tiler_meta(mali_ptr gpu_va, int job_no)
+{
+        struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
+        const struct bifrost_tiler_meta *PANDECODE_PTR_VAR(t, mem, gpu_va);
+
+        pandecode_tiler_heap_meta(t->tiler_heap_meta, job_no);
+
+        pandecode_log("struct bifrost_tiler_meta tiler_meta_%"PRIx64"_%d = {\n", gpu_va, job_no);
+        pandecode_indent++;
+
+        pandecode_prop("tiler_heap_next_start = 0x%" PRIx32, t->tiler_heap_next_start);
+        pandecode_prop("used_hierarchy_mask = 0x%" PRIx32, t->used_hierarchy_mask);
+
+        if (t->hierarchy_mask != 0xa &&
+            t->hierarchy_mask != 0x14 &&
+            t->hierarchy_mask != 0x28 &&
+            t->hierarchy_mask != 0x50 &&
+            t->hierarchy_mask != 0xa0)
+                pandecode_prop("XXX: Unexpected hierarchy_mask (not 0xa, 0x14, 0x28, 0x50 or 0xa0)!");
+
+        pandecode_prop("hierarchy_mask = 0x%" PRIx16, t->hierarchy_mask);
+
+        pandecode_prop("flags = 0x%" PRIx16, t->flags);
+
+        pandecode_prop("width = MALI_POSITIVE(%d)", t->width + 1);
+        pandecode_prop("height = MALI_POSITIVE(%d)", t->height + 1);
+
+        if (t->zero0) {
+                pandecode_msg("XXX: tiler meta zero tripped\n");
+                pandecode_prop("zero0 = 0x%" PRIx64, t->zero0);
+        }
+
+        for (int i = 0; i < 12; i++) {
+                if (t->zeros[i] != 0) {
+                        pandecode_msg("XXX: tiler heap zero %d tripped, value %" PRIx64 "\n",
+                                      i, t->zeros[i]);
+                }
+        }
+
+        pandecode_indent--;
+        pandecode_log("};\n");
+}
+
+static void
+pandecode_primitive_size(union midgard_primitive_size u, bool constant)
+{
+        if (u.pointer == 0x0)
+                return;
+
+        pandecode_log(".primitive_size = {\n");
+        pandecode_indent++;
+
+        if (constant) {
+                pandecode_prop("constant = %f", u.constant);
+        } else {
+                MEMORY_PROP((&u), pointer);
+        }
+
+        pandecode_indent--;
+        pandecode_log("},\n");
+}
+
+static void
+pandecode_tiler_only_bfr(const struct bifrost_tiler_only *t, int job_no)
+{
+        pandecode_log_cont("{\n");
+        pandecode_indent++;
+
+        /* TODO: gl_PointSize on Bifrost */
+        pandecode_primitive_size(t->primitive_size, true);
+
+        if (t->zero1 || t->zero2 || t->zero3 || t->zero4 || t->zero5
+            || t->zero6) {
+                pandecode_msg("XXX: tiler only zero tripped\n");
+                pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
+                pandecode_prop("zero2 = 0x%" PRIx64, t->zero2);
+                pandecode_prop("zero3 = 0x%" PRIx64, t->zero3);
+                pandecode_prop("zero4 = 0x%" PRIx64, t->zero4);
+                pandecode_prop("zero5 = 0x%" PRIx64, t->zero5);
+                pandecode_prop("zero6 = 0x%" PRIx64, t->zero6);
+        }
+
+        pandecode_indent--;
+        pandecode_log("},\n");
+}
+
+static int
+pandecode_vertex_job_bfr(const struct mali_job_descriptor_header *h,
+                                const struct pandecode_mapped_memory *mem,
+                                mali_ptr payload, int job_no, unsigned gpu_id)
+{
+        struct bifrost_payload_vertex *PANDECODE_PTR_VAR(v, mem, payload);
+
+        pandecode_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", true, gpu_id);
+
+        pandecode_log("struct bifrost_payload_vertex payload_%"PRIx64"_%d = {\n", payload, job_no);
+        pandecode_indent++;
+
+        pandecode_vertex_tiler_prefix(&v->prefix, job_no, false);
+        pandecode_vertex_tiler_postfix(&v->postfix, job_no, true);
+
+        pandecode_indent--;
+        pandecode_log("};\n");
+
+        return sizeof(*v);
+}
+
+static int
+pandecode_tiler_job_bfr(const struct mali_job_descriptor_header *h,
+                               const struct pandecode_mapped_memory *mem,
+                               mali_ptr payload, int job_no, unsigned gpu_id)
+{
+        struct bifrost_payload_tiler *PANDECODE_PTR_VAR(t, mem, payload);
+
+        pandecode_vertex_tiler_postfix_pre(&t->postfix, job_no, h->job_type, "", true, gpu_id);
+        pandecode_tiler_meta(t->tiler.tiler_meta, job_no);
+
+        pandecode_log("struct bifrost_payload_tiler payload_%"PRIx64"_%d = {\n", payload, job_no);
+        pandecode_indent++;
+
+        pandecode_vertex_tiler_prefix(&t->prefix, job_no, false);
+
+        pandecode_log(".tiler = ");
+        pandecode_tiler_only_bfr(&t->tiler, job_no);
+
+        pandecode_vertex_tiler_postfix(&t->postfix, job_no, true);
+
+        pandecode_indent--;
+        pandecode_log("};\n");
+
+        return sizeof(*t);
+}
+
+static int
+pandecode_vertex_or_tiler_job_mdg(const struct mali_job_descriptor_header *h,
+                const struct pandecode_mapped_memory *mem,
+                mali_ptr payload, int job_no, unsigned gpu_id)
+{
+        struct midgard_payload_vertex_tiler *PANDECODE_PTR_VAR(v, mem, payload);
+        bool is_graphics = (h->job_type == JOB_TYPE_VERTEX) || (h->job_type == JOB_TYPE_TILER);
+
+        pandecode_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", false, gpu_id);
+
+        pandecode_log("struct midgard_payload_vertex_tiler payload_%d = {\n", job_no);
+        pandecode_indent++;
+
+        pandecode_vertex_tiler_prefix(&v->prefix, job_no, is_graphics);
+        pandecode_vertex_tiler_postfix(&v->postfix, job_no, false);
+
+        bool has_primitive_pointer = v->prefix.unknown_draw & MALI_DRAW_VARYING_SIZE;
+        pandecode_primitive_size(v->primitive_size, !has_primitive_pointer);
+
+        pandecode_indent--;
+        pandecode_log("};\n");
+
+        return sizeof(*v);
+}
+
+static int
+pandecode_fragment_job(const struct pandecode_mapped_memory *mem,
+                              mali_ptr payload, int job_no,
+                              bool is_bifrost, unsigned gpu_id)
+{
+        const struct mali_payload_fragment *PANDECODE_PTR_VAR(s, mem, payload);
+
+        bool is_mfbd = s->framebuffer & MALI_MFBD;
+
+        if (!is_mfbd && is_bifrost)
+                pandecode_msg("XXX: Bifrost fragment must use MFBD\n");
+
+        struct pandecode_fbd info;
+
+        if (is_mfbd)
+                info = pandecode_mfbd_bfr(s->framebuffer & FBD_MASK, job_no, true, false, is_bifrost);
+        else
+                info = pandecode_sfbd(s->framebuffer & FBD_MASK, job_no, true, gpu_id);
+
+        /* Compute the tag for the tagged pointer. This contains the type of
+         * FBD (MFBD/SFBD), and in the case of an MFBD, information about which
+         * additional structures follow the MFBD header (an extra payload or
+         * not, as well as a count of render targets) */
+
+        unsigned expected_tag = is_mfbd ? MALI_MFBD : 0;
+
+        if (is_mfbd) {
+                if (info.has_extra)
+                        expected_tag |= MALI_MFBD_TAG_EXTRA;
+
+                expected_tag |= (MALI_POSITIVE(info.rt_count) << 2);
+        }
+
+        if ((s->min_tile_coord | s->max_tile_coord) & ~(MALI_X_COORD_MASK | MALI_Y_COORD_MASK)) {
+                pandecode_msg("XXX: unexpected tile coordinate bits\n");
+                pandecode_prop("min_tile_coord = 0x%X\n", s->min_tile_coord);
+                pandecode_prop("max_tile_coord = 0x%X\n", s->max_tile_coord);
+        }
+
+        /* Extract tile coordinates */
+
+        unsigned min_x = MALI_TILE_COORD_X(s->min_tile_coord) << MALI_TILE_SHIFT;
+        unsigned min_y = MALI_TILE_COORD_Y(s->min_tile_coord) << MALI_TILE_SHIFT;
+
+        unsigned max_x = (MALI_TILE_COORD_X(s->max_tile_coord) + 1) << MALI_TILE_SHIFT;
+        unsigned max_y = (MALI_TILE_COORD_Y(s->max_tile_coord) + 1) << MALI_TILE_SHIFT;
+
+        /* For the max, we also want the floored (rather than ceiled) version for checking */
+
+        unsigned max_x_f = (MALI_TILE_COORD_X(s->max_tile_coord)) << MALI_TILE_SHIFT;
+        unsigned max_y_f = (MALI_TILE_COORD_Y(s->max_tile_coord)) << MALI_TILE_SHIFT;
+
+        /* Validate the coordinates are well-ordered */
+
+        if (min_x == max_x)
+                pandecode_msg("XXX: empty X coordinates (%u = %u)\n", min_x, max_x);
+        else if (min_x > max_x)
+                pandecode_msg("XXX: misordered X coordinates (%u > %u)\n", min_x, max_x);
+
+        if (min_y == max_y)
+                pandecode_msg("XXX: empty X coordinates (%u = %u)\n", min_x, max_x);
+        else if (min_y > max_y)
+                pandecode_msg("XXX: misordered X coordinates (%u > %u)\n", min_x, max_x);
+
+        /* Validate the coordinates fit inside the framebuffer. We use floor,
+         * rather than ceil, for the max coordinates, since the tile
+         * coordinates for something like an 800x600 framebuffer will actually
+         * resolve to 800x608, which would otherwise trigger a Y-overflow */
+
+        if ((min_x > info.width) || (max_x_f > info.width))
+                pandecode_msg("XXX: tile coordinates overflow in X direction\n");
+
+        if ((min_y > info.height) || (max_y_f > info.height))
+                pandecode_msg("XXX: tile coordinates overflow in Y direction\n");
+
+        /* After validation, we print */
+
+        pandecode_log("fragment (%u, %u) ... (%u, %u)\n\n", min_x, min_y, max_x, max_y);
+
+        /* The FBD is a tagged pointer */
+
+        unsigned tag = (s->framebuffer & ~FBD_MASK);
+
+        if (tag != expected_tag)
+                pandecode_msg("XXX: expected FBD tag %X but got %X\n", expected_tag, tag);
+
+        return sizeof(*s);
+}
+
+/* Entrypoint to start tracing. jc_gpu_va is the GPU address for the first job
+ * in the chain; later jobs are found by walking the chain. Bifrost is, well,
+ * if it's bifrost or not. GPU ID is the more finegrained ID (at some point, we
+ * might wish to combine this with the bifrost parameter) because some details
+ * are model-specific even within a particular architecture. Minimal traces
+ * *only* examine the job descriptors, skipping printing entirely if there is
+ * no faults, and only descends into the payload if there are faults. This is
+ * useful for looking for faults without the overhead of invasive traces. */
+
+void
+pandecode_jc(mali_ptr jc_gpu_va, bool bifrost, unsigned gpu_id, bool minimal)
+{
+        pandecode_dump_file_open();
+
+        struct mali_job_descriptor_header *h;
+        unsigned job_descriptor_number = 0;
+
+        do {
+                struct pandecode_mapped_memory *mem =
+                        pandecode_find_mapped_gpu_mem_containing(jc_gpu_va);
+
+                void *payload;
+
+                h = PANDECODE_PTR(mem, jc_gpu_va, struct mali_job_descriptor_header);
+
+                /* On Midgard, for 32-bit jobs except for fragment jobs, the
+                 * high 32-bits of the 64-bit pointer are reused to store
+                 * something else.
+                 */
+                int offset = h->job_descriptor_size == MALI_JOB_32 &&
+                             h->job_type != JOB_TYPE_FRAGMENT ? 4 : 0;
+                mali_ptr payload_ptr = jc_gpu_va + sizeof(*h) - offset;
+
+                payload = pandecode_fetch_gpu_mem(mem, payload_ptr, 256);
+
+                int job_no = job_descriptor_number++;
+
+                /* If the job is good to go, skip it in minimal mode */
+                if (minimal && (h->exception_status == 0x0 || h->exception_status == 0x1))
+                        continue;
+
+                pandecode_log("struct mali_job_descriptor_header job_%"PRIx64"_%d = {\n", jc_gpu_va, job_no);
+                pandecode_indent++;
+
+                pandecode_prop("job_type = %s", pandecode_job_type(h->job_type));
+
+                if (h->job_descriptor_size)
+                        pandecode_prop("job_descriptor_size = %d", h->job_descriptor_size);
+
+                if (h->exception_status && h->exception_status != 0x1)
+                        pandecode_prop("exception_status = %x (source ID: 0x%x access: %s exception: 0x%x)",
+                                       h->exception_status,
+                                       (h->exception_status >> 16) & 0xFFFF,
+                                       pandecode_exception_access((h->exception_status >> 8) & 0x3),
+                                       h->exception_status  & 0xFF);
+
+                if (h->first_incomplete_task)
+                        pandecode_prop("first_incomplete_task = %d", h->first_incomplete_task);
+
+                if (h->fault_pointer)
+                        pandecode_prop("fault_pointer = 0x%" PRIx64, h->fault_pointer);
+
+                if (h->job_barrier)
+                        pandecode_prop("job_barrier = %d", h->job_barrier);
+
+                pandecode_prop("job_index = %d", h->job_index);
+
+                if (h->unknown_flags)
+                        pandecode_prop("unknown_flags = %d", h->unknown_flags);
+
+                if (h->job_dependency_index_1)
+                        pandecode_prop("job_dependency_index_1 = %d", h->job_dependency_index_1);
+
+                if (h->job_dependency_index_2)
+                        pandecode_prop("job_dependency_index_2 = %d", h->job_dependency_index_2);
+
+                pandecode_indent--;
+                pandecode_log("};\n");
+
+                switch (h->job_type) {
+                case JOB_TYPE_WRITE_VALUE: {
+                        struct mali_payload_write_value *s = payload;
+                        pandecode_log("struct mali_payload_write_value payload_%"PRIx64"_%d = {\n", payload_ptr, job_no);
+                        pandecode_indent++;
+                        MEMORY_PROP(s, address);
+
+                        if (s->value_descriptor != MALI_WRITE_VALUE_ZERO) {
+                                pandecode_msg("XXX: unknown value descriptor\n");
+                                pandecode_prop("value_descriptor = 0x%" PRIX32, s->value_descriptor);
+                        }
+
+                        if (s->reserved) {
+                                pandecode_msg("XXX: set value tripped\n");
+                                pandecode_prop("reserved = 0x%" PRIX32, s->reserved);
+                        }
+
+                        pandecode_prop("immediate = 0x%" PRIX64, s->immediate);
+                        pandecode_indent--;
+                        pandecode_log("};\n");
+
+                        break;
+                }
+
+                case JOB_TYPE_TILER:
+                case JOB_TYPE_VERTEX:
+                case JOB_TYPE_COMPUTE:
+                        if (bifrost) {
+                                if (h->job_type == JOB_TYPE_TILER)
+                                        pandecode_tiler_job_bfr(h, mem, payload_ptr, job_no, gpu_id);
+                                else
+                                        pandecode_vertex_job_bfr(h, mem, payload_ptr, job_no, gpu_id);
+                        } else
+                                pandecode_vertex_or_tiler_job_mdg(h, mem, payload_ptr, job_no, gpu_id);
+
+                        break;
+
+                case JOB_TYPE_FRAGMENT:
+                        pandecode_fragment_job(mem, payload_ptr, job_no, bifrost, gpu_id);
+                        break;
+
+                default:
+                        break;
+                }
+        } while ((jc_gpu_va = h->next_job));
+
+        pandecode_map_read_write();
+}
diff --git a/src/panfrost/lib/decode.h b/src/panfrost/lib/decode.h
new file mode 100644 (file)
index 0000000..2b9bcb2
--- /dev/null
@@ -0,0 +1,85 @@
+/*
+ * Copyright (C) 2017-2019 Lyude Paul
+ * Copyright (C) 2017-2019 Alyssa Rosenzweig
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef __PAN_DECODE_H__
+#define __PAN_DECODE_H__
+
+#include "wrap.h"
+#include "util/list.h"
+
+extern FILE *pandecode_dump_stream;
+
+void pandecode_dump_file_open(void);
+
+struct pandecode_mapped_memory {
+        size_t length;
+        void *addr;
+        uint64_t gpu_va;
+        bool ro;
+        char name[32];
+};
+
+char *pointer_as_memory_reference(uint64_t ptr);
+
+struct pandecode_mapped_memory *pandecode_find_mapped_gpu_mem_containing(uint64_t addr);
+
+void pandecode_map_read_write(void);
+
+static inline void *
+__pandecode_fetch_gpu_mem(const struct pandecode_mapped_memory *mem,
+                          uint64_t gpu_va, size_t size,
+                          int line, const char *filename)
+{
+        if (!mem)
+                mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
+
+        if (!mem) {
+                fprintf(stderr, "Access to unknown memory %" PRIx64 " in %s:%d",
+                        gpu_va, filename, line);
+                assert(0);
+        }
+
+        assert(mem);
+        assert(size + (gpu_va - mem->gpu_va) <= mem->length);
+
+        return mem->addr + gpu_va - mem->gpu_va;
+}
+
+#define pandecode_fetch_gpu_mem(mem, gpu_va, size) \
+       __pandecode_fetch_gpu_mem(mem, gpu_va, size, __LINE__, __FILE__)
+
+/* Returns a validated pointer to mapped GPU memory with the given pointer type,
+ * size automatically determined from the pointer type
+ */
+#define PANDECODE_PTR(mem, gpu_va, type) \
+       ((type*)(__pandecode_fetch_gpu_mem(mem, gpu_va, sizeof(type), \
+                                        __LINE__, __FILE__)))
+
+/* Usage: <variable type> PANDECODE_PTR_VAR(name, mem, gpu_va) */
+#define PANDECODE_PTR_VAR(name, mem, gpu_va) \
+       name = __pandecode_fetch_gpu_mem(mem, gpu_va, sizeof(*name), \
+                                      __LINE__, __FILE__)
+
+#endif /* __MMAP_TRACE_H__ */
diff --git a/src/panfrost/lib/decode_common.c b/src/panfrost/lib/decode_common.c
new file mode 100644 (file)
index 0000000..14cb2f0
--- /dev/null
@@ -0,0 +1,200 @@
+/*
+ * Copyright (C) 2019 Alyssa Rosenzweig
+ * Copyright (C) 2017-2018 Lyude Paul
+ * Copyright (C) 2019 Collabora, Ltd.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <assert.h>
+#include <stdint.h>
+#include <string.h>
+#include <sys/mman.h>
+
+#include "decode.h"
+#include "util/macros.h"
+#include "util/u_debug.h"
+#include "util/u_dynarray.h"
+#include "util/hash_table.h"
+
+/* Memory handling */
+
+static struct hash_table_u64 *mmap_table;
+
+static struct util_dynarray ro_mappings;
+
+static struct pandecode_mapped_memory *
+pandecode_find_mapped_gpu_mem_containing_rw(uint64_t addr)
+{
+        return _mesa_hash_table_u64_search(mmap_table, addr & ~(4096 - 1));
+}
+
+struct pandecode_mapped_memory *
+pandecode_find_mapped_gpu_mem_containing(uint64_t addr)
+{
+        struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing_rw(addr);
+
+        if (mem && mem->addr && !mem->ro) {
+                mprotect(mem->addr, mem->length, PROT_READ);
+                mem->ro = true;
+                util_dynarray_append(&ro_mappings, struct pandecode_mapped_memory *, mem);
+        }
+
+        return mem;
+}
+
+void
+pandecode_map_read_write(void)
+{
+        util_dynarray_foreach(&ro_mappings, struct pandecode_mapped_memory *, mem) {
+                (*mem)->ro = false;
+                mprotect((*mem)->addr, (*mem)->length, PROT_READ | PROT_WRITE);
+        }
+        util_dynarray_clear(&ro_mappings);
+}
+
+static void
+pandecode_add_name(struct pandecode_mapped_memory *mem, uint64_t gpu_va, const char *name)
+{
+        if (!name) {
+                /* If we don't have a name, assign one */
+
+                snprintf(mem->name, ARRAY_SIZE(mem->name) - 1,
+                         "memory_%" PRIx64, gpu_va);
+        } else {
+                assert((strlen(name) + 1) < ARRAY_SIZE(mem->name));
+                memcpy(mem->name, name, strlen(name) + 1);
+        }
+}
+
+void
+pandecode_inject_mmap(uint64_t gpu_va, void *cpu, unsigned sz, const char *name)
+{
+        /* First, search if we already mapped this and are just updating an address */
+
+        struct pandecode_mapped_memory *existing =
+                pandecode_find_mapped_gpu_mem_containing_rw(gpu_va);
+
+        if (existing && existing->gpu_va == gpu_va) {
+                existing->length = sz;
+                existing->addr = cpu;
+                pandecode_add_name(existing, gpu_va, name);
+                return;
+        }
+
+        /* Otherwise, add a fresh mapping */
+        struct pandecode_mapped_memory *mapped_mem = NULL;
+
+        mapped_mem = malloc(sizeof(*mapped_mem));
+        mapped_mem->gpu_va = gpu_va;
+        mapped_mem->length = sz;
+        mapped_mem->addr = cpu;
+        pandecode_add_name(mapped_mem, gpu_va, name);
+
+        /* Add it to the table */
+        assert((gpu_va & 4095) == 0);
+
+        for (unsigned i = 0; i < sz; i += 4096)
+                _mesa_hash_table_u64_insert(mmap_table, gpu_va + i, mapped_mem);
+}
+
+char *
+pointer_as_memory_reference(uint64_t ptr)
+{
+        struct pandecode_mapped_memory *mapped;
+        char *out = malloc(128);
+
+        /* Try to find the corresponding mapped zone */
+
+        mapped = pandecode_find_mapped_gpu_mem_containing_rw(ptr);
+
+        if (mapped) {
+                snprintf(out, 128, "%s + %d", mapped->name, (int) (ptr - mapped->gpu_va));
+                return out;
+        }
+
+        /* Just use the raw address if other options are exhausted */
+
+        snprintf(out, 128, "0x%" PRIx64, ptr);
+        return out;
+
+}
+
+static int pandecode_dump_frame_count = 0;
+
+static bool force_stderr = false;
+
+void
+pandecode_dump_file_open(void)
+{
+        if (pandecode_dump_stream)
+                return;
+
+        /* This does a getenv every frame, so it is possible to use
+         * setenv to change the base at runtime.
