#include "pan_encoder.h"
-int pandecode_jc(mali_ptr jc_gpu_va, bool bifrost);
+static void pandecode_swizzle(unsigned swizzle, enum mali_format format);
#define MEMORY_PROP(obj, p) {\
if (obj->p) { \
if (total > bo->length) {
pandecode_msg("XXX: buffer overrun. "
- "Chunk of size %d at offset %d in buffer of size %d. "
- "Overrun by %d bytes. \n",
+ "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;
}
FLAG_INFO(HAS_MSAA),
FLAG_INFO(CAN_DISCARD),
FLAG_INFO(HAS_BLEND_SHADER),
- FLAG_INFO(DEPTH_TEST),
+ FLAG_INFO(DEPTH_WRITEMASK),
{}
};
};
#undef FLAG_INFO
-#define FLAG_INFO(flag) { MALI_FRAMEBUFFER_##flag, "MALI_FRAMEBUFFER_" #flag }
-static const struct pandecode_flag_info fb_fmt_flag_info[] = {
- FLAG_INFO(MSAA_A),
- FLAG_INFO(MSAA_B),
- FLAG_INFO(MSAA_8),
- {}
-};
-#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(MSAA),
#define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
static const struct pandecode_flag_info shader_midgard1_flag_info [] = {
FLAG_INFO(EARLY_Z),
- FLAG_INFO(HELPER_INVOCATIONS),
FLAG_INFO(READS_TILEBUFFER),
FLAG_INFO(READS_ZS),
{}
};
#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
+
extern char *replace_fragment;
extern char *replace_vertex;
#undef DEFINE_CASE
-#define DEFINE_CASE(name) case MALI_ATTR_ ## name: return "MALI_ATTR_" #name
static char *pandecode_attr_mode_short(enum mali_attr_mode mode)
{
switch(mode) {
}
}
-#undef DEFINE_CASE
+static const char *
+pandecode_special_varying(uint64_t v)
+{
+ switch(v) {
+ 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 varying %" PRIx64 "\n", v);
+ return "";
+ }
+}
#define DEFINE_CASE(name) case MALI_WRAP_## name: return "MALI_WRAP_" #name
static char *
}
#undef DEFINE_CASE
-#define DEFINE_CASE(name) case MALI_MFBD_BLOCK_## name: return "MALI_MFBD_BLOCK_" #name
+#define DEFINE_CASE(name) case MALI_BLOCK_## name: return "MALI_BLOCK_" #name
static char *
-pandecode_mfbd_block_format(enum mali_mfbd_block_format fmt)
+pandecode_block_format(enum mali_block_format fmt)
{
switch (fmt) {
DEFINE_CASE(TILED);
/* 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 %d (expected %d)\n",
+ pandecode_msg("XXX: heap size %u (expected %zu)\n",
heap_size, heap->length - heap_offset);
}
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)",
+ MALI_NEGATIVE(format.nr_channels));
+
+ 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 struct pandecode_fbd
pandecode_sfbd(uint64_t gpu_va, int job_no, bool is_fragment)
{
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;
+ 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_prop("unknown1 = 0x%" PRIx32, s->unknown1);
pandecode_prop("unknown2 = 0x%" PRIx32, s->unknown2);
- pandecode_log(".format = ");
- pandecode_log_decoded_flags(fb_fmt_flag_info, s->format);
- pandecode_log_cont(",\n");
+ pandecode_sfbd_format(s->format);
info.width = s->width + 1;
info.height = s->height + 1;
- info.rt_count = 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);
pandecode_log_decoded_flags(clear_flag_info, s->clear_flags);
pandecode_log_cont(",\n");
- if (s->depth_buffer | s->depth_buffer_enable) {
+ if (s->depth_buffer) {
MEMORY_PROP(s, depth_buffer);
- pandecode_prop("depth_buffer_enable = %s", DS_ENABLE(s->depth_buffer_enable));
+ pandecode_prop("depth_stride = %d", s->depth_stride);
}
- if (s->stencil_buffer | s->stencil_buffer_enable) {
+ if (s->stencil_buffer) {
MEMORY_PROP(s, stencil_buffer);
- pandecode_prop("stencil_buffer_enable = %s", DS_ENABLE(s->stencil_buffer_enable));
+ 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("zero1 = 0x%" PRIx64, s->zero1);
pandecode_prop("zero2 = 0x%" PRIx32, s->zero2);
pandecode_prop("zero4 = 0x%" PRIx32, s->zero4);
+ pandecode_prop("zero5 = 0x%" PRIx32, s->zero5);
printf(".zero3 = {");
pandecode_prop("unk2 = 0x%" PRIx32, format.unk2);
pandecode_prop("unk3 = 0x%" PRIx32, format.unk3);
- pandecode_prop("block = %s",
- pandecode_mfbd_block_format(format.block));