+         */
+        const char *dump_file_base = debug_get_option("PANDECODE_DUMP_FILE", "pandecode.dump");
+        if (force_stderr || !strcmp(dump_file_base, "stderr"))
+                pandecode_dump_stream = stderr;
+        else {
+                char buffer[1024];
+                snprintf(buffer, sizeof(buffer), "%s.%04d", dump_file_base, pandecode_dump_frame_count);
+                printf("pandecode: dump command stream to file %s\n", buffer);
+                pandecode_dump_stream = fopen(buffer, "w");
+                if (!pandecode_dump_stream)
+                        fprintf(stderr,
+                                "pandecode: failed to open command stream log file %s\n",
+                                buffer);
+        }
+}
+
+static void
+pandecode_dump_file_close(void)
+{
+        if (pandecode_dump_stream && pandecode_dump_stream != stderr) {
+                fclose(pandecode_dump_stream);
+                pandecode_dump_stream = NULL;
+        }
+}
+
+void
+pandecode_initialize(bool to_stderr)
+{
+        force_stderr = to_stderr;
+        mmap_table = _mesa_hash_table_u64_create(NULL);
+        util_dynarray_init(&ro_mappings, NULL);
+}
+
+void
+pandecode_next_frame(void)
+{
+        pandecode_dump_file_close();
+        pandecode_dump_frame_count++;
+}
+
+void
+pandecode_close(void)
+{
+        _mesa_hash_table_u64_destroy(mmap_table, NULL);
+        util_dynarray_fini(&ro_mappings);
+        pandecode_dump_file_close();
+}
index a6b39baf4e655d7994ea99a3c9df58618a078fa3..f25b18fcf75c5f158614d24a1b7fa77ab3d634b3 100644 (file)
@@ -37,6 +37,11 @@ libpanfrost_lib_files = files(
   'pan_props.c',
 )
 
+libpanfrost_decode_files = files(
+  'decode_common.c',
+  'decode.c',
+)
+
 libpanfrost_lib = static_library(
   'panfrost_lib',
   [libpanfrost_lib_files],
@@ -46,3 +51,12 @@ libpanfrost_lib = static_library(
   dependencies: [dep_libdrm, idep_nir],
   build_by_default : false,
 )
+
+libpanfrost_decode = static_library(
+  'panfrost_decode',
+  [libpanfrost_decode_files],
+  include_directories : [inc_include, inc_src, inc_mapi, inc_mesa, inc_gallium, inc_gallium_aux, inc_panfrost],
+  c_args : [no_override_init_args],
+  gnu_symbol_visibility : 'hidden',
+  build_by_default : false,
+)
diff --git a/src/panfrost/lib/wrap.h b/src/panfrost/lib/wrap.h
new file mode 100644 (file)
index 0000000..abb18c8
--- /dev/null
@@ -0,0 +1,56 @@
+
+/*
+ * Copyright (C) 2017-2019 Lyude Paul
+ * Copyright (C) 2017-2019 Alyssa Rosenzweig
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef __PAN_DECODE_PUBLIC_H__
+#define __PAN_DECODE_PUBLIC_H__
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <stdbool.h>
+#include <inttypes.h>
+
+/* Public entrypoints for the tracing infrastructure. This API should be kept
+ * more or less stable. Don't feel bad if you have to change it; just feel
+ * slightly guilty about creating more work for me later. -Alyssa <3
+ *
+ * I'm joking. Mostly. panwrap (out-of-tree) includes this, so update that if
+ * you need to change something here. panwrap is open-source but cannot be
+ * included in-tree.
+ */
+
+void pandecode_initialize(bool to_stderr);
+
+void pandecode_next_frame(void);
+
+void pandecode_close(void);
+
+void
+pandecode_inject_mmap(uint64_t gpu_va, void *cpu, unsigned sz, const char *name);
+
+void pandecode_jc(uint64_t jc_gpu_va, bool bifrost, unsigned gpu_id, bool minimal);
+
+#endif /* __MMAP_TRACE_H__ */
index 8615bcaf7dadf46f57b990390e08fbdbbd409e03..2d257bc552bc02df86d2aaf9633c2574232406e0 100644 (file)
@@ -31,7 +31,6 @@ subdir('shared')
 subdir('util')
 subdir('midgard')
 subdir('bifrost')
-subdir('pandecode')
 subdir('lib')
 
 files_bifrost = files(
diff --git a/src/panfrost/pandecode/common.c b/src/panfrost/pandecode/common.c
deleted file mode 100644 (file)
index 14cb2f0..0000000
+++ /dev/null
@@ -1,200 +0,0 @@
-/*
- * Copyright (C) 2019 Alyssa Rosenzweig
- * Copyright (C) 2017-2018 Lyude Paul
- * Copyright (C) 2019 Collabora, Ltd.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice (including the next
- * paragraph) shall be included in all copies or substantial portions of the
- * Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <assert.h>
-#include <stdint.h>
-#include <string.h>
-#include <sys/mman.h>
-
-#include "decode.h"
-#include "util/macros.h"
-#include "util/u_debug.h"
-#include "util/u_dynarray.h"
-#include "util/hash_table.h"
-
-/* Memory handling */
-
-static struct hash_table_u64 *mmap_table;
-
-static struct util_dynarray ro_mappings;
-
-static struct pandecode_mapped_memory *
-pandecode_find_mapped_gpu_mem_containing_rw(uint64_t addr)
-{
-        return _mesa_hash_table_u64_search(mmap_table, addr & ~(4096 - 1));
-}
-
-struct pandecode_mapped_memory *
-pandecode_find_mapped_gpu_mem_containing(uint64_t addr)
-{
-        struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing_rw(addr);
-
-        if (mem && mem->addr && !mem->ro) {
-                mprotect(mem->addr, mem->length, PROT_READ);
-                mem->ro = true;
-                util_dynarray_append(&ro_mappings, struct pandecode_mapped_memory *, mem);
-        }
-
-        return mem;
-}
-
-void
-pandecode_map_read_write(void)
-{
-        util_dynarray_foreach(&ro_mappings, struct pandecode_mapped_memory *, mem) {
-                (*mem)->ro = false;
-                mprotect((*mem)->addr, (*mem)->length, PROT_READ | PROT_WRITE);
-        }
-        util_dynarray_clear(&ro_mappings);
-}
-
-static void
-pandecode_add_name(struct pandecode_mapped_memory *mem, uint64_t gpu_va, const char *name)
-{
-        if (!name) {
-                /* If we don't have a name, assign one */
-
-                snprintf(mem->name, ARRAY_SIZE(mem->name) - 1,
-                         "memory_%" PRIx64, gpu_va);
-        } else {
-                assert((strlen(name) + 1) < ARRAY_SIZE(mem->name));
-                memcpy(mem->name, name, strlen(name) + 1);
-        }
-}
-
-void
-pandecode_inject_mmap(uint64_t gpu_va, void *cpu, unsigned sz, const char *name)
-{
-        /* First, search if we already mapped this and are just updating an address */
-
-        struct pandecode_mapped_memory *existing =
-                pandecode_find_mapped_gpu_mem_containing_rw(gpu_va);
-
-        if (existing && existing->gpu_va == gpu_va) {
-                existing->length = sz;
-                existing->addr = cpu;
-                pandecode_add_name(existing, gpu_va, name);
-                return;
-        }
-
-        /* Otherwise, add a fresh mapping */
-        struct pandecode_mapped_memory *mapped_mem = NULL;
-
-        mapped_mem = malloc(sizeof(*mapped_mem));
-        mapped_mem->gpu_va = gpu_va;
-        mapped_mem->length = sz;
-        mapped_mem->addr = cpu;
-        pandecode_add_name(mapped_mem, gpu_va, name);
-
-        /* Add it to the table */
-        assert((gpu_va & 4095) == 0);
-
-        for (unsigned i = 0; i < sz; i += 4096)
-                _mesa_hash_table_u64_insert(mmap_table, gpu_va + i, mapped_mem);
-}
-
-char *
-pointer_as_memory_reference(uint64_t ptr)
-{
-        struct pandecode_mapped_memory *mapped;
-        char *out = malloc(128);
-
-        /* Try to find the corresponding mapped zone */
-
-        mapped = pandecode_find_mapped_gpu_mem_containing_rw(ptr);
-
-        if (mapped) {
-                snprintf(out, 128, "%s + %d", mapped->name, (int) (ptr - mapped->gpu_va));
-                return out;
-        }
-
-        /* Just use the raw address if other options are exhausted */
-
-        snprintf(out, 128, "0x%" PRIx64, ptr);
-        return out;
-
-}
-
-static int pandecode_dump_frame_count = 0;
-
-static bool force_stderr = false;
-
-void
-pandecode_dump_file_open(void)
-{
-        if (pandecode_dump_stream)
-                return;
-
-        /* This does a getenv every frame, so it is possible to use
-         * setenv to change the base at runtime.
-         */
-        const char *dump_file_base = debug_get_option("PANDECODE_DUMP_FILE", "pandecode.dump");
-        if (force_stderr || !strcmp(dump_file_base, "stderr"))
-                pandecode_dump_stream = stderr;
-        else {
-                char buffer[1024];
-                snprintf(buffer, sizeof(buffer), "%s.%04d", dump_file_base, pandecode_dump_frame_count);
-                printf("pandecode: dump command stream to file %s\n", buffer);
-                pandecode_dump_stream = fopen(buffer, "w");
-                if (!pandecode_dump_stream)
-                        fprintf(stderr,
-                                "pandecode: failed to open command stream log file %s\n",
-                                buffer);
-        }
-}
-
-static void
-pandecode_dump_file_close(void)
-{
-        if (pandecode_dump_stream && pandecode_dump_stream != stderr) {
-                fclose(pandecode_dump_stream);
-                pandecode_dump_stream = NULL;
-        }
-}
-
-void
-pandecode_initialize(bool to_stderr)
-{
-        force_stderr = to_stderr;
-        mmap_table = _mesa_hash_table_u64_create(NULL);
-        util_dynarray_init(&ro_mappings, NULL);
-}
-
-void
-pandecode_next_frame(void)
-{
-        pandecode_dump_file_close();
-        pandecode_dump_frame_count++;
-}
-
-void
-pandecode_close(void)
-{
-        _mesa_hash_table_u64_destroy(mmap_table, NULL);
-        util_dynarray_fini(&ro_mappings);
-        pandecode_dump_file_close();
-}
diff --git a/src/panfrost/pandecode/decode.c b/src/panfrost/pandecode/decode.c
deleted file mode 100644 (file)
index f3ce25e..0000000
+++ /dev/null
@@ -1,3421 +0,0 @@
-/*
- * Copyright (C) 2017-2019 Alyssa Rosenzweig
- * Copyright (C) 2017-2019 Connor Abbott
- * Copyright (C) 2019 Collabora, Ltd.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice (including the next
- * paragraph) shall be included in all copies or substantial portions of the
- * Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-
-#include <panfrost-job.h>
-#include <stdio.h>
-#include <stdlib.h>
-#include <memory.h>
-#include <stdbool.h>
-#include <stdarg.h>
-#include <ctype.h>
-#include "decode.h"
-#include "util/macros.h"
-#include "util/u_math.h"
-
-#include "midgard/disassemble.h"
-#include "bifrost/disassemble.h"
-
-#include "pan_encoder.h"
-
-static void pandecode_swizzle(unsigned swizzle, enum mali_format format);
-
-#define MEMORY_PROP(obj, p) {\
-        if (obj->p) { \
-                char *a = pointer_as_memory_reference(obj->p); \
-                pandecode_prop("%s = %s", #p, a); \
-                free(a); \
-        } \
-}
-
-#define MEMORY_PROP_DIR(obj, p) {\
-        if (obj.p) { \
-                char *a = pointer_as_memory_reference(obj.p); \
-                pandecode_prop("%s = %s", #p, a); \
-                free(a); \
-        } \
-}
-
-FILE *pandecode_dump_stream;
-
-/* Semantic logging type.
- *
- * Raw: for raw messages to be printed as is.
- * Message: for helpful information to be commented out in replays.
- * Property: for properties of a struct
- *
- * Use one of pandecode_log, pandecode_msg, or pandecode_prop as syntax sugar.
- */
-
-enum pandecode_log_type {
-        PANDECODE_RAW,
-        PANDECODE_MESSAGE,
-        PANDECODE_PROPERTY
-};
-
-#define pandecode_log(...)  pandecode_log_typed(PANDECODE_RAW,      __VA_ARGS__)
-#define pandecode_msg(...)  pandecode_log_typed(PANDECODE_MESSAGE,  __VA_ARGS__)
-#define pandecode_prop(...) pandecode_log_typed(PANDECODE_PROPERTY, __VA_ARGS__)
-
-unsigned pandecode_indent = 0;
-
-static void
-pandecode_make_indent(void)
-{
-        for (unsigned i = 0; i < pandecode_indent; ++i)
-                fprintf(pandecode_dump_stream, "    ");
-}
-
-static void
-pandecode_log_typed(enum pandecode_log_type type, const char *format, ...)
-{
-        va_list ap;
-
-        pandecode_make_indent();
-
-        if (type == PANDECODE_MESSAGE)
-                fprintf(pandecode_dump_stream, "// ");
-        else if (type == PANDECODE_PROPERTY)
-                fprintf(pandecode_dump_stream, ".");
-
-        va_start(ap, format);
-        vfprintf(pandecode_dump_stream, format, ap);
-        va_end(ap);
-
-        if (type == PANDECODE_PROPERTY)
-                fprintf(pandecode_dump_stream, ",\n");
-}
-
-static void
-pandecode_log_cont(const char *format, ...)
-{
-        va_list ap;
-
-        va_start(ap, format);
-        vfprintf(pandecode_dump_stream, format, ap);
-        va_end(ap);
-}
-
-/* To check for memory safety issues, validates that the given pointer in GPU
- * memory is valid, containing at least sz bytes. The goal is to eliminate
- * GPU-side memory bugs (NULL pointer dereferences, buffer overflows, or buffer
- * overruns) by statically validating pointers.
- */
-
-static void
-pandecode_validate_buffer(mali_ptr addr, size_t sz)
-{
-        if (!addr) {
-                pandecode_msg("XXX: null pointer deref");
-                return;
-        }
-
-        /* Find a BO */
-
-        struct pandecode_mapped_memory *bo =
-                pandecode_find_mapped_gpu_mem_containing(addr);
-
-        if (!bo) {
-                pandecode_msg("XXX: invalid memory dereference\n");
-                return;
-        }
-
-        /* Bounds check */
-
-        unsigned offset = addr - bo->gpu_va;
-        unsigned total = offset + sz;
-
-        if (total > bo->length) {
-                pandecode_msg("XXX: buffer overrun. "
-                                "Chunk of size %zu at offset %d in buffer of size %zu. "
-                                "Overrun by %zu bytes. \n",
-                                sz, offset, bo->length, total - bo->length);
-                return;
-        }
-}
-
-struct pandecode_flag_info {
-        u64 flag;
-        const char *name;
-};
-
-static void
-pandecode_log_decoded_flags(const struct pandecode_flag_info *flag_info,
-                            u64 flags)
-{
-        bool decodable_flags_found = false;
-
-        for (int i = 0; flag_info[i].name; i++) {
-                if ((flags & flag_info[i].flag) != flag_info[i].flag)
-                        continue;
-
-                if (!decodable_flags_found) {
-                        decodable_flags_found = true;
-                } else {
-                        pandecode_log_cont(" | ");
-                }
-
-                pandecode_log_cont("%s", flag_info[i].name);
-
-                flags &= ~flag_info[i].flag;
-        }
-
-        if (decodable_flags_found) {
-                if (flags)
-                        pandecode_log_cont(" | 0x%" PRIx64, flags);
-        } else {
-                pandecode_log_cont("0x%" PRIx64, flags);
-        }
-}
-
-#define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
-static const struct pandecode_flag_info gl_enable_flag_info[] = {
-        FLAG_INFO(OCCLUSION_QUERY),
-        FLAG_INFO(OCCLUSION_PRECISE),
-        FLAG_INFO(FRONT_CCW_TOP),
-        FLAG_INFO(CULL_FACE_FRONT),
-        FLAG_INFO(CULL_FACE_BACK),
-        {}
-};
-#undef FLAG_INFO
-
-#define FLAG_INFO(flag) { MALI_CLEAR_##flag, "MALI_CLEAR_" #flag }
-static const struct pandecode_flag_info clear_flag_info[] = {
-        FLAG_INFO(FAST),
-        FLAG_INFO(SLOW),
-        FLAG_INFO(SLOW_STENCIL),
-        {}
-};
-#undef FLAG_INFO
-
-#define FLAG_INFO(flag) { MALI_MASK_##flag, "MALI_MASK_" #flag }
-static const struct pandecode_flag_info mask_flag_info[] = {
-        FLAG_INFO(R),
-        FLAG_INFO(G),
-        FLAG_INFO(B),
-        FLAG_INFO(A),
-        {}
-};
-#undef FLAG_INFO
-
-#define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
-static const struct pandecode_flag_info u3_flag_info[] = {
-        FLAG_INFO(HAS_MSAA),
-        FLAG_INFO(PER_SAMPLE),
-        FLAG_INFO(CAN_DISCARD),
-        FLAG_INFO(HAS_BLEND_SHADER),
-        FLAG_INFO(DEPTH_WRITEMASK),
-        FLAG_INFO(DEPTH_CLIP_NEAR),
-        FLAG_INFO(DEPTH_CLIP_FAR),
-        {}
-};
-
-static const struct pandecode_flag_info u4_flag_info[] = {
-        FLAG_INFO(NO_MSAA),
-        FLAG_INFO(NO_DITHER),
-        FLAG_INFO(DEPTH_RANGE_A),
-        FLAG_INFO(DEPTH_RANGE_B),
-        FLAG_INFO(STENCIL_TEST),
-        FLAG_INFO(ALPHA_TO_COVERAGE),
-        {}
-};
-#undef FLAG_INFO
-
-#define FLAG_INFO(flag) { MALI_MFBD_FORMAT_##flag, "MALI_MFBD_FORMAT_" #flag }
-static const struct pandecode_flag_info mfbd_fmt_flag_info[] = {
-        FLAG_INFO(SRGB),
-        {}
-};
-#undef FLAG_INFO
-
-#define FLAG_INFO(flag) { MALI_AFBC_##flag, "MALI_AFBC_" #flag }
-static const struct pandecode_flag_info afbc_fmt_flag_info[] = {
-        FLAG_INFO(YTR),
-        {}
-};
-#undef FLAG_INFO
-
-#define FLAG_INFO(flag) { MALI_EXTRA_##flag, "MALI_EXTRA_" #flag }
-static const struct pandecode_flag_info mfbd_extra_flag_hi_info[] = {
-        FLAG_INFO(PRESENT),
-        {}
-};
-#undef FLAG_INFO
-
-#define FLAG_INFO(flag) { MALI_EXTRA_##flag, "MALI_EXTRA_" #flag }
-static const struct pandecode_flag_info mfbd_extra_flag_lo_info[] = {
-        FLAG_INFO(ZS),
-        {}
-};
-#undef FLAG_INFO
-
-#define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
-static const struct pandecode_flag_info shader_midgard1_flag_lo_info [] = {
-        FLAG_INFO(WRITES_Z),
-        FLAG_INFO(EARLY_Z),
-        FLAG_INFO(READS_TILEBUFFER),
-        FLAG_INFO(WRITES_GLOBAL),
-        FLAG_INFO(READS_ZS),
-        {}
-};
-
-static const struct pandecode_flag_info shader_midgard1_flag_hi_info [] = {
-        FLAG_INFO(WRITES_S),
-        FLAG_INFO(SUPPRESS_INF_NAN),
-        {}
-};
-#undef FLAG_INFO
-
-#define FLAG_INFO(flag) { MALI_BIFROST_##flag, "MALI_BIFROST_" #flag }
-static const struct pandecode_flag_info shader_bifrost_info [] = {
-        FLAG_INFO(FULL_THREAD),
-        FLAG_INFO(EARLY_Z),
-        FLAG_INFO(FIRST_ATEST),
-        {}
-};
-
-#undef FLAG_INFO
-
-#define FLAG_INFO(flag) { MALI_MFBD_##flag, "MALI_MFBD_" #flag }
-static const struct pandecode_flag_info mfbd_flag_info [] = {
-        FLAG_INFO(DEPTH_WRITE),
-        FLAG_INFO(EXTRA),
-        {}
-};
-#undef FLAG_INFO
-
-#define FLAG_INFO(flag) { MALI_SAMP_##flag, "MALI_SAMP_" #flag }
-static const struct pandecode_flag_info sampler_flag_info [] = {
-        FLAG_INFO(MAG_NEAREST),
-        FLAG_INFO(MIN_NEAREST),
-        FLAG_INFO(MIP_LINEAR_1),
-        FLAG_INFO(MIP_LINEAR_2),
-        FLAG_INFO(NORM_COORDS),
-        {}
-};
-#undef FLAG_INFO
-
-#define FLAG_INFO(flag) { MALI_SFBD_FORMAT_##flag, "MALI_SFBD_FORMAT_" #flag }
-static const struct pandecode_flag_info sfbd_unk1_info [] = {
-        FLAG_INFO(MSAA_8),
-        FLAG_INFO(MSAA_A),
-        {}
-};
-#undef FLAG_INFO
-
-#define FLAG_INFO(flag) { MALI_SFBD_FORMAT_##flag, "MALI_SFBD_FORMAT_" #flag }
-static const struct pandecode_flag_info sfbd_unk2_info [] = {
-        FLAG_INFO(MSAA_B),
-        FLAG_INFO(SRGB),
-        {}
-};
-#undef FLAG_INFO
-
-#define DEFINE_CASE(name) case MALI_## name: return "MALI_" #name
-static char *pandecode_format(enum mali_format format)
-{
-        static char unk_format_str[10];
-
-        switch (format) {
-                DEFINE_CASE(ETC2_RGB8);
-                DEFINE_CASE(ETC2_R11_UNORM);
-                DEFINE_CASE(ETC2_RGBA8);
-                DEFINE_CASE(ETC2_RG11_UNORM);
-                DEFINE_CASE(ETC2_R11_SNORM);
-                DEFINE_CASE(ETC2_RG11_SNORM);
-                DEFINE_CASE(ETC2_RGB8A1);
-                DEFINE_CASE(NXR);
-                DEFINE_CASE(BC1_UNORM);
-                DEFINE_CASE(BC2_UNORM);
-                DEFINE_CASE(BC3_UNORM);
-                DEFINE_CASE(BC4_UNORM);
-                DEFINE_CASE(BC4_SNORM);
-                DEFINE_CASE(BC5_UNORM);
-                DEFINE_CASE(BC5_SNORM);
-                DEFINE_CASE(BC6H_UF16);
-                DEFINE_CASE(BC6H_SF16);
-                DEFINE_CASE(BC7_UNORM);
-                DEFINE_CASE(ASTC_3D_LDR);
-                DEFINE_CASE(ASTC_3D_HDR);
-                DEFINE_CASE(ASTC_2D_LDR);
-                DEFINE_CASE(ASTC_2D_HDR);
-                DEFINE_CASE(RGB565);
-                DEFINE_CASE(RGB5_X1_UNORM);
-                DEFINE_CASE(RGB5_A1_UNORM);
-                DEFINE_CASE(RGB10_A2_UNORM);
-                DEFINE_CASE(RGB10_A2_SNORM);
-                DEFINE_CASE(RGB10_A2UI);
-                DEFINE_CASE(RGB10_A2I);
-                DEFINE_CASE(RGB332_UNORM);
-                DEFINE_CASE(RGB233_UNORM);
-                DEFINE_CASE(Z24X8_UNORM);
-                DEFINE_CASE(R32_FIXED);
-                DEFINE_CASE(RG32_FIXED);
-                DEFINE_CASE(RGB32_FIXED);
-                DEFINE_CASE(RGBA32_FIXED);
-                DEFINE_CASE(R11F_G11F_B10F);
-                DEFINE_CASE(R9F_G9F_B9F_E5F);
-                DEFINE_CASE(VARYING_POS);
-                DEFINE_CASE(VARYING_DISCARD);
-
-                DEFINE_CASE(R8_SNORM);
-                DEFINE_CASE(R16_SNORM);
-                DEFINE_CASE(R32_SNORM);
-                DEFINE_CASE(RG8_SNORM);
-                DEFINE_CASE(RG16_SNORM);
-                DEFINE_CASE(RG32_SNORM);
-                DEFINE_CASE(RGB8_SNORM);
-                DEFINE_CASE(RGB16_SNORM);
-                DEFINE_CASE(RGB32_SNORM);
-                DEFINE_CASE(RGBA8_SNORM);
-                DEFINE_CASE(RGBA16_SNORM);
-                DEFINE_CASE(RGBA32_SNORM);
-
-                DEFINE_CASE(R8UI);
-                DEFINE_CASE(R16UI);
-                DEFINE_CASE(R32UI);
-                DEFINE_CASE(RG8UI);
-                DEFINE_CASE(RG16UI);
-                DEFINE_CASE(RG32UI);
-                DEFINE_CASE(RGB8UI);
-                DEFINE_CASE(RGB16UI);
-                DEFINE_CASE(RGB32UI);
-                DEFINE_CASE(RGBA8UI);
-                DEFINE_CASE(RGBA16UI);
-                DEFINE_CASE(RGBA32UI);
-
-                DEFINE_CASE(R8_UNORM);
-                DEFINE_CASE(R16_UNORM);
-                DEFINE_CASE(R32_UNORM);
-                DEFINE_CASE(R32F);