+ pandecode_prop("block = %s", pandecode_block_format(format.block));
/* TODO: Map formats so we can check swizzles and print nicely */
pandecode_log("swizzle");
pandecode_rt_format(rt->format);
- if (rt->format.block == MALI_MFBD_BLOCK_AFBC) {
+ if (rt->format.block == MALI_BLOCK_AFBC) {
pandecode_log(".afbc = {\n");
pandecode_indent++;
static void
pandecode_attributes(const struct pandecode_mapped_memory *mem,
mali_ptr addr, int job_no, char *suffix,
- int count, bool varying)
+ int count, bool varying, enum mali_job_type job_type)
{
char *prefix = varying ? "varying" : "attribute";
+ assert(addr);
- if (!addr) {
- pandecode_msg("no %s\n", prefix);
+ 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_VARYING_SPECIAL) {
+ /* Special records are always varyings */
+
+ if (!varying)
+ pandecode_msg("XXX: Special varying in attribute field\n");
+
+ if (job_type != JOB_TYPE_TILER)
+ pandecode_msg("XXX: Special varying in non-FS\n");
+
+ /* We're special, so all fields should be zero */
+ unsigned zero = attr[i].stride | attr[i].size;
+ zero |= attr[i].shift | attr[i].extra_flags;
+
+ if (zero)
+ pandecode_msg("XXX: Special varying has non-zero fields\n");
+ else {
+ /* Print the special varying name */
+ pandecode_log("varying_%d = %s;", i, pandecode_special_varying(attr[i].elements));
+ continue;
+ }
+ }
+
enum mali_attr_mode mode = attr[i].elements & 7;
if (mode == MALI_ATTR_UNUSED)
/* Print the stride and size */
pandecode_log_cont("<%u>[%u]", attr[i].stride, attr[i].size);
- /* Check: the size must be divisible by the stride */
- if (attr[i].size % attr[i].stride)
- pandecode_msg("XXX: size not divisible by stride\n");
-
- /* TODO: Sanity check the quotient itself -- it should equal
- * vertex count (or something computed from it for instanced)
- * which means we can check and elide */
+ /* 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;
snprintf(base, sizeof(base), "%s_meta", prefix);
struct mali_attr_meta *attr_meta;
- mali_ptr p = varying ? (v->varying_meta & ~0xF) : v->attribute_meta;
+ mali_ptr p = varying ? v->varying_meta : v->attribute_meta;
struct pandecode_mapped_memory *attr_mem = pandecode_find_mapped_gpu_mem_containing(p);
if (!canonical) {
pandecode_msg("XXX: non-canonical workgroups packing\n");
- pandecode_msg("expected: %X, %d, %d, %d, %d, %d\n",
+ pandecode_msg("expected: %X, %d, %d, %d, %d, %d, %d\n",
ref.invocation_count,
ref.size_y_shift,
ref.size_z_shift,
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 void
pandecode_scratchpad(uintptr_t pscratchpad, int job_no, char *suffix)
{
static struct midgard_disasm_stats
pandecode_shader_disassemble(mali_ptr shader_ptr, int shader_no, int type,
- bool is_bifrost)
+ 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);
stats.instruction_count = 0;
stats.bundle_count = 0;
stats.quadword_count = 0;
+ stats.helper_invocations = false;
} else {
- stats = disassemble_midgard(code, sz);
+ stats = disassemble_midgard(code, sz, gpu_id);
}
/* Print shader-db stats */
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_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)
+ char *suffix, bool is_bifrost, unsigned gpu_id)
{
- mali_ptr shader_meta_ptr = (u64) (uintptr_t) (p->_shader_upper << 4);
struct pandecode_mapped_memory *attr_mem;
/* On Bifrost, since the tiler heap (for tiler jobs) and the scratchpad
int varying_count = 0, attribute_count = 0, uniform_count = 0, uniform_buffer_count = 0;
int texture_count = 0, sampler_count = 0;
- if (shader_meta_ptr) {
- struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(shader_meta_ptr);
- struct mali_shader_meta *PANDECODE_PTR_VAR(s, smem, shader_meta_ptr);
+ 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
};
if (s->shader & ~0xF)
- info = pandecode_shader_disassemble(s->shader & ~0xF, job_no, job_type, is_bifrost);
+ 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", shader_meta_ptr, job_no, suffix);