-                DEFINE_CASE(RG8_UNORM);
-                DEFINE_CASE(RG16_UNORM);
-                DEFINE_CASE(RG32_UNORM);
-                DEFINE_CASE(RG32F);
-                DEFINE_CASE(RGB8_UNORM);
-                DEFINE_CASE(RGB16_UNORM);
-                DEFINE_CASE(RGB32_UNORM);
-                DEFINE_CASE(RGB32F);
-                DEFINE_CASE(RGBA4_UNORM);
-                DEFINE_CASE(RGBA8_UNORM);
-                DEFINE_CASE(RGBA16_UNORM);
-                DEFINE_CASE(RGBA32_UNORM);
-                DEFINE_CASE(RGBA32F);
-
-                DEFINE_CASE(R8I);
-                DEFINE_CASE(R16I);
-                DEFINE_CASE(R32I);
-                DEFINE_CASE(RG8I);
-                DEFINE_CASE(R16F);
-                DEFINE_CASE(RG16I);
-                DEFINE_CASE(RG32I);
-                DEFINE_CASE(RG16F);
-                DEFINE_CASE(RGB8I);
-                DEFINE_CASE(RGB16I);
-                DEFINE_CASE(RGB32I);
-                DEFINE_CASE(RGB16F);
-                DEFINE_CASE(RGBA8I);
-                DEFINE_CASE(RGBA16I);
-                DEFINE_CASE(RGBA32I);
-                DEFINE_CASE(RGBA16F);
-
-                DEFINE_CASE(RGBA4);
-                DEFINE_CASE(RGBA8_2);
-                DEFINE_CASE(RGB10_A2_2);
-        default:
-                snprintf(unk_format_str, sizeof(unk_format_str), "MALI_0x%02x", format);
-                return unk_format_str;
-        }
-}
-
-#undef DEFINE_CASE
-
-static char *
-pandecode_job_type(enum mali_job_type type)
-{
-#define DEFINE_CASE(name) case JOB_TYPE_ ## name: return "JOB_TYPE_" #name
-
-        switch (type) {
-                DEFINE_CASE(NULL);
-                DEFINE_CASE(WRITE_VALUE);
-                DEFINE_CASE(CACHE_FLUSH);
-                DEFINE_CASE(COMPUTE);
-                DEFINE_CASE(VERTEX);
-                DEFINE_CASE(TILER);
-                DEFINE_CASE(FUSED);
-                DEFINE_CASE(FRAGMENT);
-
-        case JOB_NOT_STARTED:
-                return "NOT_STARTED";
-
-        default:
-                pandecode_log("Warning! Unknown job type %x\n", type);
-                return "!?!?!?";
-        }
-
-#undef DEFINE_CASE
-}
-
-static char *
-pandecode_draw_mode(enum mali_draw_mode mode)
-{
-#define DEFINE_CASE(name) case MALI_ ## name: return "MALI_" #name
-
-        switch (mode) {
-                DEFINE_CASE(DRAW_NONE);
-                DEFINE_CASE(POINTS);
-                DEFINE_CASE(LINES);
-                DEFINE_CASE(TRIANGLES);
-                DEFINE_CASE(TRIANGLE_STRIP);
-                DEFINE_CASE(TRIANGLE_FAN);
-                DEFINE_CASE(LINE_STRIP);
-                DEFINE_CASE(LINE_LOOP);
-                DEFINE_CASE(POLYGON);
-                DEFINE_CASE(QUADS);
-                DEFINE_CASE(QUAD_STRIP);
-
-        default:
-                pandecode_msg("XXX: invalid draw mode %X\n", mode);
-                return "";
-        }
-
-#undef DEFINE_CASE
-}
-
-#define DEFINE_CASE(name) case MALI_FUNC_ ## name: return "MALI_FUNC_" #name
-static char *
-pandecode_func(enum mali_func mode)
-{
-        switch (mode) {
-                DEFINE_CASE(NEVER);
-                DEFINE_CASE(LESS);
-                DEFINE_CASE(EQUAL);
-                DEFINE_CASE(LEQUAL);
-                DEFINE_CASE(GREATER);
-                DEFINE_CASE(NOTEQUAL);
-                DEFINE_CASE(GEQUAL);
-                DEFINE_CASE(ALWAYS);
-
-        default:
-                pandecode_msg("XXX: invalid func %X\n", mode);
-                return "";
-        }
-}
-#undef DEFINE_CASE
-
-#define DEFINE_CASE(name) case MALI_MSAA_ ## name: return "MALI_MSAA_" #name
-static char *
-pandecode_msaa_mode(enum mali_msaa_mode mode)
-{
-        switch (mode) {
-                DEFINE_CASE(SINGLE);
-                DEFINE_CASE(AVERAGE);
-                DEFINE_CASE(MULTIPLE);
-                DEFINE_CASE(LAYERED);
-        default:
-                unreachable("Impossible");
-                return "";
-        }
-}
-#undef DEFINE_CASE
-
-#define DEFINE_CASE(name) case MALI_STENCIL_ ## name: return "MALI_STENCIL_" #name
-static char *
-pandecode_stencil_op(enum mali_stencil_op op)
-{
-        switch (op) {
-                DEFINE_CASE(KEEP);
-                DEFINE_CASE(REPLACE);
-                DEFINE_CASE(ZERO);
-                DEFINE_CASE(INVERT);
-                DEFINE_CASE(INCR_WRAP);
-                DEFINE_CASE(DECR_WRAP);
-                DEFINE_CASE(INCR);
-                DEFINE_CASE(DECR);
-
-        default:
-                pandecode_msg("XXX: invalid stencil op %X\n", op);
-                return "";
-        }
-}
-
-#undef DEFINE_CASE
-
-static char *pandecode_attr_mode_short(enum mali_attr_mode mode)
-{
-        switch(mode) {
-                /* TODO: Combine to just "instanced" once this can be done
-                 * unambiguously in all known cases */
-        case MALI_ATTR_POT_DIVIDE:
-                return "instanced_pot";
-        case MALI_ATTR_MODULO:
-                return "instanced_mod";
-        case MALI_ATTR_NPOT_DIVIDE:
-                return "instanced_npot";
-        case MALI_ATTR_IMAGE:
-                return "image";
-        default:
-                pandecode_msg("XXX: invalid attribute mode %X\n", mode);
-                return "";
-        }
-}
-
-static const char *
-pandecode_special_record(uint64_t v, bool* attribute)
-{
-        switch(v) {
-        case MALI_ATTR_VERTEXID:
-                *attribute = true;
-                return "gl_VertexID";
-        case MALI_ATTR_INSTANCEID:
-                *attribute = true;
-                return "gl_InstanceID";
-        case MALI_VARYING_FRAG_COORD:
-                return "gl_FragCoord";
-        case MALI_VARYING_FRONT_FACING:
-                return "gl_FrontFacing";
-        case MALI_VARYING_POINT_COORD:
-                return "gl_PointCoord";
-        default:
-                pandecode_msg("XXX: invalid special record %" PRIx64 "\n", v);
-                return "";
-        }
-}
-
-#define DEFINE_CASE(name) case MALI_WRAP_## name: return "MALI_WRAP_" #name
-static char *
-pandecode_wrap_mode(enum mali_wrap_mode op)
-{
-        switch (op) {
-                DEFINE_CASE(REPEAT);
-                DEFINE_CASE(CLAMP_TO_EDGE);
-                DEFINE_CASE(CLAMP);
-                DEFINE_CASE(CLAMP_TO_BORDER);
-                DEFINE_CASE(MIRRORED_REPEAT);
-                DEFINE_CASE(MIRRORED_CLAMP_TO_EDGE);
-                DEFINE_CASE(MIRRORED_CLAMP);
-                DEFINE_CASE(MIRRORED_CLAMP_TO_BORDER);
-
-        default:
-                pandecode_msg("XXX: invalid wrap mode %X\n", op);
-                return "";
-        }
-}
-#undef DEFINE_CASE
-
-#define DEFINE_CASE(name) case MALI_BLOCK_## name: return "MALI_BLOCK_" #name
-static char *
-pandecode_block_format(enum mali_block_format fmt)
-{
-        switch (fmt) {
-                DEFINE_CASE(TILED);
-                DEFINE_CASE(UNKNOWN);
-                DEFINE_CASE(LINEAR);
-                DEFINE_CASE(AFBC);
-
-        default:
-                unreachable("Invalid case");
-        }
-}
-#undef DEFINE_CASE
-
-#define DEFINE_CASE(name) case MALI_EXCEPTION_ACCESS_## name: return ""#name
-static char *
-pandecode_exception_access(unsigned access)
-{
-        switch (access) {
-                DEFINE_CASE(NONE);
-                DEFINE_CASE(EXECUTE);
-                DEFINE_CASE(READ);
-                DEFINE_CASE(WRITE);
-
-        default:
-                unreachable("Invalid case");
-        }
-}
-#undef DEFINE_CASE
-
-/* Midgard's tiler descriptor is embedded within the
- * larger FBD */
-
-static void
-pandecode_midgard_tiler_descriptor(
-                const struct midgard_tiler_descriptor *t,
-                unsigned width,
-                unsigned height,
-                bool is_fragment,
-                bool has_hierarchy)
-{
-        pandecode_log(".tiler = {\n");
-        pandecode_indent++;
-
-        if (t->hierarchy_mask == MALI_TILER_DISABLED)
-                pandecode_prop("hierarchy_mask = MALI_TILER_DISABLED");
-        else
-                pandecode_prop("hierarchy_mask = 0x%" PRIx16, t->hierarchy_mask);
-
-        /* We know this name from the kernel, but we never see it nonzero */
-
-        if (t->flags)
-                pandecode_msg("XXX: unexpected tiler flags 0x%" PRIx16, t->flags);
-
-        MEMORY_PROP(t, polygon_list);
-
-        /* The body is offset from the base of the polygon list */
-        //assert(t->polygon_list_body > t->polygon_list);
-        unsigned body_offset = t->polygon_list_body - t->polygon_list;
-
-        /* It needs to fit inside the reported size */
-        //assert(t->polygon_list_size >= body_offset);
-
-        /* Now that we've sanity checked, we'll try to calculate the sizes
-         * ourselves for comparison */
-
-        unsigned ref_header = panfrost_tiler_header_size(width, height, t->hierarchy_mask, has_hierarchy);
-        unsigned ref_size = panfrost_tiler_full_size(width, height, t->hierarchy_mask, has_hierarchy);
-
-        if (!((ref_header == body_offset) && (ref_size == t->polygon_list_size))) {
-                pandecode_msg("XXX: bad polygon list size (expected %d / 0x%x)\n",
-                                ref_header, ref_size);
-                pandecode_prop("polygon_list_size = 0x%x", t->polygon_list_size);
-                pandecode_msg("body offset %d\n", body_offset);
-        }
-
-        /* The tiler heap has a start and end specified -- it should be
-         * identical to what we have in the BO. The exception is if tiling is
-         * disabled. */
-
-        MEMORY_PROP(t, heap_start);
-        assert(t->heap_end >= t->heap_start);
-
-        struct pandecode_mapped_memory *heap =
-                pandecode_find_mapped_gpu_mem_containing(t->heap_start);
-
-        unsigned heap_size = t->heap_end - t->heap_start;
-
-        /* Tiling is enabled with a special flag */
-        unsigned hierarchy_mask = t->hierarchy_mask & MALI_HIERARCHY_MASK;
-        unsigned tiler_flags = t->hierarchy_mask ^ hierarchy_mask;
-
-        bool tiling_enabled = hierarchy_mask;
-
-        if (tiling_enabled) {
-                /* When tiling is enabled, the heap should be a tight fit */
-                unsigned heap_offset = t->heap_start - heap->gpu_va;
-                if ((heap_offset + heap_size) != heap->length) {
-                        pandecode_msg("XXX: heap size %u (expected %zu)\n",
-                                        heap_size, heap->length - heap_offset);
-                }
-
-                /* We should also have no other flags */
-                if (tiler_flags)
-                        pandecode_msg("XXX: unexpected tiler %X\n", tiler_flags);
-        } else {
-                /* When tiling is disabled, we should have that flag and no others */
-
-                if (tiler_flags != MALI_TILER_DISABLED) {
-                        pandecode_msg("XXX: unexpected tiler flag %X, expected MALI_TILER_DISABLED\n",
-                                        tiler_flags);
-                }
-
-                /* We should also have an empty heap */
-                if (heap_size) {
-                        pandecode_msg("XXX: tiler heap size %d given, expected empty\n",
-                                        heap_size);
-                }
-
-                /* Disabled tiling is used only for clear-only jobs, which are
-                 * purely FRAGMENT, so we should never see this for
-                 * non-FRAGMENT descriptors. */
-
-                if (!is_fragment)
-                        pandecode_msg("XXX: tiler disabled for non-FRAGMENT job\n");
-        }
-
-        /* We've never seen weights used in practice, but we know from the
-         * kernel these fields is there */
-
-        bool nonzero_weights = false;
-
-        for (unsigned w = 0; w < ARRAY_SIZE(t->weights); ++w) {
-                nonzero_weights |= t->weights[w] != 0x0;
-        }
-
-        if (nonzero_weights) {
-                pandecode_log(".weights = { ");
-
-                for (unsigned w = 0; w < ARRAY_SIZE(t->weights); ++w) {
-                        pandecode_log_cont("%d, ", t->weights[w]);
-                }
-
-                pandecode_log("},");
-        }
-
-        pandecode_indent--;
-        pandecode_log("}\n");
-}
-
-/* TODO: The Bifrost tiler is not understood at all yet */
-
-static void
-pandecode_bifrost_tiler_descriptor(const struct mali_framebuffer *fb)
-{
-        pandecode_log(".tiler = {\n");
-        pandecode_indent++;
-
-        MEMORY_PROP(fb, tiler_meta);
-
-        for (int i = 0; i < 16; i++) {
-                if (fb->zeros[i] != 0) {
-                        pandecode_msg("XXX: tiler descriptor zero %d tripped, value %x\n",
-                                      i, fb->zeros[i]);
-                }
-        }
-
-        pandecode_log("},\n");
-
-        pandecode_indent--;
-        pandecode_log("}\n");
-
-}
-
-/* Information about the framebuffer passed back for
- * additional analysis */
-
-struct pandecode_fbd {
-        unsigned width;
-        unsigned height;
-        unsigned rt_count;
-        bool has_extra;
-};
-
-static void
-pandecode_sfbd_format(struct mali_sfbd_format format)
-{
-        pandecode_log(".format = {\n");
-        pandecode_indent++;
-
-        pandecode_log(".unk1 = ");
-        pandecode_log_decoded_flags(sfbd_unk1_info, format.unk1);
-        pandecode_log_cont(",\n");
-
-        /* TODO: Map formats so we can check swizzles and print nicely */
-        pandecode_log("swizzle");
-        pandecode_swizzle(format.swizzle, MALI_RGBA8_UNORM);
-        pandecode_log_cont(",\n");
-
-        pandecode_prop("nr_channels = MALI_POSITIVE(%d)",
-                       (format.nr_channels + 1));
-
-        pandecode_log(".unk2 = ");
-        pandecode_log_decoded_flags(sfbd_unk2_info, format.unk2);
-        pandecode_log_cont(",\n");
-
-        pandecode_prop("block = %s", pandecode_block_format(format.block));
-
-        pandecode_prop("unk3 = 0x%" PRIx32, format.unk3);
-
-        pandecode_indent--;
-        pandecode_log("},\n");
-}
-
-static void
-pandecode_shared_memory(const struct mali_shared_memory *desc, bool is_compute)
-{
-        pandecode_prop("stack_shift = 0x%x", desc->stack_shift);
-
-        if (desc->unk0)
-                pandecode_prop("unk0 = 0x%x", desc->unk0);
-
-        if (desc->shared_workgroup_count != 0x1F) {
-                pandecode_prop("shared_workgroup_count = %d", desc->shared_workgroup_count);
-                if (!is_compute)
-                        pandecode_msg("XXX: wrong workgroup count for noncompute\n");
-        }
-
-        if (desc->shared_unk1 || desc->shared_shift) {
-                pandecode_prop("shared_unk1 = %X", desc->shared_unk1);
-                pandecode_prop("shared_shift = %X", desc->shared_shift);
-
-                if (!is_compute)
-                        pandecode_msg("XXX: shared memory configured in noncompute shader");
-        }
-
-        if (desc->shared_zero) {
-                pandecode_msg("XXX: shared memory zero tripped\n");
-                pandecode_prop("shared_zero = 0x%" PRIx32, desc->shared_zero);
-        }
-
-        if (desc->shared_memory && !is_compute)
-                pandecode_msg("XXX: shared memory used in noncompute shader\n");
-
-        MEMORY_PROP(desc, scratchpad);
-        MEMORY_PROP(desc, shared_memory);
-        MEMORY_PROP(desc, unknown1);
-
-        if (desc->scratchpad) {
-                struct pandecode_mapped_memory *smem =
-                        pandecode_find_mapped_gpu_mem_containing(desc->scratchpad);
-
-                pandecode_msg("scratchpad size %u\n", smem->length);
-        }
-
-}
-
-static struct pandecode_fbd
-pandecode_sfbd(uint64_t gpu_va, int job_no, bool is_fragment, unsigned gpu_id)
-{
-        struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
-        const struct mali_single_framebuffer *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va);
-
-        struct pandecode_fbd info = {
-                .has_extra = false,
-                .rt_count = 1
-        };
-
-        pandecode_log("struct mali_single_framebuffer framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
-        pandecode_indent++;
-
-        pandecode_log(".shared_memory = {\n");
-        pandecode_indent++;
-        pandecode_shared_memory(&s->shared_memory, false);
-        pandecode_indent--;
-        pandecode_log("},\n");
-
-        pandecode_sfbd_format(s->format);
-
-        info.width = s->width + 1;
-        info.height = s->height + 1;
-
-        pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", info.width);
-        pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", info.height);
-
-        MEMORY_PROP(s, checksum);
-
-        if (s->checksum_stride)
-                pandecode_prop("checksum_stride = %d", s->checksum_stride);
-
-        MEMORY_PROP(s, framebuffer);
-        pandecode_prop("stride = %d", s->stride);
-
-        /* Earlier in the actual commandstream -- right before width -- but we
-         * delay to flow nicer */
-
-        pandecode_log(".clear_flags = ");
-        pandecode_log_decoded_flags(clear_flag_info, s->clear_flags);
-        pandecode_log_cont(",\n");
-
-        if (s->depth_buffer) {
-                MEMORY_PROP(s, depth_buffer);
-                pandecode_prop("depth_stride = %d", s->depth_stride);
-        }
-
-        if (s->stencil_buffer) {
-                MEMORY_PROP(s, stencil_buffer);
-                pandecode_prop("stencil_stride = %d", s->stencil_stride);
-        }
-
-        if (s->depth_stride_zero ||
-            s->stencil_stride_zero ||
-            s->zero7 || s->zero8) {
-                pandecode_msg("XXX: Depth/stencil zeros tripped\n");
-                pandecode_prop("depth_stride_zero = 0x%x",
-                               s->depth_stride_zero);
-                pandecode_prop("stencil_stride_zero = 0x%x",
-                               s->stencil_stride_zero);
-                pandecode_prop("zero7 = 0x%" PRIx32,
-                               s->zero7);
-                pandecode_prop("zero8 = 0x%" PRIx32,
-                               s->zero8);
-        }
-
-        if (s->clear_color_1 | s->clear_color_2 | s->clear_color_3 | s->clear_color_4) {
-                pandecode_prop("clear_color_1 = 0x%" PRIx32, s->clear_color_1);
-                pandecode_prop("clear_color_2 = 0x%" PRIx32, s->clear_color_2);
-                pandecode_prop("clear_color_3 = 0x%" PRIx32, s->clear_color_3);
-                pandecode_prop("clear_color_4 = 0x%" PRIx32, s->clear_color_4);
-        }
-
-        if (s->clear_depth_1 != 0 || s->clear_depth_2 != 0 || s->clear_depth_3 != 0 || s->clear_depth_4 != 0) {
-                pandecode_prop("clear_depth_1 = %f", s->clear_depth_1);
-                pandecode_prop("clear_depth_2 = %f", s->clear_depth_2);
-                pandecode_prop("clear_depth_3 = %f", s->clear_depth_3);
-                pandecode_prop("clear_depth_4 = %f", s->clear_depth_4);
-        }
-
-        if (s->clear_stencil) {
-                pandecode_prop("clear_stencil = 0x%x", s->clear_stencil);
-        }
-
-        const struct midgard_tiler_descriptor t = s->tiler;
-
-        bool has_hierarchy = !(gpu_id == 0x0720 || gpu_id == 0x0820 || gpu_id == 0x0830);
-        pandecode_midgard_tiler_descriptor(&t, s->width + 1, s->height + 1, is_fragment, has_hierarchy);
-
-        pandecode_indent--;
-        pandecode_log("};\n");
-
-        pandecode_prop("zero2 = 0x%" PRIx32, s->zero2);
-        pandecode_prop("zero4 = 0x%" PRIx32, s->zero4);
-        pandecode_prop("zero5 = 0x%" PRIx32, s->zero5);
-
-        pandecode_log_cont(".zero3 = {");
-
-        for (int i = 0; i < sizeof(s->zero3) / sizeof(s->zero3[0]); ++i)
-                pandecode_log_cont("%X, ", s->zero3[i]);
-
-        pandecode_log_cont("},\n");
-
-        pandecode_log_cont(".zero6 = {");
-
-        for (int i = 0; i < sizeof(s->zero6) / sizeof(s->zero6[0]); ++i)
-                pandecode_log_cont("%X, ", s->zero6[i]);
-
-        pandecode_log_cont("},\n");
-
-        return info;
-}
-
-static void
-pandecode_compute_fbd(uint64_t gpu_va, int job_no)
-{
-        struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
-        const struct mali_shared_memory *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va);
-
-        pandecode_log("struct mali_shared_memory shared_%"PRIx64"_%d = {\n", gpu_va, job_no);
-        pandecode_indent++;
-        pandecode_shared_memory(s, true);
-        pandecode_indent--;
-        pandecode_log("},\n");
-}
-
-/* Extracts the number of components associated with a Mali format */
-
-static unsigned
-pandecode_format_component_count(enum mali_format fmt)
-{
-        /* Mask out the format class */
-        unsigned top = fmt & 0b11100000;
-
-        switch (top) {
-        case MALI_FORMAT_SNORM:
-        case MALI_FORMAT_UINT:
-        case MALI_FORMAT_UNORM:
-        case MALI_FORMAT_SINT:
-                return ((fmt >> 3) & 3) + 1;
-        default:
-                /* TODO: Validate */
-                return 4;
-        }
-}
-
-/* Extracts a mask of accessed components from a 12-bit Mali swizzle */
-
-static unsigned
-pandecode_access_mask_from_channel_swizzle(unsigned swizzle)
-{
-        unsigned mask = 0;
-        assert(MALI_CHANNEL_RED == 0);
-
-        for (unsigned c = 0; c < 4; ++c) {
-                enum mali_channel chan = (swizzle >> (3*c)) & 0x7;
-
-                if (chan <= MALI_CHANNEL_ALPHA)
-                        mask |= (1 << chan);
-        }
-
-        return mask;
-}
-
-/* Validates that a (format, swizzle) pair is valid, in the sense that the
- * swizzle doesn't access any components that are undefined in the format.
- * Returns whether the swizzle is trivial (doesn't do any swizzling) and can be
- * omitted */
-
-static bool
-pandecode_validate_format_swizzle(enum mali_format fmt, unsigned swizzle)
-{
-        unsigned nr_comp = pandecode_format_component_count(fmt);
-        unsigned access_mask = pandecode_access_mask_from_channel_swizzle(swizzle);
-        unsigned valid_mask = (1 << nr_comp) - 1;
-        unsigned invalid_mask = ~valid_mask;
-
-        if (access_mask & invalid_mask) {
-                pandecode_msg("XXX: invalid components accessed\n");
-                return false;
-        }
-
-        /* Check for the default non-swizzling swizzle so we can suppress
-         * useless printing for the defaults */
-
-        unsigned default_swizzles[4] = {
-                MALI_CHANNEL_RED | (MALI_CHANNEL_ZERO  << 3) | (MALI_CHANNEL_ZERO << 6) | (MALI_CHANNEL_ONE   << 9),
-                MALI_CHANNEL_RED | (MALI_CHANNEL_GREEN << 3) | (MALI_CHANNEL_ZERO << 6) | (MALI_CHANNEL_ONE   << 9),
-                MALI_CHANNEL_RED | (MALI_CHANNEL_GREEN << 3) | (MALI_CHANNEL_BLUE << 6) | (MALI_CHANNEL_ONE   << 9),
-                MALI_CHANNEL_RED | (MALI_CHANNEL_GREEN << 3) | (MALI_CHANNEL_BLUE << 6) | (MALI_CHANNEL_ALPHA << 9)
-        };
-
-        return (swizzle == default_swizzles[nr_comp - 1]);
-}
-
-/* Maps MALI_RGBA32F to rgba32f, etc */
-
-static void
-pandecode_format_short(enum mali_format fmt, bool srgb)
-{
-        /* We want a type-like format, so cut off the initial MALI_ */
-        char *format = pandecode_format(fmt);
-        format += strlen("MALI_");
-
-        unsigned len = strlen(format);
-        char *lower_format = calloc(1, len + 1);
-
-        for (unsigned i = 0; i < len; ++i)
-                lower_format[i] = tolower(format[i]);
-
-        /* Sanity check sRGB flag is applied to RGB, per the name */
-        if (srgb && lower_format[0] != 'r')
-                pandecode_msg("XXX: sRGB applied to non-colour format\n");
-
-        /* Just prefix with an s, so you get formats like srgba8_unorm */
-        if (srgb)
-                pandecode_log_cont("s");
-
-        pandecode_log_cont("%s", lower_format);
-        free(lower_format);
-}
-
-static void
-pandecode_swizzle(unsigned swizzle, enum mali_format format)
-{
-        /* First, do some validation */
-        bool trivial_swizzle = pandecode_validate_format_swizzle(
-                        format, swizzle);
-
-        if (trivial_swizzle)
-                return;
-
-        /* Next, print the swizzle */
-        pandecode_log_cont(".");
-
-        static const char components[] = "rgba01";
-
-        for (unsigned c = 0; c < 4; ++c) {
-                enum mali_channel chan = (swizzle >> (3 * c)) & 0x7;
-
-                if (chan >= MALI_CHANNEL_RESERVED_0) {
-                        pandecode_log("XXX: invalid swizzle channel %d\n", chan);
-                        continue;
-                }
-                pandecode_log_cont("%c", components[chan]);
-        }
-}
-
-static void
-pandecode_rt_format(struct mali_rt_format format)
-{
-        pandecode_log(".format = {\n");
-        pandecode_indent++;
-
-        pandecode_prop("unk1 = 0x%" PRIx32, format.unk1);
-        pandecode_prop("unk2 = 0x%" PRIx32, format.unk2);
-        pandecode_prop("unk3 = 0x%" PRIx32, format.unk3);
-        pandecode_prop("unk4 = 0x%" PRIx32, format.unk4);
-
-        pandecode_prop("block = %s", pandecode_block_format(format.block));
-
-        /* TODO: Map formats so we can check swizzles and print nicely */
-        pandecode_log("swizzle");
-        pandecode_swizzle(format.swizzle, MALI_RGBA8_UNORM);
-        pandecode_log_cont(",\n");
-
-        pandecode_prop("nr_channels = MALI_POSITIVE(%d)",
-                       (format.nr_channels + 1));
-
-        pandecode_log(".flags = ");
-        pandecode_log_decoded_flags(mfbd_fmt_flag_info, format.flags);
-        pandecode_log_cont(",\n");
-
-        pandecode_prop("msaa = %s", pandecode_msaa_mode(format.msaa));
-
-        /* In theory, the no_preload bit can be cleared to enable MFBD preload,
-         * which is a faster hardware-based alternative to the wallpaper method
-         * to preserve framebuffer contents across frames. In practice, MFBD
-         * preload is buggy on Midgard, and so this is a chicken bit. If this
-         * bit isn't set, most likely something broke unrelated to preload */
-
-        if (!format.no_preload) {
-                pandecode_msg("XXX: buggy MFBD preload enabled - chicken bit should be clear\n");
-                pandecode_prop("no_preload = 0x%" PRIx32, format.no_preload);
-        }
-
-        if (format.zero)
-                pandecode_prop("zero = 0x%" PRIx32, format.zero);
-
-        pandecode_indent--;
-        pandecode_log("},\n");
-}
-
-static void
-pandecode_render_target(uint64_t gpu_va, unsigned job_no, const struct mali_framebuffer *fb)
-{
-        pandecode_log("struct mali_render_target rts_list_%"PRIx64"_%d[] = {\n", gpu_va, job_no);
-        pandecode_indent++;
-
-        for (int i = 0; i < (fb->rt_count_1 + 1); i++) {
-                mali_ptr rt_va = gpu_va + i * sizeof(struct mali_render_target);
-                struct pandecode_mapped_memory *mem =
-                        pandecode_find_mapped_gpu_mem_containing(rt_va);
-                const struct mali_render_target *PANDECODE_PTR_VAR(rt, mem, (mali_ptr) rt_va);
-
-                pandecode_log("{\n");
-                pandecode_indent++;
-
-                pandecode_rt_format(rt->format);
-
-                if (rt->format.block == MALI_BLOCK_AFBC) {
-                        pandecode_log(".afbc = {\n");
-                        pandecode_indent++;
-
-                        char *a = pointer_as_memory_reference(rt->afbc.metadata);
-                        pandecode_prop("metadata = %s", a);
-                        free(a);
-
-                        pandecode_prop("stride = %d", rt->afbc.stride);
-
-                        pandecode_log(".flags = ");
-                        pandecode_log_decoded_flags(afbc_fmt_flag_info, rt->afbc.flags);
-                        pandecode_log_cont(",\n");
-
-                        pandecode_indent--;
-                        pandecode_log("},\n");
-                } else if (rt->afbc.metadata || rt->afbc.stride || rt->afbc.flags) {
-                        pandecode_msg("XXX: AFBC disabled but AFBC field set (0x%lX, 0x%x, 0x%x)\n",
-                                        rt->afbc.metadata,
-                                        rt->afbc.stride,
-                                        rt->afbc.flags);
-                }
-
-                MEMORY_PROP(rt, framebuffer);
-                pandecode_prop("framebuffer_stride = %d", rt->framebuffer_stride);
-
-                if (rt->layer_stride)
-                        pandecode_prop("layer_stride = %d", rt->layer_stride);
-
-                if (rt->clear_color_1 | rt->clear_color_2 | rt->clear_color_3 | rt->clear_color_4) {
-                        pandecode_prop("clear_color_1 = 0x%" PRIx32, rt->clear_color_1);
-                        pandecode_prop("clear_color_2 = 0x%" PRIx32, rt->clear_color_2);
-                        pandecode_prop("clear_color_3 = 0x%" PRIx32, rt->clear_color_3);
-                        pandecode_prop("clear_color_4 = 0x%" PRIx32, rt->clear_color_4);
-                }
-
-                if (rt->zero1 || rt->zero2) {
-                        pandecode_msg("XXX: render target zeros tripped\n");
-                        pandecode_prop("zero1 = 0x%" PRIx64, rt->zero1);
-                        pandecode_prop("zero2 = 0x%" PRIx32, rt->zero2);
-                }
-
-                pandecode_indent--;
-                pandecode_log("},\n");
-        }
-
-        pandecode_indent--;
-        pandecode_log("};\n");
-}
-
-static struct pandecode_fbd
-pandecode_mfbd_bfr(uint64_t gpu_va, int job_no, bool is_fragment, bool is_compute, bool is_bifrost)
-{
-        struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
-        const struct mali_framebuffer *PANDECODE_PTR_VAR(fb, mem, (mali_ptr) gpu_va);
-
-        struct pandecode_fbd info;
-
-        if (is_bifrost && fb->msaa.sample_locations) {
-                /* The blob stores all possible sample locations in a single buffer
-                 * allocated on startup, and just switches the pointer when switching
-                 * MSAA state. For now, we just put the data into the cmdstream, but we
-                 * should do something like what the blob does with a real driver.
-                 *
-                 * There seem to be 32 slots for sample locations, followed by another
-                 * 16. The second 16 is just the center location followed by 15 zeros
-                 * in all the cases I've identified (maybe shader vs. depth/color
-                 * samples?).
-                 */
-
-                struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(fb->msaa.sample_locations);
-
-                const u16 *PANDECODE_PTR_VAR(samples, smem, fb->msaa.sample_locations);
-
-                pandecode_log("uint16_t sample_locations_%d[] = {\n", job_no);
-                pandecode_indent++;
-
-                for (int i = 0; i < 32 + 16; i++) {
-                        pandecode_log("%d, %d,\n", samples[2 * i], samples[2 * i + 1]);
-                }
-
-                pandecode_indent--;
-                pandecode_log("};\n");
-        }
-        pandecode_log("struct mali_framebuffer framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
-        pandecode_indent++;
-
-        if (is_bifrost) {
-                pandecode_log(".msaa = {\n");
-                pandecode_indent++;
-
-                if (fb->msaa.sample_locations)
-                        pandecode_prop("sample_locations = sample_locations_%d", job_no);
-                else
-                        pandecode_msg("XXX: sample_locations missing\n");
-
-                if (fb->msaa.zero1 || fb->msaa.zero2 || fb->msaa.zero4) {
-                        pandecode_msg("XXX: multisampling zero tripped\n");
-                        pandecode_prop("zero1 = %" PRIx64, fb->msaa.zero1);
-                        pandecode_prop("zero2 = %" PRIx64, fb->msaa.zero2);
-                        pandecode_prop("zero4 = %" PRIx64, fb->msaa.zero4);
-                }
-
-                pandecode_indent--;
-                pandecode_log("},\n");
-        } else {
-                pandecode_log(".shared_memory = {\n");
-                pandecode_indent++;
-                pandecode_shared_memory(&fb->shared_memory, is_compute);
-                pandecode_indent--;
-                pandecode_log("},\n");
-        }
-
-        info.width = fb->width1 + 1;
-        info.height = fb->height1 + 1;
-        info.rt_count = fb->rt_count_1 + 1;
-
-        pandecode_prop("width1 = MALI_POSITIVE(%d)", fb->width1 + 1);
-        pandecode_prop("height1 = MALI_POSITIVE(%d)", fb->height1 + 1);
-        pandecode_prop("width2 = MALI_POSITIVE(%d)", fb->width2 + 1);
-        pandecode_prop("height2 = MALI_POSITIVE(%d)", fb->height2 + 1);
-
-        pandecode_prop("unk1 = 0x%x", fb->unk1);
-        pandecode_prop("unk2 = 0x%x", fb->unk2);
-        pandecode_prop("rt_count_1 = MALI_POSITIVE(%d)", fb->rt_count_1 + 1);
-        pandecode_prop("rt_count_2 = %d", fb->rt_count_2);
-
-        pandecode_log(".mfbd_flags = ");
-        pandecode_log_decoded_flags(mfbd_flag_info, fb->mfbd_flags);
-        pandecode_log_cont(",\n");
-
-        if (fb->clear_stencil)
-                pandecode_prop("clear_stencil = 0x%x", fb->clear_stencil);
-
-        if (fb->clear_depth)
-                pandecode_prop("clear_depth = %f", fb->clear_depth);
-
-        if (!is_compute)
-                if (is_bifrost)
-                        pandecode_bifrost_tiler_descriptor(fb);
-                else {
-                        const struct midgard_tiler_descriptor t = fb->tiler;
-                        pandecode_midgard_tiler_descriptor(&t, fb->width1 + 1, fb->height1 + 1, is_fragment, true);
-                }
-        else
-                pandecode_msg("XXX: skipping compute MFBD, fixme\n");
-
-        if (fb->zero3 || fb->zero4) {
-                pandecode_msg("XXX: framebuffer zeros tripped\n");
-                pandecode_prop("zero3 = 0x%" PRIx32, fb->zero3);
-                pandecode_prop("zero4 = 0x%" PRIx32, fb->zero4);
-        }
-
-        pandecode_indent--;
-        pandecode_log("};\n");
-
-        gpu_va += sizeof(struct mali_framebuffer);
-
-        info.has_extra = (fb->mfbd_flags & MALI_MFBD_EXTRA) && is_fragment;
-
-        if (info.has_extra) {
-                mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
-                const struct mali_framebuffer_extra *PANDECODE_PTR_VAR(fbx, mem, (mali_ptr) gpu_va);
-
-                pandecode_log("struct mali_framebuffer_extra fb_extra_%"PRIx64"_%d = {\n", gpu_va, job_no);
-                pandecode_indent++;
-
-                MEMORY_PROP(fbx, checksum);
-
-                if (fbx->checksum_stride)
-                        pandecode_prop("checksum_stride = %d", fbx->checksum_stride);
-
-                pandecode_log(".flags_hi = ");
-                pandecode_log_decoded_flags(mfbd_extra_flag_hi_info, fbx->flags_hi);
-                pandecode_log_cont(",\n");
-
-                pandecode_log(".flags_lo = ");
-                pandecode_log_decoded_flags(mfbd_extra_flag_lo_info, fbx->flags_lo);
-                pandecode_log_cont(",\n");
-
-                pandecode_prop("zs_block = %s", pandecode_block_format(fbx->zs_block));
-                pandecode_prop("zs_samples = MALI_POSITIVE(%u)", fbx->zs_samples + 1);
-
-                if (fbx->zs_block == MALI_BLOCK_AFBC) {
-                        pandecode_log(".ds_afbc = {\n");
-                        pandecode_indent++;
-
-                        MEMORY_PROP_DIR(fbx->ds_afbc, depth_stencil_afbc_metadata);
-                        pandecode_prop("depth_stencil_afbc_stride = %d",
-                                       fbx->ds_afbc.depth_stencil_afbc_stride);
-                        MEMORY_PROP_DIR(fbx->ds_afbc, depth_stencil);
-
-                        pandecode_log(".flags = ");
-                        pandecode_log_decoded_flags(afbc_fmt_flag_info, fbx->ds_afbc.flags);
-                        pandecode_log_cont(",\n");
-
-                        if (fbx->ds_afbc.padding) {
-                                pandecode_msg("XXX: Depth/stencil AFBC zeros tripped\n");
-                                pandecode_prop("padding = 0x%" PRIx64, fbx->ds_afbc.padding);
-                        }
-
-                        pandecode_indent--;
-                        pandecode_log("},\n");
-                } else {
-                        pandecode_log(".ds_linear = {\n");
-                        pandecode_indent++;
-
-                        if (fbx->ds_linear.depth) {
-                                MEMORY_PROP_DIR(fbx->ds_linear, depth);
-                                pandecode_prop("depth_stride = %d",
-                                               fbx->ds_linear.depth_stride);
-                                pandecode_prop("depth_layer_stride = %d",
-                                               fbx->ds_linear.depth_layer_stride);
-                        } else if (fbx->ds_linear.depth_stride || fbx->ds_linear.depth_layer_stride) {
-                                pandecode_msg("XXX: depth stride zero tripped %d %d\n", fbx->ds_linear.depth_stride, fbx->ds_linear.depth_layer_stride);
-                        }
-
-                        if (fbx->ds_linear.stencil) {
-                                MEMORY_PROP_DIR(fbx->ds_linear, stencil);
-                                pandecode_prop("stencil_stride = %d",
-                                               fbx->ds_linear.stencil_stride);
-                                pandecode_prop("stencil_layer_stride = %d",
-                                               fbx->ds_linear.stencil_layer_stride);
-                        } else if (fbx->ds_linear.stencil_stride || fbx->ds_linear.stencil_layer_stride) {
-                                pandecode_msg("XXX: stencil stride zero tripped %d %d\n", fbx->ds_linear.stencil_stride, fbx->ds_linear.stencil_layer_stride);
-                        }
-
-                        if (fbx->ds_linear.depth_stride_zero ||
-                            fbx->ds_linear.stencil_stride_zero) {
-                                pandecode_msg("XXX: Depth/stencil zeros tripped\n");
-                                pandecode_prop("depth_stride_zero = 0x%x",
-                                               fbx->ds_linear.depth_stride_zero);
-                                pandecode_prop("stencil_stride_zero = 0x%x",
-                                               fbx->ds_linear.stencil_stride_zero);
-                        }
-
-                        pandecode_indent--;
-                        pandecode_log("},\n");
-                }
-
-                if (fbx->clear_color_1 | fbx->clear_color_2) {
-                        pandecode_prop("clear_color_1 = 0x%" PRIx32, fbx->clear_color_1);
-                        pandecode_prop("clear_color_2 = 0x%" PRIx32, fbx->clear_color_2);
-                }
-
-                if (fbx->zero3) {
-                        pandecode_msg("XXX: fb_extra zeros tripped\n");
-                        pandecode_prop("zero3 = 0x%" PRIx64, fbx->zero3);
-                }
-
-                pandecode_indent--;
-                pandecode_log("};\n");
-
-                gpu_va += sizeof(struct mali_framebuffer_extra);
-        }
-
-        if (is_fragment)
-                pandecode_render_target(gpu_va, job_no, fb);
-
-        return info;
-}
-
-/* Just add a comment decoding the shift/odd fields forming the padded vertices
- * count */
-
-static void
-pandecode_padded_vertices(unsigned shift, unsigned k)
-{
-        unsigned odd = 2*k + 1;
-        unsigned pot = 1 << shift;
-        pandecode_msg("padded_num_vertices = %d\n", odd * pot);
-}
-
-/* Given a magic divisor, recover what we were trying to divide by.
- *
- * Let m represent the magic divisor. By definition, m is an element on Z, whre
- * 0 <= m < 2^N, for N bits in m.
- *
- * Let q represent the number we would like to divide by.
- *
- * By definition of a magic divisor for N-bit unsigned integers (a number you
- * multiply by to magically get division), m is a number such that:
- *
- *      (m * x) & (2^N - 1) = floor(x/q).
- *      for all x on Z where 0 <= x < 2^N
- *
- * Ignore the case where any of the above values equals zero; it is irrelevant
- * for our purposes (instanced arrays).
- *
- * Choose x = q. Then:
- *
- *      (m * x) & (2^N - 1) = floor(x/q).
- *      (m * q) & (2^N - 1) = floor(q/q).
- *
- *      floor(q/q) = floor(1) = 1, therefore:
- *
- *      (m * q) & (2^N - 1) = 1
- *
- * Recall the identity that the bitwise AND of one less than a power-of-two
- * equals the modulo with that power of two, i.e. for all x:
- *
- *      x & (2^N - 1) = x % N
- *
- * Therefore:
- *
- *      mq % (2^N) = 1
- *
- * By definition, a modular multiplicative inverse of a number m is the number
- * q such that with respect to a modulos M:
- *
- *      mq % M = 1
- *
- * Therefore, q is the modular multiplicative inverse of m with modulus 2^N.
- *
- */
-
-static void
-pandecode_magic_divisor(uint32_t magic, unsigned shift, unsigned orig_divisor, unsigned extra)
-{
-#if 0
-        /* Compute the modular inverse of `magic` with respect to 2^(32 -
-         * shift) the most lame way possible... just repeatedly add.
-         * Asymptoptically slow but nobody cares in practice, unless you have
-         * massive numbers of vertices or high divisors. */
-
-        unsigned inverse = 0;
-
-        /* Magic implicitly has the highest bit set */
-        magic |= (1 << 31);
-
-        /* Depending on rounding direction */
-        if (extra)
-                magic++;
-
-        for (;;) {
-                uint32_t product = magic * inverse;
-
-                if (shift) {
-                        product >>= shift;
-                }
-
-                if (product == 1)
-                        break;
-
-                ++inverse;
-        }
-
-        pandecode_msg("dividing by %d (maybe off by two)\n", inverse);
-
-        /* Recall we're supposed to divide by (gl_level_divisor *
-         * padded_num_vertices) */
-
-        unsigned padded_num_vertices = inverse / orig_divisor;
-
-        pandecode_msg("padded_num_vertices = %d\n", padded_num_vertices);
-#endif
-}
-
-static void
-pandecode_attributes(const struct pandecode_mapped_memory *mem,
-                            mali_ptr addr, int job_no, char *suffix,
-                            int count, bool varying, enum mali_job_type job_type)
-{
-        char *prefix = varying ? "varying" : "attribute";
-        assert(addr);
-
-        if (!count) {
-                pandecode_msg("warn: No %s records\n", prefix);
-                return;
-        }
-
-        union mali_attr *attr = pandecode_fetch_gpu_mem(mem, addr, sizeof(union mali_attr) * count);
-
-        for (int i = 0; i < count; ++i) {
-                /* First, check for special records */
-                if (attr[i].elements < MALI_RECORD_SPECIAL) {
-                        if (attr[i].size)
-                                pandecode_msg("XXX: tripped size=%d\n", attr[i].size);
-
-                        if (attr[i].stride) {
-                                /* gl_InstanceID passes a magic divisor in the
-                                 * stride field to divide by the padded vertex
-                                 * count. No other records should do so, so
-                                 * stride should otherwise be zero. Note that
-                                 * stride in the usual attribute sense doesn't
-                                 * apply to special records. */
-
-                                bool has_divisor = attr[i].elements == MALI_ATTR_INSTANCEID;
-
-                                pandecode_log_cont("/* %smagic divisor = %X */ ",
-                                                has_divisor ? "" : "XXX: ", attr[i].stride);
-                        }
-
-                        if (attr[i].shift || attr[i].extra_flags) {
-                                /* Attributes use these fields for
-                                 * instancing/padding/etc type issues, but
-                                 * varyings don't */
-
-                                pandecode_log_cont("/* %sshift=%d, extra=%d */ ",
-                                                varying ? "XXX: " : "",
-                                                attr[i].shift, attr[i].extra_flags);
-                        }
-
-                        /* Print the special record name */
-                        bool attribute = false;
-                        pandecode_log("%s_%d = %s;\n", prefix, i, pandecode_special_record(attr[i].elements, &attribute));
-
-                        /* Sanity check */
-                        if (attribute == varying)
-                                pandecode_msg("XXX: mismatched special record\n");
-
-                        continue;
-        }
-
-                enum mali_attr_mode mode = attr[i].elements & 7;
-
-                if (mode == MALI_ATTR_UNUSED)
-                        pandecode_msg("XXX: unused attribute record\n");
-
-                /* For non-linear records, we need to print the type of record */
-                if (mode != MALI_ATTR_LINEAR)
-                        pandecode_log_cont("%s ", pandecode_attr_mode_short(mode));
-
-                /* Print the name to link with attr_meta */
-                pandecode_log_cont("%s_%d", prefix, i);
-
-                /* Print the stride and size */
-                pandecode_log_cont("<%u>[%u]", attr[i].stride, attr[i].size);
-
-                /* TODO: Sanity check the quotient itself. It must be equal to
-                 * (or be greater than, if the driver added padding) the padded
-                 * vertex count. */
-
-                /* Finally, print the pointer */
-                mali_ptr raw_elements = attr[i].elements & ~7;
-                char *a = pointer_as_memory_reference(raw_elements);
-                pandecode_log_cont(" = (%s);\n", a);
-                free(a);
-
-                /* Check the pointer */
-                pandecode_validate_buffer(raw_elements, attr[i].size);
-
-                /* shift/extra_flags exist only for instanced */
-                if (attr[i].shift | attr[i].extra_flags) {
-                        /* These are set to random values by the blob for
-                         * varyings, most likely a symptom of uninitialized
-                         * memory where the hardware masked the bug. As such we
-                         * put this at a warning, not an error. */
-
-                        if (mode == MALI_ATTR_LINEAR)
-                                pandecode_msg("warn: instancing fields set for linear\n");
-
-                        pandecode_prop("shift = %d", attr[i].shift);
-                        pandecode_prop("extra_flags = %d", attr[i].extra_flags);
-                }
-
-                /* Decode further where possible */
-
-                if (mode == MALI_ATTR_MODULO) {
-                        pandecode_padded_vertices(
-                                attr[i].shift,
-                                attr[i].extra_flags);
-                }
-
-                if (mode == MALI_ATTR_NPOT_DIVIDE) {
-                        i++;
-                        pandecode_log("{\n");
-                        pandecode_indent++;
-                        pandecode_prop("unk = 0x%x", attr[i].unk);
-                        pandecode_prop("magic_divisor = 0x%08x", attr[i].magic_divisor);
-                        if (attr[i].zero != 0)
-                                pandecode_prop("XXX: zero tripped (0x%x)\n", attr[i].zero);
-                        pandecode_prop("divisor = %d", attr[i].divisor);
-                        pandecode_magic_divisor(attr[i].magic_divisor, attr[i - 1].shift, attr[i].divisor, attr[i - 1].extra_flags);
-                        pandecode_indent--;
-                        pandecode_log("}, \n");
-                }
-
-        }
-
-        pandecode_log("\n");
-}
-
-static mali_ptr
-pandecode_shader_address(const char *name, mali_ptr ptr)
-{
-        /* TODO: Decode flags */
-        mali_ptr shader_ptr = ptr & ~15;
-
-        char *a = pointer_as_memory_reference(shader_ptr);
-        pandecode_prop("%s = (%s) | %d", name, a, (int) (ptr & 15));
-        free(a);
-
-        return shader_ptr;
-}
-
-static void
-pandecode_stencil(const char *name, const struct mali_stencil_test *stencil)
-{
-        unsigned any_nonzero =
-                stencil->ref | stencil->mask | stencil->func |
-                stencil->sfail | stencil->dpfail | stencil->dppass;
-
-        if (any_nonzero == 0)
-                return;
-
-        const char *func = pandecode_func(stencil->func);
-        const char *sfail = pandecode_stencil_op(stencil->sfail);
-        const char *dpfail = pandecode_stencil_op(stencil->dpfail);
-        const char *dppass = pandecode_stencil_op(stencil->dppass);
-
-        if (stencil->zero)
-                pandecode_msg("XXX: stencil zero tripped: %X\n", stencil->zero);
-
-        pandecode_log(".stencil_%s = {\n", name);
-        pandecode_indent++;
-        pandecode_prop("ref = %d", stencil->ref);
-        pandecode_prop("mask = 0x%02X", stencil->mask);
-        pandecode_prop("func = %s", func);
-        pandecode_prop("sfail = %s", sfail);
-        pandecode_prop("dpfail = %s", dpfail);
-        pandecode_prop("dppass = %s", dppass);
-        pandecode_indent--;
-        pandecode_log("},\n");
-}
-
-static void
-pandecode_blend_equation(const struct mali_blend_equation *blend)
-{
-        if (blend->zero1)
-                pandecode_msg("XXX: blend zero tripped: %X\n", blend->zero1);
-
-        pandecode_log(".equation = {\n");
-        pandecode_indent++;
-
-        pandecode_prop("rgb_mode = 0x%X", blend->rgb_mode);
-        pandecode_prop("alpha_mode = 0x%X", blend->alpha_mode);
-
-        pandecode_log(".color_mask = ");
-        pandecode_log_decoded_flags(mask_flag_info, blend->color_mask);
-        pandecode_log_cont(",\n");
-
-        pandecode_indent--;
-        pandecode_log("},\n");
-}
-
-/* Decodes a Bifrost blend constant. See the notes in bifrost_blend_rt */
-
-static unsigned
-decode_bifrost_constant(u16 constant)
-{
-        float lo = (float) (constant & 0xFF);
-        float hi = (float) (constant >> 8);
-
-        return (hi / 255.0) + (lo / 65535.0);
-}
-
-static mali_ptr
-pandecode_bifrost_blend(void *descs, int job_no, int rt_no)
-{
-        struct bifrost_blend_rt *b =
-                ((struct bifrost_blend_rt *) descs) + rt_no;
-
-        pandecode_log("struct bifrost_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
-        pandecode_indent++;
-
-        pandecode_prop("flags = 0x%" PRIx16, b->flags);
-        pandecode_prop("constant = 0x%" PRIx8 " /* %f */",
-                       b->constant, decode_bifrost_constant(b->constant));
-
-        /* TODO figure out blend shader enable bit */
-        pandecode_blend_equation(&b->equation);
-
-        pandecode_prop("unk2 = 0x%" PRIx16, b->unk2);
-        pandecode_prop("index = 0x%" PRIx16, b->index);
-
-        pandecode_log(".format = ");
-        pandecode_format_short(b->format, false);
-        pandecode_swizzle(b->swizzle, b->format);
-        pandecode_log_cont(",\n");
-
-        pandecode_prop("swizzle = 0x%" PRIx32, b->swizzle);
-        pandecode_prop("format = 0x%" PRIx32, b->format);
-
-        if (b->zero1) {
-                pandecode_msg("XXX: pandecode_bifrost_blend zero1 tripped\n");
-                pandecode_prop("zero1 = 0x%" PRIx32, b->zero1);
-        }
-
-        pandecode_log(".shader_type = ");
-        switch(b->shader_type) {
-        case BIFROST_BLEND_F16:
-                pandecode_log_cont("BIFROST_BLEND_F16");
-                break;
-        case BIFROST_BLEND_F32:
-                pandecode_log_cont("BIFROST_BLEND_F32");
-                break;
-        case BIFROST_BLEND_I32:
-                pandecode_log_cont("BIFROST_BLEND_I32");
-                break;
-        case BIFROST_BLEND_U32:
-                pandecode_log_cont("BIFROST_BLEND_U32");
-                break;
-        case BIFROST_BLEND_I16:
-                pandecode_log_cont("BIFROST_BLEND_I16");
-                break;
-        case BIFROST_BLEND_U16:
-                pandecode_log_cont("BIFROST_BLEND_U16");
-                break;
-        }
-        pandecode_log_cont(",\n");
-
-        if (b->zero2) {
-                pandecode_msg("XXX: pandecode_bifrost_blend zero2 tripped\n");
-                pandecode_prop("zero2 = 0x%" PRIx32, b->zero2);
-        }
-
-        pandecode_prop("shader = 0x%" PRIx32, b->shader);
-
-        pandecode_indent--;
-        pandecode_log("},\n");
-
-        return 0;
-}
-
-static mali_ptr
-pandecode_midgard_blend(union midgard_blend *blend, bool is_shader)
-{
-        /* constant/equation is in a union */
-        if (!blend->shader)
-                return 0;
-
-        pandecode_log(".blend = {\n");
-        pandecode_indent++;
-
-        if (is_shader) {
-                pandecode_shader_address("shader", blend->shader);
-        } else {
-                pandecode_blend_equation(&blend->equation);
-                pandecode_prop("constant = %f", blend->constant);
-        }
-
-        pandecode_indent--;
-        pandecode_log("},\n");
-
-        /* Return blend shader to disassemble if present */
-        return is_shader ? (blend->shader & ~0xF) : 0;
-}
-
-static mali_ptr
-pandecode_midgard_blend_mrt(void *descs, int job_no, int rt_no)
-{
-        struct midgard_blend_rt *b =
-                ((struct midgard_blend_rt *) descs) + rt_no;
-
-        /* Flags determine presence of blend shader */
-        bool is_shader = (b->flags & 0xF) >= 0x2;
-
-        pandecode_log("struct midgard_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
-        pandecode_indent++;
-
-        pandecode_prop("flags = 0x%" PRIx64, b->flags);
-
-        union midgard_blend blend = b->blend;
-        mali_ptr shader = pandecode_midgard_blend(&blend, is_shader);
-
-        pandecode_indent--;
-        pandecode_log("};\n");
-
-        return shader;
-}
-
-/* Attributes and varyings have descriptor records, which contain information
- * about their format and ordering with the attribute/varying buffers. We'll
- * want to validate that the combinations specified are self-consistent.
- */
-
-static int
-pandecode_attribute_meta(int job_no, int count, const struct mali_vertex_tiler_postfix *v, bool varying, char *suffix)
-{
-        char base[128];
-        char *prefix = varying ? "varying" : "attribute";
-        unsigned max_index = 0;
-        snprintf(base, sizeof(base), "%s_meta", prefix);
-
-        struct mali_attr_meta *attr_meta;
-        mali_ptr p = varying ? v->varying_meta : v->attribute_meta;
-
-        struct pandecode_mapped_memory *attr_mem = pandecode_find_mapped_gpu_mem_containing(p);
-
-        for (int i = 0; i < count; ++i, p += sizeof(struct mali_attr_meta)) {
-                attr_meta = pandecode_fetch_gpu_mem(attr_mem, p,
-                                                    sizeof(*attr_mem));
-
-                /* If the record is discard, it should be zero for everything else */
-
-                if (attr_meta->format == MALI_VARYING_DISCARD) {
-                        uint64_t zero =
-                                attr_meta->index |
-                                attr_meta->unknown1 |
-                                attr_meta->unknown3 |
-                                attr_meta->src_offset;
-
-                        if (zero)
-                                pandecode_msg("XXX: expected empty record for varying discard\n");
-
-                        /* We want to look for a literal 0000 swizzle -- this
-                         * is not encoded with all zeroes, however */
-
-                        enum mali_channel z = MALI_CHANNEL_ZERO;
-                        unsigned zero_swizzle = z | (z << 3) | (z << 6) | (z << 9);
-                        bool good_swizzle = attr_meta->swizzle == zero_swizzle;
-
-                        if (!good_swizzle)
-                                pandecode_msg("XXX: expected zero swizzle for discard\n");
-
-                        if (!varying)
-                                pandecode_msg("XXX: cannot discard attribute\n");
-
-                        /* If we're all good, omit the record */
-                        if (!zero && varying && good_swizzle) {
-                                pandecode_log("/* discarded varying */\n");
-                                continue;
-                        }
-                }
-
-                if (attr_meta->index > max_index)
-                        max_index = attr_meta->index;
-
-                if (attr_meta->unknown1 != 0x2) {
-                        pandecode_msg("XXX: expected unknown1 = 0x2\n");
-                        pandecode_prop("unknown1 = 0x%" PRIx64, (u64) attr_meta->unknown1);
-                }
-
-                if (attr_meta->unknown3) {
-                        pandecode_msg("XXX: unexpected unknown3 set\n");
-                        pandecode_prop("unknown3 = 0x%" PRIx64, (u64) attr_meta->unknown3);
-                }
-
-                pandecode_format_short(attr_meta->format, false);
-                pandecode_log_cont(" %s_%u", prefix, attr_meta->index);
-
-                if (attr_meta->src_offset)
-                        pandecode_log_cont("[%u]", attr_meta->src_offset);
-
-                pandecode_swizzle(attr_meta->swizzle, attr_meta->format);
-
-                pandecode_log_cont(";\n");
-        }
-
-        pandecode_log("\n");
-
-        return count ? (max_index + 1) : 0;
-}
-
-/* return bits [lo, hi) of word */
-static u32
-bits(u32 word, u32 lo, u32 hi)
-{
-        if (hi - lo >= 32)
-                return word; // avoid undefined behavior with the shift
-
-        return (word >> lo) & ((1 << (hi - lo)) - 1);
-}
-
-static void
-pandecode_vertex_tiler_prefix(struct mali_vertex_tiler_prefix *p, int job_no, bool graphics)
-{
-        pandecode_log(".prefix = {\n");
-        pandecode_indent++;
-
-        /* Decode invocation_count. See the comment before the definition of
-         * invocation_count for an explanation.
-         */
-
-        unsigned size_y_shift = bits(p->invocation_shifts, 0, 5);
-        unsigned size_z_shift = bits(p->invocation_shifts, 5, 10);
-        unsigned workgroups_x_shift = bits(p->invocation_shifts, 10, 16);
-        unsigned workgroups_y_shift = bits(p->invocation_shifts, 16, 22);
-        unsigned workgroups_z_shift = bits(p->invocation_shifts, 22, 28);
-        unsigned workgroups_x_shift_2 = bits(p->invocation_shifts, 28, 32);
-
-        unsigned size_x = bits(p->invocation_count, 0, size_y_shift) + 1;
-        unsigned size_y = bits(p->invocation_count, size_y_shift, size_z_shift) + 1;
-        unsigned size_z = bits(p->invocation_count, size_z_shift, workgroups_x_shift) + 1;
-
-        unsigned groups_x = bits(p->invocation_count, workgroups_x_shift, workgroups_y_shift) + 1;
-        unsigned groups_y = bits(p->invocation_count, workgroups_y_shift, workgroups_z_shift) + 1;
-        unsigned groups_z = bits(p->invocation_count, workgroups_z_shift, 32) + 1;
-
-        /* Even though we have this decoded, we want to ensure that the
-         * representation is "unique" so we don't lose anything by printing only
-         * the final result. More specifically, we need to check that we were
-         * passed something in canonical form, since the definition per the
-         * hardware is inherently not unique. How? Well, take the resulting
-         * decode and pack it ourselves! If it is bit exact with what we
-         * decoded, we're good to go. */
-
-        struct mali_vertex_tiler_prefix ref;
-        panfrost_pack_work_groups_compute(&ref, groups_x, groups_y, groups_z, size_x, size_y, size_z, graphics);
-
-        bool canonical =
-                (p->invocation_count == ref.invocation_count) &&
-                (p->invocation_shifts == ref.invocation_shifts);
-
-        if (!canonical) {
-                pandecode_msg("XXX: non-canonical workgroups packing\n");
-                pandecode_msg("expected: %X, %X",
-                                ref.invocation_count,
-                                ref.invocation_shifts);
-
-                pandecode_prop("invocation_count = 0x%" PRIx32, p->invocation_count);
-                pandecode_prop("size_y_shift = %d", size_y_shift);
-                pandecode_prop("size_z_shift = %d", size_z_shift);
-                pandecode_prop("workgroups_x_shift = %d", workgroups_x_shift);
-                pandecode_prop("workgroups_y_shift = %d", workgroups_y_shift);
-                pandecode_prop("workgroups_z_shift = %d", workgroups_z_shift);
-                pandecode_prop("workgroups_x_shift_2 = %d", workgroups_x_shift_2);
-        }
-
-        /* Regardless, print the decode */
-        pandecode_msg("size (%d, %d, %d), count (%d, %d, %d)\n",
-                        size_x, size_y, size_z,
-                        groups_x, groups_y, groups_z);
-
-        /* TODO: Decode */
-        if (p->unknown_draw)
-                pandecode_prop("unknown_draw = 0x%" PRIx32, p->unknown_draw);
-
-        pandecode_prop("workgroups_x_shift_3 = 0x%" PRIx32, p->workgroups_x_shift_3);
-
-        if (p->draw_mode != MALI_DRAW_NONE)
-                pandecode_prop("draw_mode = %s", pandecode_draw_mode(p->draw_mode));
-
-        /* Index count only exists for tiler jobs anyway */
-
-        if (p->index_count)
-                pandecode_prop("index_count = MALI_POSITIVE(%" PRId32 ")", p->index_count + 1);
-
-
-        unsigned index_raw_size = (p->unknown_draw & MALI_DRAW_INDEXED_SIZE);
-        index_raw_size >>= MALI_DRAW_INDEXED_SHIFT;
-
-        /* Validate an index buffer is present if we need one. TODO: verify
-         * relationship between invocation_count and index_count */
-
-        if (p->indices) {
-                unsigned count = p->index_count;
-
-                /* Grab the size */
-                unsigned size = (index_raw_size == 0x3) ? 4 : index_raw_size;
-
-                /* Ensure we got a size, and if so, validate the index buffer
-                 * is large enough to hold a full set of indices of the given
-                 * size */
-
-                if (!index_raw_size)
-                        pandecode_msg("XXX: index size missing\n");
-                else
-                        pandecode_validate_buffer(p->indices, count * size);
-        } else if (index_raw_size)
-                pandecode_msg("XXX: unexpected index size %u\n", index_raw_size);
-
-        if (p->offset_bias_correction)
-                pandecode_prop("offset_bias_correction = %d", p->offset_bias_correction);
-
-        /* TODO: Figure out what this is. It's not zero */
-        pandecode_prop("zero1 = 0x%" PRIx32, p->zero1);
-
-        pandecode_indent--;
-        pandecode_log("},\n");
-}
-
-static void
-pandecode_uniform_buffers(mali_ptr pubufs, int ubufs_count, int job_no)
-{
-        struct pandecode_mapped_memory *umem = pandecode_find_mapped_gpu_mem_containing(pubufs);
-        uint64_t *PANDECODE_PTR_VAR(ubufs, umem, pubufs);
-
-        for (int i = 0; i < ubufs_count; i++) {
-                unsigned size = (ubufs[i] & ((1 << 10) - 1)) * 16;
-                mali_ptr addr = (ubufs[i] >> 10) << 2;
-
-                pandecode_validate_buffer(addr, size);
-
-                char *ptr = pointer_as_memory_reference(addr);
-                pandecode_log("ubuf_%d[%u] = %s;\n", i, size, ptr);
-                free(ptr);
-        }
-
-        pandecode_log("\n");
-}
-
-static void
-pandecode_uniforms(mali_ptr uniforms, unsigned uniform_count)
-{
-        pandecode_validate_buffer(uniforms, uniform_count * 16);
-
-        char *ptr = pointer_as_memory_reference(uniforms);
-        pandecode_log("vec4 uniforms[%u] = %s;\n", uniform_count, ptr);
-        free(ptr);
-}
-
-static const char *
-shader_type_for_job(unsigned type)
-{
-        switch (type) {
-        case JOB_TYPE_VERTEX:  return "VERTEX";
-        case JOB_TYPE_TILER:   return "FRAGMENT";
-        case JOB_TYPE_COMPUTE: return "COMPUTE";
-        default:
-                               return "UNKNOWN";
-        }
-}
-
-static unsigned shader_id = 0;
-
-static struct midgard_disasm_stats
-pandecode_shader_disassemble(mali_ptr shader_ptr, int shader_no, int type,
-                             bool is_bifrost, unsigned gpu_id)
-{
-        struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(shader_ptr);
-        uint8_t *PANDECODE_PTR_VAR(code, mem, shader_ptr);
-
-        /* Compute maximum possible size */
-        size_t sz = mem->length - (shader_ptr - mem->gpu_va);
-
-        /* Print some boilerplate to clearly denote the assembly (which doesn't
-         * obey indentation rules), and actually do the disassembly! */
-
-        pandecode_log_cont("\n\n");
-
-        struct midgard_disasm_stats stats;
-
-        if (is_bifrost) {
-                disassemble_bifrost(pandecode_dump_stream, code, sz, true);
-
-                /* TODO: Extend stats to Bifrost */
-                stats.texture_count = -128;
-                stats.sampler_count = -128;
-                stats.attribute_count = -128;
-                stats.varying_count = -128;
-                stats.uniform_count = -128;
-                stats.uniform_buffer_count = -128;
-                stats.work_count = -128;
-
-                stats.instruction_count = 0;
-                stats.bundle_count = 0;
-                stats.quadword_count = 0;
-                stats.helper_invocations = false;
-        } else {
-                stats = disassemble_midgard(pandecode_dump_stream,
-                                code, sz, gpu_id,
-                                type == JOB_TYPE_TILER ?
-                                MESA_SHADER_FRAGMENT : MESA_SHADER_VERTEX);
-        }
-
-        /* Print shader-db stats. Skip COMPUTE jobs since they are used for
-         * driver-internal purposes with the blob and interfere */
-
-        bool should_shaderdb = type != JOB_TYPE_COMPUTE;
-
-        if (should_shaderdb) {
-                unsigned nr_threads =
-                        (stats.work_count <= 4) ? 4 :
-                        (stats.work_count <= 8) ? 2 :
-                        1;
-
-                pandecode_log_cont("shader%d - MESA_SHADER_%s shader: "
-                        "%u inst, %u bundles, %u quadwords, "
-                        "%u registers, %u threads, 0 loops, 0:0 spills:fills\n\n\n",
-                        shader_id++,
-                        shader_type_for_job(type),
-                        stats.instruction_count, stats.bundle_count, stats.quadword_count,
-                        stats.work_count, nr_threads);
-        }
-
-
-        return stats;
-}
-
-static void
-pandecode_texture_payload(mali_ptr payload,
-                          enum mali_texture_type type,
-                          enum mali_texture_layout layout,
-                          bool manual_stride,
-                          uint8_t levels,
-                          uint16_t depth,
-                          uint16_t array_size,
-                          struct pandecode_mapped_memory *tmem)
-{
-        pandecode_log(".payload = {\n");
-        pandecode_indent++;
-
-        /* A bunch of bitmap pointers follow.
-         * We work out the correct number,
-         * based on the mipmap/cubemap
-         * properties, but dump extra
-         * possibilities to futureproof */
-
-        int bitmap_count = levels + 1;
-
-        /* Miptree for each face */
-        if (type == MALI_TEX_CUBE)
-                bitmap_count *= 6;
-
-        /* Array of layers */
-        bitmap_count *= (depth + 1);
-
-        /* Array of textures */
-        bitmap_count *= (array_size + 1);
-
-        /* Stride for each element */
-        if (manual_stride)
-                bitmap_count *= 2;
-
-        mali_ptr *pointers_and_strides = pandecode_fetch_gpu_mem(tmem,
-                payload, sizeof(mali_ptr) * bitmap_count);
-        for (int i = 0; i < bitmap_count; ++i) {
-                /* How we dump depends if this is a stride or a pointer */
-
-                if (manual_stride && (i & 1)) {
-                        /* signed 32-bit snuck in as a 64-bit pointer */
-                        uint64_t stride_set = pointers_and_strides[i];
-                        uint32_t clamped_stride = stride_set;
-                        int32_t stride = clamped_stride;
-                        assert(stride_set == clamped_stride);
-                        pandecode_log("(mali_ptr) %d /* stride */, \n", stride);
-                } else {
-                        char *a = pointer_as_memory_reference(pointers_and_strides[i]);
-                        pandecode_log("%s, \n", a);
-                        free(a);
-                }
-        }
-
-        pandecode_indent--;
-        pandecode_log("},\n");
-}
-
-static void
-pandecode_texture(mali_ptr u,
-                struct pandecode_mapped_memory *tmem,
-                unsigned job_no, unsigned tex)
-{
-        struct mali_texture_descriptor *PANDECODE_PTR_VAR(t, tmem, u);
-
-        pandecode_log("struct mali_texture_descriptor texture_descriptor_%"PRIx64"_%d_%d = {\n", u, job_no, tex);
-        pandecode_indent++;
-
-        pandecode_prop("width = %" PRId32, t->width);
-        pandecode_prop("height = %" PRId32, t->height);
-        pandecode_prop("depth = %" PRId32, t->depth);
-        pandecode_prop("array_size = %" PRId32, t->array_size);
-
-        pandecode_log("\n");
-        pandecode_prop("f.swizzle = 0x%" PRIx32, t->format.swizzle);
-        pandecode_prop("f.format = 0x%" PRIx32, t->format.format);
-        pandecode_prop("f.srgb = 0x%" PRIx32, t->format.srgb);
-        pandecode_prop("f.unknown1 = 0x%" PRIx32, t->format.unknown1);
-        pandecode_prop("f.type = %" PRId32, t->format.type);
-        pandecode_prop("f.layout = %" PRId32, t->format.layout);
-        pandecode_prop("f.unknown2 = 0x%" PRIx32, t->format.unknown2);
-        pandecode_prop("f.manual_stride = %" PRId32, t->format.manual_stride);
-        pandecode_prop("f.zero = 0x%" PRIx32, t->format.zero);
-        pandecode_log("\n");
-
-        pandecode_prop("unknown3 = 0x%" PRIx32, t->unknown3);
-        pandecode_prop("unknown3A = 0x%" PRIx32, t->unknown3A);
-        pandecode_prop("levels = %" PRId32, t->levels);
-        pandecode_prop("swizzle = 0x%" PRIx32, t->swizzle);
-        pandecode_prop("swizzle_zero = 0x%" PRIx32, t->swizzle_zero);
-
-        pandecode_prop("unknown5 = 0x%" PRIx32, t->unknown5);
-        pandecode_prop("unknown6 = 0x%" PRIx32, t->unknown6);
-        pandecode_prop("unknown7 = 0x%" PRIx32, t->unknown7);
-        pandecode_log("\n");
-
-        struct mali_texture_format f = t->format;
-
-        /* See the definiton of enum mali_texture_type */
-
-        bool is_cube = f.type == MALI_TEX_CUBE;
-        unsigned dimension = is_cube ? 2 : f.type;
-
-        pandecode_make_indent();
-
-        /* Print the layout. Default is linear; a modifier can denote AFBC or
-         * u-interleaved/tiled modes */
-
-        if (f.layout == MALI_TEXTURE_AFBC)
-                pandecode_log_cont("afbc");
-        else if (f.layout == MALI_TEXTURE_TILED)
-                pandecode_log_cont("tiled");
-        else if (f.layout == MALI_TEXTURE_LINEAR)
-                pandecode_log_cont("linear");
-        else
-                pandecode_msg("XXX: invalid texture layout 0x%X\n", f.layout);
-
-        pandecode_swizzle(t->swizzle, f.format);
-        pandecode_log_cont(" ");
-
-        /* Distinguish cube/2D with modifier */
-
-        if (is_cube)
-                pandecode_log_cont("cube ");
-
-        pandecode_format_short(f.format, f.srgb);
-        pandecode_swizzle(f.swizzle, f.format);
-
-        /* All four width/height/depth/array_size dimensions are present
-         * regardless of the type of texture, but it is an error to have
-         * non-zero dimensions for unused dimensions. Verify this. array_size
-         * can always be set, as can width. Depth used for MSAA. */
-
-        if (t->height && dimension < 2)
-                pandecode_msg("XXX: nonzero height for <2D texture\n");
-
-        /* Print only the dimensions that are actually there */
-
-        pandecode_log_cont(": %d", t->width + 1);
-
-        if (t->height || t->depth)
-                pandecode_log_cont("x%u", t->height + 1);
-
-        if (t->depth)
-                pandecode_log_cont("x%u", t->depth + 1);
-
-        if (t->array_size)
-                pandecode_log_cont("[%u]", t->array_size + 1);
-
-        if (t->levels)
-                pandecode_log_cont(" mip %u", t->levels);
-
-        pandecode_log_cont("\n");
-
-        if (f.unknown1 | f.zero) {
-                pandecode_msg("XXX: texture format zero tripped\n");
-                pandecode_prop("unknown1 = %" PRId32, f.unknown1);
-                pandecode_prop("zero = %" PRId32, f.zero);
-        }
-
-        if (!f.unknown2) {
-                pandecode_msg("XXX: expected unknown texture bit set\n");
-                pandecode_prop("unknown2 = %" PRId32, f.unknown2);
-        }
-
-        if (t->swizzle_zero) {
-                pandecode_msg("XXX: swizzle zero tripped\n");
-                pandecode_prop("swizzle_zero = %d", t->swizzle_zero);
-        }
-
-        if (t->unknown3 | t->unknown3A | t->unknown5 | t->unknown6 | t->unknown7) {
-                pandecode_msg("XXX: texture zero tripped\n");
-                pandecode_prop("unknown3 = %" PRId16, t->unknown3);
-                pandecode_prop("unknown3A = %" PRId8, t->unknown3A);
-                pandecode_prop("unknown5 = 0x%" PRIx32, t->unknown5);
-                pandecode_prop("unknown6 = 0x%" PRIx32, t->unknown6);
-                pandecode_prop("unknown7 = 0x%" PRIx32, t->unknown7);
-        }
-
-        pandecode_texture_payload(u + sizeof(*t), f.type, f.layout, f.manual_stride, t->levels, t->depth, t->array_size, tmem);
-
-        pandecode_indent--;
-        pandecode_log("};\n");
-}
-
-static void
-pandecode_bifrost_texture(
-                const struct bifrost_texture_descriptor *t,
-                unsigned job_no,
-                unsigned tex)
-{
-        pandecode_log("struct bifrost_texture_descriptor texture_descriptor_%d_%d = {\n", job_no, tex);
-        pandecode_indent++;
-
-        pandecode_prop("format_unk = 0x%" PRIx32, t->format_unk);
-        pandecode_prop("type = %" PRId32, t->type);
-
-        if (t->zero) {
-                pandecode_msg("XXX: zero tripped\n");
-                pandecode_prop("zero = 0x%" PRIx32, t->zero);
-        }
-
-        pandecode_prop("format_swizzle = 0x%" PRIx32, t->format_swizzle);
-        pandecode_prop("format = 0x%" PRIx32, t->format);
-        pandecode_prop("srgb = 0x%" PRIx32, t->srgb);
-        pandecode_prop("format_unk3 = 0x%" PRIx32, t->format_unk3);
-        pandecode_prop("width = %" PRId32, t->width);
-        pandecode_prop("height = %" PRId32, t->height);
-        pandecode_prop("swizzle = 0x%" PRIx32, t->swizzle);
-        pandecode_prop("levels = %" PRId32, t->levels);
-        pandecode_prop("unk1 = 0x%" PRIx32, t->unk1);
-        pandecode_prop("levels_unk = %" PRId32, t->levels_unk);
-        pandecode_prop("level_2 = %" PRId32, t->level_2);
-        pandecode_prop("payload = 0x%" PRIx64, t->payload);
-        pandecode_prop("array_size = %" PRId32, t->array_size);
-        pandecode_prop("unk4 = 0x%" PRIx32, t->unk4);
-        pandecode_prop("depth = %" PRId32, t->depth);
-        pandecode_prop("unk5 = 0x%" PRIx32, t->unk5);
-        pandecode_log("\n");
-
-        /* See the definiton of enum mali_texture_type */
-
-        bool is_cube = t->type == MALI_TEX_CUBE;
-        unsigned dimension = is_cube ? 2 : t->type;
-
-        /* Print the layout. Default is linear; a modifier can denote AFBC or
-         * u-interleaved/tiled modes */
-
-        if (t->layout == MALI_TEXTURE_AFBC)
-                pandecode_log_cont("afbc");
-        else if (t->layout == MALI_TEXTURE_TILED)
-                pandecode_log_cont("tiled");
-        else if (t->layout == MALI_TEXTURE_LINEAR)
-                pandecode_log_cont("linear");
-        else
-                pandecode_msg("XXX: invalid texture layout 0x%X\n", t->layout);
-
-        pandecode_swizzle(t->swizzle, t->format);
-        pandecode_log_cont(" ");
-
-        /* Distinguish cube/2D with modifier */
-
-        if (is_cube)
-                pandecode_log_cont("cube ");
-
-        pandecode_format_short(t->format, t->srgb);
-
-        /* All four width/height/depth/array_size dimensions are present
-         * regardless of the type of texture, but it is an error to have
-         * non-zero dimensions for unused dimensions. Verify this. array_size
-         * can always be set, as can width. */
-
-        if (t->height && dimension < 2)
-                pandecode_msg("XXX: nonzero height for <2D texture\n");
-
-        if (t->depth && dimension < 3)
-                pandecode_msg("XXX: nonzero depth for <2D texture\n");
-
-        /* Print only the dimensions that are actually there */
-
-        pandecode_log_cont(": %d", t->width + 1);
-
-        if (dimension >= 2)
-                pandecode_log_cont("x%u", t->height + 1);
-
-        if (dimension >= 3)
-                pandecode_log_cont("x%u", t->depth + 1);
-
-        if (t->array_size)
-                pandecode_log_cont("[%u]", t->array_size + 1);
-
-        if (t->levels)
-                pandecode_log_cont(" mip %u", t->levels);
-
-        pandecode_log_cont("\n");
-
-        struct pandecode_mapped_memory *tmem = pandecode_find_mapped_gpu_mem_containing(t->payload);
-        if (t->payload) {
-                pandecode_texture_payload(t->payload, t->type, t->layout,
-                                          true, t->levels, t->depth,
-                                          t->array_size, tmem);
-        }
-
-        pandecode_indent--;
-        pandecode_log("};\n");
-}
-
-/* For shader properties like texture_count, we have a claimed property in the shader_meta, and the actual Truth from static analysis (this may just be an upper limit). We validate accordingly */
-
-static void
-pandecode_shader_prop(const char *name, unsigned claim, signed truth, bool fuzzy)
-{
-        /* Nothing to do */
-        if (claim == truth)
-                return;
-
-        if (fuzzy && (truth < 0))
-                pandecode_msg("XXX: fuzzy %s, claimed %d, expected %d\n", name, claim, truth);
-
-        if ((truth >= 0) && !fuzzy) {
-                pandecode_msg("%s: expected %s = %d, claimed %u\n",
-                                (truth < claim) ? "warn" : "XXX",
-                                name, truth, claim);
-        } else if ((claim > -truth) && !fuzzy) {
-                pandecode_msg("XXX: expected %s <= %u, claimed %u\n",
-                                name, -truth, claim);
-        } else if (fuzzy && (claim < truth))
-                pandecode_msg("XXX: expected %s >= %u, claimed %u\n",
-                                name, truth, claim);
-
-        pandecode_log(".%s = %" PRId16, name, claim);
-
-        if (fuzzy)
-                pandecode_log_cont(" /* %u used */", truth);
-
-        pandecode_log_cont(",\n");
-}
-
-static void
-pandecode_blend_shader_disassemble(mali_ptr shader, int job_no, int job_type,
-                                   bool is_bifrost, unsigned gpu_id)
-{
-        struct midgard_disasm_stats stats =
-                pandecode_shader_disassemble(shader, job_no, job_type, is_bifrost, gpu_id);
-
-        bool has_texture = (stats.texture_count > 0);
-        bool has_sampler = (stats.sampler_count > 0);
-        bool has_attribute = (stats.attribute_count > 0);
-        bool has_varying = (stats.varying_count > 0);
-        bool has_uniform = (stats.uniform_count > 0);
-        bool has_ubo = (stats.uniform_buffer_count > 0);
-
-        if (has_texture || has_sampler)
-                pandecode_msg("XXX: blend shader accessing textures\n");
-
-        if (has_attribute || has_varying)
-                pandecode_msg("XXX: blend shader accessing interstage\n");
-
-        if (has_uniform || has_ubo)
-                pandecode_msg("XXX: blend shader accessing uniforms\n");
-}
-
-static void
-pandecode_textures(mali_ptr textures, unsigned texture_count, int job_no, bool is_bifrost)
-{
-        struct pandecode_mapped_memory *mmem = pandecode_find_mapped_gpu_mem_containing(textures);
-
-        if (!mmem)
-                return;
-
-        if (is_bifrost) {
-                const struct bifrost_texture_descriptor *PANDECODE_PTR_VAR(t, mmem, textures);
-
-                pandecode_log("uint64_t textures_%"PRIx64"_%d[] = {\n", textures, job_no);
-                pandecode_indent++;
-
-                for (unsigned tex = 0; tex < texture_count; ++tex)
-                        pandecode_bifrost_texture(&t[tex], job_no, tex);
-
-                pandecode_indent--;
-                pandecode_log("};\n");
-        } else {
-                mali_ptr *PANDECODE_PTR_VAR(u, mmem, textures);
-
-                pandecode_log("uint64_t textures_%"PRIx64"_%d[] = {\n", textures, job_no);
-                pandecode_indent++;
-
-                for (int tex = 0; tex < texture_count; ++tex) {
-                        mali_ptr *PANDECODE_PTR_VAR(u, mmem, textures + tex * sizeof(mali_ptr));
-                        char *a = pointer_as_memory_reference(*u);
-                        pandecode_log("%s,\n", a);
-                        free(a);
-                }
-
-                pandecode_indent--;
-                pandecode_log("};\n");
-
-                /* Now, finally, descend down into the texture descriptor */
-                for (unsigned tex = 0; tex < texture_count; ++tex) {
-                        mali_ptr *PANDECODE_PTR_VAR(u, mmem, textures + tex * sizeof(mali_ptr));
-                        struct pandecode_mapped_memory *tmem = pandecode_find_mapped_gpu_mem_containing(*u);
-                        if (tmem)
-                                pandecode_texture(*u, tmem, job_no, tex);
-                }
-        }
-}
-
-static void
-pandecode_samplers(mali_ptr samplers, unsigned sampler_count, int job_no, bool is_bifrost)
-{
-        struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(samplers);
-
-        if (!smem)
-                return;
-
-        if (is_bifrost) {
-                struct bifrost_sampler_descriptor *s;
-
-                for (int i = 0; i < sampler_count; ++i) {
-                        s = pandecode_fetch_gpu_mem(smem, samplers + sizeof(*s) * i, sizeof(*s));
-
-                        pandecode_log("struct bifrost_sampler_descriptor sampler_descriptor_%"PRIx64"_%d_%d = {\n", samplers + sizeof(*s) * i, job_no, i);
-                        pandecode_indent++;
-
-                        if (s->unk1 != 1) {
-                                pandecode_msg("XXX: unk1 tripped\n");
-                                pandecode_prop("unk1 = 0x%x", s->unk1);
-                        }
-
-                        pandecode_prop("wrap_s = %s", pandecode_wrap_mode(s->wrap_s));
-                        pandecode_prop("wrap_t = %s", pandecode_wrap_mode(s->wrap_t));
-                        pandecode_prop("wrap_r = %s", pandecode_wrap_mode(s->wrap_r));
-
-                        if (s->unk8 != 0x8) {
-                                pandecode_msg("XXX: unk8 tripped\n");
-                                pandecode_prop("unk8 = 0x%x", s->unk8);
-                        }
-
-                        pandecode_prop("unk2 = 0x%x", s->unk2);
-                        pandecode_prop("unk3 = 0x%x", s->unk3);
-                        pandecode_prop("min_filter = %s", s->min_filter ? "nearest" : "linear");
-                        pandecode_prop("norm_coords = 0x%x", s->norm_coords & 0x1);
-                        pandecode_prop("zero1 = 0x%x", s->zero1 & 0x1);
-                        pandecode_prop("mip_filter = %s", s->mip_filter ? "linear" : "nearest");
-                        pandecode_prop("mag_filter = %s", s->mag_filter ? "linear" : "nearest");
-
-                        pandecode_prop("min_lod = FIXED_16(%f)", DECODE_FIXED_16(s->min_lod));
-                        pandecode_prop("max_lod = FIXED_16(%f)", DECODE_FIXED_16(s->max_lod));
-
-                        if (s->zero1 || s->zero2 || s->zero3 || s->zero4) {
-                                pandecode_msg("XXX: sampler zero tripped\n");
-                                pandecode_prop("zero = 0x%" PRIx8 ", 0x%" PRIx64 ", 0x%" PRIx64 ", 0x%" PRIx64 "\n", s->zero1, s->zero2, s->zero3, s->zero4);
-                        }
-
-                        pandecode_indent--;
-                        pandecode_log("};\n");
-                }
-        } else {
-                struct mali_sampler_descriptor *s;
-
-                for (int i = 0; i < sampler_count; ++i) {
-                        s = pandecode_fetch_gpu_mem(smem, samplers + sizeof(*s) * i, sizeof(*s));
-
-                        pandecode_log("struct mali_sampler_descriptor sampler_descriptor_%"PRIx64"_%d_%d = {\n", samplers + sizeof(*s) * i, job_no, i);
-                        pandecode_indent++;
-
-                        pandecode_log(".filter_mode = ");
-                        pandecode_log_decoded_flags(sampler_flag_info, s->filter_mode);
-                        pandecode_log_cont(",\n");
-
-                        pandecode_prop("min_lod = FIXED_16(%f)", DECODE_FIXED_16(s->min_lod));
-                        pandecode_prop("max_lod = FIXED_16(%f)", DECODE_FIXED_16(s->max_lod));
-
-                        if (s->lod_bias)
-                                pandecode_prop("lod_bias = FIXED_16(%f)", DECODE_FIXED_16(s->lod_bias));
-
-                        pandecode_prop("wrap_s = %s", pandecode_wrap_mode(s->wrap_s));
-                        pandecode_prop("wrap_t = %s", pandecode_wrap_mode(s->wrap_t));
-                        pandecode_prop("wrap_r = %s", pandecode_wrap_mode(s->wrap_r));
-
-                        pandecode_prop("compare_func = %s", pandecode_func(s->compare_func));
-
-                        if (s->zero || s->zero2) {
-                                pandecode_msg("XXX: sampler zero tripped\n");
-                                pandecode_prop("zero = 0x%X, 0x%X\n", s->zero, s->zero2);
-                        }
-
-                        pandecode_prop("seamless_cube_map = %d", s->seamless_cube_map);
-
-                        pandecode_prop("border_color = { %f, %f, %f, %f }",
-                                       s->border_color[0],
-                                       s->border_color[1],
-                                       s->border_color[2],
-                                       s->border_color[3]);
-
-                        pandecode_indent--;
-                        pandecode_log("};\n");
-                }
-        }
-}
-
-static void
-pandecode_vertex_tiler_postfix_pre(
-                const struct mali_vertex_tiler_postfix *p,
-                int job_no, enum mali_job_type job_type,
-                char *suffix, bool is_bifrost, unsigned gpu_id)
-{
-        struct pandecode_mapped_memory *attr_mem;
-
-        /* On Bifrost, since the tiler heap (for tiler jobs) and the scratchpad
-         * are the only things actually needed from the FBD, vertex/tiler jobs
-         * no longer reference the FBD -- instead, this field points to some
-         * info about the scratchpad.
-         */
-
-        struct pandecode_fbd fbd_info = {
-                /* Default for Bifrost */
-                .rt_count = 1
-        };
-
-        if (is_bifrost) {
-                pandecode_log_cont("\t/* %X %/\n", p->shared_memory & 1);
-                pandecode_compute_fbd(p->shared_memory & ~1, job_no);
-        } else if (p->shared_memory & MALI_MFBD)
-                fbd_info = pandecode_mfbd_bfr((u64) ((uintptr_t) p->shared_memory) & FBD_MASK, job_no, false, job_type == JOB_TYPE_COMPUTE, false);
-        else if (job_type == JOB_TYPE_COMPUTE)
-                pandecode_compute_fbd((u64) (uintptr_t) p->shared_memory, job_no);
-        else
-                fbd_info = pandecode_sfbd((u64) (uintptr_t) p->shared_memory, job_no, false, gpu_id);
-
-        int varying_count = 0, attribute_count = 0, uniform_count = 0, uniform_buffer_count = 0;
-        int texture_count = 0, sampler_count = 0;
-
-        if (p->shader) {
-                struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(p->shader);
-                struct mali_shader_meta *PANDECODE_PTR_VAR(s, smem, p->shader);
-
-                /* Disassemble ahead-of-time to get stats. Initialize with
-                 * stats for the missing-shader case so we get validation
-                 * there, too */
-
-                struct midgard_disasm_stats info = {
-                        .texture_count = 0,
-                        .sampler_count = 0,
-                        .attribute_count = 0,
-                        .varying_count = 0,
-                        .work_count = 1,
-
-                        .uniform_count = -128,
-                        .uniform_buffer_count = 0
-                };
-
-                if (s->shader & ~0xF)
-                        info = pandecode_shader_disassemble(s->shader & ~0xF, job_no, job_type, is_bifrost, gpu_id);
-
-                pandecode_log("struct mali_shader_meta shader_meta_%"PRIx64"_%d%s = {\n", p->shader, job_no, suffix);
-                pandecode_indent++;
-
-                /* Save for dumps */
-                attribute_count = s->attribute_count;
-                varying_count = s->varying_count;
-                texture_count = s->texture_count;
-                sampler_count = s->sampler_count;
-
-                if (is_bifrost) {
-                        uniform_count = s->bifrost2.uniform_count;
-                        uniform_buffer_count = s->bifrost1.uniform_buffer_count;
-                } else {
-                        uniform_count = s->midgard1.uniform_count;
-                        uniform_buffer_count = s->midgard1.uniform_buffer_count;
-                }
-
-                pandecode_shader_address("shader", s->shader);
-
-                pandecode_shader_prop("texture_count", s->texture_count, info.texture_count, false);
-                pandecode_shader_prop("sampler_count", s->sampler_count, info.sampler_count, false);
-                pandecode_shader_prop("attribute_count", s->attribute_count, info.attribute_count, false);
-                pandecode_shader_prop("varying_count", s->varying_count, info.varying_count, false);
-                pandecode_shader_prop("uniform_buffer_count",
-                                uniform_buffer_count,
-                                info.uniform_buffer_count, true);
-
-                if (!is_bifrost) {
-                        pandecode_shader_prop("uniform_count",
-                                        uniform_count,
-                                        info.uniform_count, false);
-
-                        pandecode_shader_prop("work_count",
-                                        s->midgard1.work_count, info.work_count, false);
-                }
-
-                if (is_bifrost) {
-                        pandecode_log("bifrost1.unk1 = ");
-                        pandecode_log_decoded_flags(shader_bifrost_info, s->bifrost1.unk1);
-                        pandecode_log_cont(",\n");
-                } else {
-                        bool helpers = s->midgard1.flags_lo & MALI_HELPER_INVOCATIONS;
-
-                        if (helpers != info.helper_invocations) {
-                                pandecode_msg("XXX: expected helpers %u but got %u\n",
-                                                info.helper_invocations, helpers);
-                        }
-
-                        pandecode_log(".midgard1.flags_lo = ");
-                        pandecode_log_decoded_flags(shader_midgard1_flag_lo_info,
-                                                    s->midgard1.flags_lo & ~MALI_HELPER_INVOCATIONS);
-                        pandecode_log_cont(",\n");
-
-                        pandecode_log(".midgard1.flags_hi = ");
-                        pandecode_log_decoded_flags(shader_midgard1_flag_hi_info, s->midgard1.flags_hi);
-                        pandecode_log_cont(",\n");
-                }
-
-                if (s->depth_units || s->depth_factor) {
-                        pandecode_prop("depth_factor = %f", s->depth_factor);
-                        pandecode_prop("depth_units = %f", s->depth_units);
-                }
-
-                if (s->coverage_mask)
-                        pandecode_prop("coverage_mask = 0x%X", s->coverage_mask);
-
-                if (s->unknown2_2)
-                        pandecode_prop(".unknown2_2 = %X", s->unknown2_2);
-
-                if (s->unknown2_3 || s->unknown2_4) {
-                        pandecode_log(".unknown2_3 = ");
-
-                        int unknown2_3 = s->unknown2_3;
-                        int unknown2_4 = s->unknown2_4;
-
-                        /* We're not quite sure what these flags mean without the depth test, if anything */
-
-                        if (unknown2_3 & (MALI_DEPTH_WRITEMASK | MALI_DEPTH_FUNC_MASK)) {
-                                const char *func = pandecode_func(MALI_GET_DEPTH_FUNC(unknown2_3));
-                                unknown2_3 &= ~MALI_DEPTH_FUNC_MASK;
-
-                                pandecode_log_cont("MALI_DEPTH_FUNC(%s) | ", func);
-                        }
-
-                        pandecode_log_decoded_flags(u3_flag_info, unknown2_3);
-                        pandecode_log_cont(",\n");
-
-                        pandecode_log(".unknown2_4 = ");
-                        pandecode_log_decoded_flags(u4_flag_info, unknown2_4);
-                        pandecode_log_cont(",\n");
-                }
-
-                if (s->stencil_mask_front || s->stencil_mask_back) {
-                        pandecode_prop("stencil_mask_front = 0x%02X", s->stencil_mask_front);
-                        pandecode_prop("stencil_mask_back = 0x%02X", s->stencil_mask_back);
-                }
-
-                pandecode_stencil("front", &s->stencil_front);
-                pandecode_stencil("back", &s->stencil_back);
-
-                if (is_bifrost) {
-                        pandecode_log(".bifrost2 = {\n");
-                        pandecode_indent++;
-
-                        pandecode_prop("unk3 = 0x%" PRIx32, s->bifrost2.unk3);
-                        pandecode_prop("preload_regs = 0x%" PRIx32, s->bifrost2.preload_regs);
-                        pandecode_prop("uniform_count = %" PRId32, s->bifrost2.uniform_count);
-                        pandecode_prop("unk4 = 0x%" PRIx32, s->bifrost2.unk4);
-
-                        pandecode_indent--;
-                        pandecode_log("},\n");
-                } else if (s->midgard2.unknown2_7) {
-                        pandecode_log(".midgard2 = {\n");
-                        pandecode_indent++;
-
-                        pandecode_prop("unknown2_7 = 0x%" PRIx32, s->midgard2.unknown2_7);
-                        pandecode_indent--;
-                        pandecode_log("},\n");
-                }
-
-                if (s->padding) {
-                        pandecode_msg("XXX: shader padding tripped\n");
-                        pandecode_prop("padding = 0x%" PRIx32, s->padding);
-                }
-
-                if (!is_bifrost) {
-                        /* TODO: Blend shaders routing/disasm */
-                        union midgard_blend blend = s->blend;
-                        mali_ptr shader = pandecode_midgard_blend(&blend, s->unknown2_3 & MALI_HAS_BLEND_SHADER);
-                        if (shader & ~0xF)
-                                pandecode_blend_shader_disassemble(shader, job_no, job_type, false, gpu_id);
-                } else {
-                        pandecode_msg("mdg_blend = %" PRIx64 "\n", s->blend.shader);
-                }
-
-                pandecode_indent--;
-                pandecode_log("};\n");
-
-                /* MRT blend fields are used whenever MFBD is used, with
-                 * per-RT descriptors */
-
-                if (job_type == JOB_TYPE_TILER && (is_bifrost || p->shared_memory & MALI_MFBD)) {
-                        void* blend_base = (void *) (s + 1);
-
-                        for (unsigned i = 0; i < fbd_info.rt_count; i++) {
-                                mali_ptr shader = 0;
-
-                                if (is_bifrost)
-                                        shader = pandecode_bifrost_blend(blend_base, job_no, i);
-                                else
-                                        shader = pandecode_midgard_blend_mrt(blend_base, job_no, i);
-
-                                if (shader & ~0xF)
-                                        pandecode_blend_shader_disassemble(shader, job_no, job_type, false, gpu_id);
-
-                        }
-                }
-        } else
-                pandecode_msg("XXX: missing shader descriptor\n");
-
-        if (p->viewport) {
-                struct pandecode_mapped_memory *fmem = pandecode_find_mapped_gpu_mem_containing(p->viewport);
-                struct mali_viewport *PANDECODE_PTR_VAR(f, fmem, p->viewport);
-
-                pandecode_log("struct mali_viewport viewport_%"PRIx64"_%d%s = {\n", p->viewport, job_no, suffix);
-                pandecode_indent++;
-
-                pandecode_prop("clip_minx = %f", f->clip_minx);
-                pandecode_prop("clip_miny = %f", f->clip_miny);
-                pandecode_prop("clip_minz = %f", f->clip_minz);
-                pandecode_prop("clip_maxx = %f", f->clip_maxx);
-                pandecode_prop("clip_maxy = %f", f->clip_maxy);
-                pandecode_prop("clip_maxz = %f", f->clip_maxz);
-
-                /* Only the higher coordinates are MALI_POSITIVE scaled */
-
-                pandecode_prop("viewport0 = { %d, %d }",
-                               f->viewport0[0], f->viewport0[1]);
-
-                pandecode_prop("viewport1 = { MALI_POSITIVE(%d), MALI_POSITIVE(%d) }",
-                               f->viewport1[0] + 1, f->viewport1[1] + 1);
-
-                pandecode_indent--;
-                pandecode_log("};\n");
-        }
-
-        unsigned max_attr_index = 0;
-
-        if (p->attribute_meta)
-                max_attr_index = pandecode_attribute_meta(job_no, attribute_count, p, false, suffix);
-
-        if (p->attributes) {
-                attr_mem = pandecode_find_mapped_gpu_mem_containing(p->attributes);
-                pandecode_attributes(attr_mem, p->attributes, job_no, suffix, max_attr_index, false, job_type);
-        }
-
-        /* Varyings are encoded like attributes but not actually sent; we just
-         * pass a zero buffer with the right stride/size set, (or whatever)
-         * since the GPU will write to it itself */
-
-        if (p->varying_meta) {
-                varying_count = pandecode_attribute_meta(job_no, varying_count, p, true, suffix);
-        }
-
-        if (p->varyings) {
-                attr_mem = pandecode_find_mapped_gpu_mem_containing(p->varyings);
-
-                /* Number of descriptors depends on whether there are
-                 * non-internal varyings */
-
-                pandecode_attributes(attr_mem, p->varyings, job_no, suffix, varying_count, true, job_type);
-        }
-
-        if (p->uniform_buffers) {
-                if (uniform_buffer_count)
-                        pandecode_uniform_buffers(p->uniform_buffers, uniform_buffer_count, job_no);
-                else
-                        pandecode_msg("warn: UBOs specified but not referenced\n");
-        } else if (uniform_buffer_count)
-                pandecode_msg("XXX: UBOs referenced but not specified\n");
-
-        /* We don't want to actually dump uniforms, but we do need to validate
-         * that the counts we were given are sane */
-
-        if (p->uniforms) {
-                if (uniform_count)
-                        pandecode_uniforms(p->uniforms, uniform_count);
-                else
-                        pandecode_msg("warn: Uniforms specified but not referenced\n");
-        } else if (uniform_count)
-                pandecode_msg("XXX: Uniforms referenced but not specified\n");
-
-        if (p->textures)
-                pandecode_textures(p->textures, texture_count, job_no, is_bifrost);
-
-        if (p->sampler_descriptor)
-                pandecode_samplers(p->sampler_descriptor, sampler_count, job_no, is_bifrost);
-}
-
-static void
-pandecode_gl_enables(uint32_t gl_enables, int job_type)
-{
-        pandecode_log(".gl_enables = ");
-
-        pandecode_log_decoded_flags(gl_enable_flag_info, gl_enables);
-
-        pandecode_log_cont(",\n");
-}
-
-static void
-pandecode_vertex_tiler_postfix(const struct mali_vertex_tiler_postfix *p, int job_no, bool is_bifrost)
-{
-        if (p->shader & 0xF)
-                pandecode_msg("warn: shader tagged %X\n", (unsigned) (p->shader & 0xF));
-
-        pandecode_log(".postfix = {\n");
-        pandecode_indent++;
-
-        pandecode_gl_enables(p->gl_enables, JOB_TYPE_TILER);
-        pandecode_prop("instance_shift = 0x%x", p->instance_shift);
-        pandecode_prop("instance_odd = 0x%x", p->instance_odd);
-
-        if (p->zero4) {
-                pandecode_msg("XXX: vertex only zero tripped");
-                pandecode_prop("zero4 = 0x%" PRIx32, p->zero4);
-        }
-
-        pandecode_prop("offset_start = 0x%x", p->offset_start);
-
-        if (p->zero5) {
-                pandecode_msg("XXX: vertex only zero tripped");
-                pandecode_prop("zero5 = 0x%" PRIx64, p->zero5);
-        }
-
-        MEMORY_PROP(p, position_varying);
-        MEMORY_PROP(p, occlusion_counter);
-
-        pandecode_indent--;
-        pandecode_log("},\n");
-}
-
-static void
-pandecode_tiler_heap_meta(mali_ptr gpu_va, int job_no)
-{
-        struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
-        const struct bifrost_tiler_heap_meta *PANDECODE_PTR_VAR(h, mem, gpu_va);
-
-        pandecode_log("struct bifrost_tiler_heap_meta tiler_heap_meta_%"PRIx64"_%d = {\n", gpu_va, job_no);
-        pandecode_indent++;
-
-        if (h->zero) {
-                pandecode_msg("XXX: tiler heap zero tripped\n");
-                pandecode_prop("zero = 0x%x", h->zero);
-        }
-
-        pandecode_prop("heap_size = 0x%x", h->heap_size);
-        MEMORY_PROP(h, tiler_heap_start);
-        MEMORY_PROP(h, tiler_heap_free);
-
-        /* this might point to the beginning of another buffer, when it's
-         * really the end of the tiler heap buffer, so we have to be careful
-         * here. but for zero length, we need the same pointer.
-         */
-
-        if (h->tiler_heap_end == h->tiler_heap_start) {
-                MEMORY_PROP(h, tiler_heap_start);
-        } else {
-                char *a = pointer_as_memory_reference(h->tiler_heap_end - 1);
-                pandecode_prop("tiler_heap_end = %s + 1", a);
-                free(a);
-        }
-
-        for (int i = 0; i < 10; i++) {
-                if (h->zeros[i] != 0) {
-                        pandecode_msg("XXX: tiler heap zero %d tripped, value %x\n",
-                                      i, h->zeros[i]);
-                }
-        }
-
-        if (h->unk1 != 0x1) {
-                pandecode_msg("XXX: tiler heap unk1 tripped\n");
-                pandecode_prop("unk1 = 0x%x", h->unk1);
-        }
-
-        if (h->unk7e007e != 0x7e007e) {
-                pandecode_msg("XXX: tiler heap unk7e007e tripped\n");
-                pandecode_prop("unk7e007e = 0x%x", h->unk7e007e);
-        }
-
-        pandecode_indent--;
-        pandecode_log("};\n");
-}
-
-static void
-pandecode_tiler_meta(mali_ptr gpu_va, int job_no)
-{
-        struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
-        const struct bifrost_tiler_meta *PANDECODE_PTR_VAR(t, mem, gpu_va);
-
-        pandecode_tiler_heap_meta(t->tiler_heap_meta, job_no);
-
-        pandecode_log("struct bifrost_tiler_meta tiler_meta_%"PRIx64"_%d = {\n", gpu_va, job_no);
-        pandecode_indent++;
-
-        pandecode_prop("tiler_heap_next_start = 0x%" PRIx32, t->tiler_heap_next_start);
-        pandecode_prop("used_hierarchy_mask = 0x%" PRIx32, t->used_hierarchy_mask);
-
-        if (t->hierarchy_mask != 0xa &&
-            t->hierarchy_mask != 0x14 &&
-            t->hierarchy_mask != 0x28 &&
-            t->hierarchy_mask != 0x50 &&
-            t->hierarchy_mask != 0xa0)
-                pandecode_prop("XXX: Unexpected hierarchy_mask (not 0xa, 0x14, 0x28, 0x50 or 0xa0)!");
-
-        pandecode_prop("hierarchy_mask = 0x%" PRIx16, t->hierarchy_mask);
-
-        pandecode_prop("flags = 0x%" PRIx16, t->flags);
-
-        pandecode_prop("width = MALI_POSITIVE(%d)", t->width + 1);
-        pandecode_prop("height = MALI_POSITIVE(%d)", t->height + 1);
-
-        if (t->zero0) {
-                pandecode_msg("XXX: tiler meta zero tripped\n");
-                pandecode_prop("zero0 = 0x%" PRIx64, t->zero0);
-        }
-
-        for (int i = 0; i < 12; i++) {
-                if (t->zeros[i] != 0) {
-                        pandecode_msg("XXX: tiler heap zero %d tripped, value %" PRIx64 "\n",
-                                      i, t->zeros[i]);
-                }
-        }
-
-        pandecode_indent--;
-        pandecode_log("};\n");
-}
-
-static void
-pandecode_primitive_size(union midgard_primitive_size u, bool constant)
-{
-        if (u.pointer == 0x0)
-                return;
-
-        pandecode_log(".primitive_size = {\n");
-        pandecode_indent++;
-
-        if (constant) {
-                pandecode_prop("constant = %f", u.constant);
-        } else {
-                MEMORY_PROP((&u), pointer);
-        }
-
-        pandecode_indent--;
-        pandecode_log("},\n");
-}
-
-static void
-pandecode_tiler_only_bfr(const struct bifrost_tiler_only *t, int job_no)
-{
-        pandecode_log_cont("{\n");
-        pandecode_indent++;
-
-        /* TODO: gl_PointSize on Bifrost */
-        pandecode_primitive_size(t->primitive_size, true);
-
-        if (t->zero1 || t->zero2 || t->zero3 || t->zero4 || t->zero5
-            || t->zero6) {
-                pandecode_msg("XXX: tiler only zero tripped\n");
-                pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
-                pandecode_prop("zero2 = 0x%" PRIx64, t->zero2);
-                pandecode_prop("zero3 = 0x%" PRIx64, t->zero3);
-                pandecode_prop("zero4 = 0x%" PRIx64, t->zero4);
-                pandecode_prop("zero5 = 0x%" PRIx64, t->zero5);
-                pandecode_prop("zero6 = 0x%" PRIx64, t->zero6);
-        }
-
-        pandecode_indent--;
-        pandecode_log("},\n");
-}
-
-static int
-pandecode_vertex_job_bfr(const struct mali_job_descriptor_header *h,
-                                const struct pandecode_mapped_memory *mem,
-                                mali_ptr payload, int job_no, unsigned gpu_id)
-{
-        struct bifrost_payload_vertex *PANDECODE_PTR_VAR(v, mem, payload);
-
-        pandecode_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", true, gpu_id);
-
-        pandecode_log("struct bifrost_payload_vertex payload_%"PRIx64"_%d = {\n", payload, job_no);
-        pandecode_indent++;
-
-        pandecode_vertex_tiler_prefix(&v->prefix, job_no, false);
-        pandecode_vertex_tiler_postfix(&v->postfix, job_no, true);
-
-        pandecode_indent--;
-        pandecode_log("};\n");
-
-        return sizeof(*v);
-}
-
-static int
-pandecode_tiler_job_bfr(const struct mali_job_descriptor_header *h,
-                               const struct pandecode_mapped_memory *mem,
-                               mali_ptr payload, int job_no, unsigned gpu_id)
-{
-        struct bifrost_payload_tiler *PANDECODE_PTR_VAR(t, mem, payload);
-
-        pandecode_vertex_tiler_postfix_pre(&t->postfix, job_no, h->job_type, "", true, gpu_id);
-        pandecode_tiler_meta(t->tiler.tiler_meta, job_no);
-
-        pandecode_log("struct bifrost_payload_tiler payload_%"PRIx64"_%d = {\n", payload, job_no);
-        pandecode_indent++;
-
-        pandecode_vertex_tiler_prefix(&t->prefix, job_no, false);
-
-        pandecode_log(".tiler = ");
-        pandecode_tiler_only_bfr(&t->tiler, job_no);
-
-        pandecode_vertex_tiler_postfix(&t->postfix, job_no, true);
-
-        pandecode_indent--;
-        pandecode_log("};\n");
-
-        return sizeof(*t);
-}
-
-static int
-pandecode_vertex_or_tiler_job_mdg(const struct mali_job_descriptor_header *h,
-                const struct pandecode_mapped_memory *mem,
-                mali_ptr payload, int job_no, unsigned gpu_id)
-{
-        struct midgard_payload_vertex_tiler *PANDECODE_PTR_VAR(v, mem, payload);
-        bool is_graphics = (h->job_type == JOB_TYPE_VERTEX) || (h->job_type == JOB_TYPE_TILER);
-
-        pandecode_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", false, gpu_id);
-
-        pandecode_log("struct midgard_payload_vertex_tiler payload_%d = {\n", job_no);
-        pandecode_indent++;
-
-        pandecode_vertex_tiler_prefix(&v->prefix, job_no, is_graphics);
-        pandecode_vertex_tiler_postfix(&v->postfix, job_no, false);
-
-        bool has_primitive_pointer = v->prefix.unknown_draw & MALI_DRAW_VARYING_SIZE;
-        pandecode_primitive_size(v->primitive_size, !has_primitive_pointer);
-
-        pandecode_indent--;
-        pandecode_log("};\n");
-
-        return sizeof(*v);
-}
-
-static int
-pandecode_fragment_job(const struct pandecode_mapped_memory *mem,
-                              mali_ptr payload, int job_no,
-                              bool is_bifrost, unsigned gpu_id)
-{
-        const struct mali_payload_fragment *PANDECODE_PTR_VAR(s, mem, payload);
-
-        bool is_mfbd = s->framebuffer & MALI_MFBD;
-
-        if (!is_mfbd && is_bifrost)
-                pandecode_msg("XXX: Bifrost fragment must use MFBD\n");
-
-        struct pandecode_fbd info;
-
-        if (is_mfbd)
-                info = pandecode_mfbd_bfr(s->framebuffer & FBD_MASK, job_no, true, false, is_bifrost);
-        else
-                info = pandecode_sfbd(s->framebuffer & FBD_MASK, job_no, true, gpu_id);
-
-        /* Compute the tag for the tagged pointer. This contains the type of
-         * FBD (MFBD/SFBD), and in the case of an MFBD, information about which
-         * additional structures follow the MFBD header (an extra payload or
-         * not, as well as a count of render targets) */
-
-        unsigned expected_tag = is_mfbd ? MALI_MFBD : 0;
-
-        if (is_mfbd) {
-                if (info.has_extra)
-                        expected_tag |= MALI_MFBD_TAG_EXTRA;
-
-                expected_tag |= (MALI_POSITIVE(info.rt_count) << 2);
-        }
-
-        if ((s->min_tile_coord | s->max_tile_coord) & ~(MALI_X_COORD_MASK | MALI_Y_COORD_MASK)) {
-                pandecode_msg("XXX: unexpected tile coordinate bits\n");
-                pandecode_prop("min_tile_coord = 0x%X\n", s->min_tile_coord);
-                pandecode_prop("max_tile_coord = 0x%X\n", s->max_tile_coord);
-        }
-
-        /* Extract tile coordinates */
-
-        unsigned min_x = MALI_TILE_COORD_X(s->min_tile_coord) << MALI_TILE_SHIFT;
-        unsigned min_y = MALI_TILE_COORD_Y(s->min_tile_coord) << MALI_TILE_SHIFT;
-
-        unsigned max_x = (MALI_TILE_COORD_X(s->max_tile_coord) + 1) << MALI_TILE_SHIFT;
-        unsigned max_y = (MALI_TILE_COORD_Y(s->max_tile_coord) + 1) << MALI_TILE_SHIFT;
-
-        /* For the max, we also want the floored (rather than ceiled) version for checking */
-
-        unsigned max_x_f = (MALI_TILE_COORD_X(s->max_tile_coord)) << MALI_TILE_SHIFT;
-        unsigned max_y_f = (MALI_TILE_COORD_Y(s->max_tile_coord)) << MALI_TILE_SHIFT;
-
-        /* Validate the coordinates are well-ordered */
-
-        if (min_x == max_x)
-                pandecode_msg("XXX: empty X coordinates (%u = %u)\n", min_x, max_x);
-        else if (min_x > max_x)
-                pandecode_msg("XXX: misordered X coordinates (%u > %u)\n", min_x, max_x);
-
-        if (min_y == max_y)
-                pandecode_msg("XXX: empty X coordinates (%u = %u)\n", min_x, max_x);
-        else if (min_y > max_y)
-                pandecode_msg("XXX: misordered X coordinates (%u > %u)\n", min_x, max_x);
-
-        /* Validate the coordinates fit inside the framebuffer. We use floor,
-         * rather than ceil, for the max coordinates, since the tile
-         * coordinates for something like an 800x600 framebuffer will actually
-         * resolve to 800x608, which would otherwise trigger a Y-overflow */
-
-        if ((min_x > info.width) || (max_x_f > info.width))
-                pandecode_msg("XXX: tile coordinates overflow in X direction\n");
-
-        if ((min_y > info.height) || (max_y_f > info.height))
-                pandecode_msg("XXX: tile coordinates overflow in Y direction\n");
-
-        /* After validation, we print */
-
-        pandecode_log("fragment (%u, %u) ... (%u, %u)\n\n", min_x, min_y, max_x, max_y);
-
-        /* The FBD is a tagged pointer */
-
-        unsigned tag = (s->framebuffer & ~FBD_MASK);
-
-        if (tag != expected_tag)
-                pandecode_msg("XXX: expected FBD tag %X but got %X\n", expected_tag, tag);
-
-        return sizeof(*s);
-}
-
-/* Entrypoint to start tracing. jc_gpu_va is the GPU address for the first job
- * in the chain; later jobs are found by walking the chain. Bifrost is, well,
- * if it's bifrost or not. GPU ID is the more finegrained ID (at some point, we
- * might wish to combine this with the bifrost parameter) because some details
- * are model-specific even within a particular architecture. Minimal traces
- * *only* examine the job descriptors, skipping printing entirely if there is
- * no faults, and only descends into the payload if there are faults. This is
- * useful for looking for faults without the overhead of invasive traces. */
-
-void
-pandecode_jc(mali_ptr jc_gpu_va, bool bifrost, unsigned gpu_id, bool minimal)
-{
-        pandecode_dump_file_open();
-
-        struct mali_job_descriptor_header *h;
-        unsigned job_descriptor_number = 0;
-
-        do {
-                struct pandecode_mapped_memory *mem =
-                        pandecode_find_mapped_gpu_mem_containing(jc_gpu_va);
-
-                void *payload;
-
-                h = PANDECODE_PTR(mem, jc_gpu_va, struct mali_job_descriptor_header);
-
-                /* On Midgard, for 32-bit jobs except for fragment jobs, the
-                 * high 32-bits of the 64-bit pointer are reused to store
-                 * something else.
-                 */
-                int offset = h->job_descriptor_size == MALI_JOB_32 &&
-                             h->job_type != JOB_TYPE_FRAGMENT ? 4 : 0;
-                mali_ptr payload_ptr = jc_gpu_va + sizeof(*h) - offset;
-
-                payload = pandecode_fetch_gpu_mem(mem, payload_ptr, 256);
-
-                int job_no = job_descriptor_number++;
-
-                /* If the job is good to go, skip it in minimal mode */
-                if (minimal && (h->exception_status == 0x0 || h->exception_status == 0x1))
-                        continue;
-
-                pandecode_log("struct mali_job_descriptor_header job_%"PRIx64"_%d = {\n", jc_gpu_va, job_no);
-                pandecode_indent++;
-
-                pandecode_prop("job_type = %s", pandecode_job_type(h->job_type));
-
-                if (h->job_descriptor_size)
-                        pandecode_prop("job_descriptor_size = %d", h->job_descriptor_size);
-
-                if (h->exception_status && h->exception_status != 0x1)
-                        pandecode_prop("exception_status = %x (source ID: 0x%x access: %s exception: 0x%x)",
-                                       h->exception_status,
-                                       (h->exception_status >> 16) & 0xFFFF,
-                                       pandecode_exception_access((h->exception_status >> 8) & 0x3),
-                                       h->exception_status  & 0xFF);
-
-                if (h->first_incomplete_task)
-                        pandecode_prop("first_incomplete_task = %d", h->first_incomplete_task);
-
-                if (h->fault_pointer)
-                        pandecode_prop("fault_pointer = 0x%" PRIx64, h->fault_pointer);
-
-                if (h->job_barrier)
-                        pandecode_prop("job_barrier = %d", h->job_barrier);
-
-                pandecode_prop("job_index = %d", h->job_index);
-
-                if (h->unknown_flags)
-                        pandecode_prop("unknown_flags = %d", h->unknown_flags);
-
-                if (h->job_dependency_index_1)
-                        pandecode_prop("job_dependency_index_1 = %d", h->job_dependency_index_1);
-
-                if (h->job_dependency_index_2)
-                        pandecode_prop("job_dependency_index_2 = %d", h->job_dependency_index_2);
-
-                pandecode_indent--;
-                pandecode_log("};\n");
-
-                switch (h->job_type) {
-                case JOB_TYPE_WRITE_VALUE: {
-                        struct mali_payload_write_value *s = payload;
-                        pandecode_log("struct mali_payload_write_value payload_%"PRIx64"_%d = {\n", payload_ptr, job_no);
-                        pandecode_indent++;
-                        MEMORY_PROP(s, address);
-
-                        if (s->value_descriptor != MALI_WRITE_VALUE_ZERO) {
-                                pandecode_msg("XXX: unknown value descriptor\n");
-                                pandecode_prop("value_descriptor = 0x%" PRIX32, s->value_descriptor);
-                        }
-
-                        if (s->reserved) {
-                                pandecode_msg("XXX: set value tripped\n");
-                                pandecode_prop("reserved = 0x%" PRIX32, s->reserved);
-                        }
-
-                        pandecode_prop("immediate = 0x%" PRIX64, s->immediate);
-                        pandecode_indent--;
-                        pandecode_log("};\n");
-
-                        break;
-                }
-
-                case JOB_TYPE_TILER:
-                case JOB_TYPE_VERTEX:
-                case JOB_TYPE_COMPUTE:
-                        if (bifrost) {
-                                if (h->job_type == JOB_TYPE_TILER)
-                                        pandecode_tiler_job_bfr(h, mem, payload_ptr, job_no, gpu_id);
-                                else
-                                        pandecode_vertex_job_bfr(h, mem, payload_ptr, job_no, gpu_id);
-                        } else
-                                pandecode_vertex_or_tiler_job_mdg(h, mem, payload_ptr, job_no, gpu_id);
-
-                        break;
-
-                case JOB_TYPE_FRAGMENT:
-                        pandecode_fragment_job(mem, payload_ptr, job_no, bifrost, gpu_id);
-                        break;
-
-                default:
-                        break;
-                }
-        } while ((jc_gpu_va = h->next_job));
-
-        pandecode_map_read_write();
-}
diff --git a/src/panfrost/pandecode/decode.h b/src/panfrost/pandecode/decode.h
deleted file mode 100644 (file)
index 61345a2..0000000
+++ /dev/null
@@ -1,85 +0,0 @@
-/*
- * Copyright (C) 2017-2019 Lyude Paul
- * Copyright (C) 2017-2019 Alyssa Rosenzweig
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice (including the next
- * paragraph) shall be included in all copies or substantial portions of the
- * Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- *
- */
-
-#ifndef __PAN_DECODE_H__
-#define __PAN_DECODE_H__
-
-#include "public.h"
-#include "util/list.h"
-
-extern FILE *pandecode_dump_stream;
-
-void pandecode_dump_file_open(void);
-
-struct pandecode_mapped_memory {
-        size_t length;
-        void *addr;
-        uint64_t gpu_va;
-        bool ro;
-        char name[32];
-};
-
-char *pointer_as_memory_reference(uint64_t ptr);
-
-struct pandecode_mapped_memory *pandecode_find_mapped_gpu_mem_containing(uint64_t addr);
-
-void pandecode_map_read_write(void);
-
-static inline void *
-__pandecode_fetch_gpu_mem(const struct pandecode_mapped_memory *mem,
-                          uint64_t gpu_va, size_t size,
-                          int line, const char *filename)
-{
-        if (!mem)
-                mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
-
-        if (!mem) {
-                fprintf(stderr, "Access to unknown memory %" PRIx64 " in %s:%d",
-                        gpu_va, filename, line);
-                assert(0);
-        }
-
-        assert(mem);
-        assert(size + (gpu_va - mem->gpu_va) <= mem->length);
-
-        return mem->addr + gpu_va - mem->gpu_va;
-}
-
-#define pandecode_fetch_gpu_mem(mem, gpu_va, size) \
-       __pandecode_fetch_gpu_mem(mem, gpu_va, size, __LINE__, __FILE__)
-
-/* Returns a validated pointer to mapped GPU memory with the given pointer type,
- * size automatically determined from the pointer type
- */
-#define PANDECODE_PTR(mem, gpu_va, type) \
-       ((type*)(__pandecode_fetch_gpu_mem(mem, gpu_va, sizeof(type), \
-                                        __LINE__, __FILE__)))
-
-/* Usage: <variable type> PANDECODE_PTR_VAR(name, mem, gpu_va) */
-#define PANDECODE_PTR_VAR(name, mem, gpu_va) \
-       name = __pandecode_fetch_gpu_mem(mem, gpu_va, sizeof(*name), \
-                                      __LINE__, __FILE__)
-
-#endif /* __MMAP_TRACE_H__ */
diff --git a/src/panfrost/pandecode/meson.build b/src/panfrost/pandecode/meson.build
deleted file mode 100644 (file)
index e4bc6aa..0000000
+++ /dev/null
@@ -1,34 +0,0 @@
-# Copyright © 2018 Rob Clark
-# Copyright © 2019 Collabora
-
-# Permission is hereby granted, free of charge, to any person obtaining a copy
-# of this software and associated documentation files (the "Software"), to deal
-# in the Software without restriction, including without limitation the rights
-# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-# copies of the Software, and to permit persons to whom the Software is
-# furnished to do so, subject to the following conditions:
-
-# The above copyright notice and this permission notice shall be included in
-# all copies or substantial portions of the Software.
-
-# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-# SOFTWARE.
-
-libpanfrost_decode_files = files(
-  'common.c',
-  'decode.c',
-)
-
-libpanfrost_decode = static_library(
-  'panfrost_decode',
-  [libpanfrost_decode_files],
-  include_directories : [inc_include, inc_src, inc_mapi, inc_mesa, inc_gallium, inc_gallium_aux, inc_panfrost],
-  c_args : [no_override_init_args],
-  gnu_symbol_visibility : 'hidden',
-  build_by_default : false,
-)
diff --git a/src/panfrost/pandecode/public.h b/src/panfrost/pandecode/public.h
deleted file mode 100644 (file)
index abb18c8..0000000
+++ /dev/null
@@ -1,56 +0,0 @@
-
-/*
- * Copyright (C) 2017-2019 Lyude Paul
- * Copyright (C) 2017-2019 Alyssa Rosenzweig
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice (including the next
- * paragraph) shall be included in all copies or substantial portions of the
- * Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- *
- */
-
-#ifndef __PAN_DECODE_PUBLIC_H__
-#define __PAN_DECODE_PUBLIC_H__
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <stddef.h>
-#include <stdbool.h>
-#include <inttypes.h>
-
-/* Public entrypoints for the tracing infrastructure. This API should be kept
- * more or less stable. Don't feel bad if you have to change it; just feel
- * slightly guilty about creating more work for me later. -Alyssa <3
- *
- * I'm joking. Mostly. panwrap (out-of-tree) includes this, so update that if
- * you need to change something here. panwrap is open-source but cannot be
- * included in-tree.
- */
-
-void pandecode_initialize(bool to_stderr);
-
-void pandecode_next_frame(void);
-
-void pandecode_close(void);
-
-void
-pandecode_inject_mmap(uint64_t gpu_va, void *cpu, unsigned sz, const char *name);
-
-void pandecode_jc(uint64_t jc_gpu_va, bool bifrost, unsigned gpu_id, bool minimal);
-
-#endif /* __MMAP_TRACE_H__ */