+ pandecode_log("struct mali_shader_meta shader_meta_%"PRIx64"_%d%s = {\n", p->shader, job_no, suffix);
pandecode_indent++;
/* Save for dumps */
if (is_bifrost) {
pandecode_prop("bifrost1.unk1 = 0x%" PRIx32, s->bifrost1.unk1);
} else {
+ bool helpers = s->midgard1.flags & MALI_HELPER_INVOCATIONS;
+ s->midgard1.flags &= ~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 = ");
pandecode_log_decoded_flags(shader_midgard1_flag_info, s->midgard1.flags);
pandecode_log_cont(",\n");
/* We're not quite sure what these flags mean without the depth test, if anything */
- if (unknown2_3 & (MALI_DEPTH_TEST | MALI_DEPTH_FUNC_MASK)) {
+ 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;
if (!is_bifrost) {
/* TODO: Blend shaders routing/disasm */
-
union midgard_blend blend = s->blend;
- pandecode_midgard_blend(&blend, false);
+ 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);
}
pandecode_indent--;
else
shader = pandecode_midgard_blend_mrt(blend_base, job_no, i);
- if (shader & ~0xF) {
- struct midgard_disasm_stats stats =
- pandecode_shader_disassemble(shader, job_no, job_type, false);
-
- 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");
- }
+ if (shader & ~0xF)
+ pandecode_blend_shader_disassemble(shader, job_no, job_type, false, gpu_id);
}
}
pandecode_log("};\n");
}
- if (p->attribute_meta) {
- unsigned max_attr_index = pandecode_attribute_meta(job_no, attribute_count, p, false, suffix);
+ 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);
+ 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
/* Number of descriptors depends on whether there are
* non-internal varyings */
- pandecode_attributes(attr_mem, p->varyings, job_no, suffix, varying_count, true);
+ pandecode_attributes(attr_mem, p->varyings, job_no, suffix, varying_count, true, job_type);
}
if (p->uniform_buffers) {
if (p->uniforms) {
if (uniform_count)
- pandecode_validate_buffer(p->uniforms, uniform_count * 16);
+ pandecode_uniforms(p->uniforms, uniform_count);
else
pandecode_msg("warn: Uniforms specified but not referenced\n");
} else if (uniform_count)
static void
pandecode_vertex_tiler_postfix(const struct mali_vertex_tiler_postfix *p, int job_no, bool is_bifrost)
{
- if (!(p->position_varying || p->occlusion_counter || p->flags))
+ if (p->shader & 0xF)
+ pandecode_msg("warn: shader tagged %X\n", p->shader & 0xF);
+
+ if (!(p->position_varying || p->occlusion_counter))
return;
pandecode_log(".postfix = {\n");
MEMORY_PROP(p, position_varying);
MEMORY_PROP(p, occlusion_counter);
- if (p->flags)
- pandecode_prop("flags = %d", p->flags);
-
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)
+ 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);
+ pandecode_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", true, gpu_id);
pandecode_log("struct bifrost_payload_vertex payload_%d = {\n", job_no);
pandecode_indent++;
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)
+ 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);
+ 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_%d = {\n", job_no);
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)
+ mali_ptr payload, int job_no, unsigned gpu_id)
{
struct midgard_payload_vertex_tiler *PANDECODE_PTR_VAR(v, mem, payload);
- pandecode_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", false);
+ 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++;
static int job_descriptor_number = 0;
int
-pandecode_jc(mali_ptr jc_gpu_va, bool bifrost)
+pandecode_jc(mali_ptr jc_gpu_va, bool bifrost, unsigned gpu_id)
{
struct mali_job_descriptor_header *h;
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,
- MALI_PAYLOAD_SIZE);
+ payload = pandecode_fetch_gpu_mem(mem, payload_ptr, 256);
int job_no = job_descriptor_number++;
case JOB_TYPE_COMPUTE:
if (bifrost) {
if (h->job_type == JOB_TYPE_TILER)
- pandecode_tiler_job_bfr(h, mem, payload_ptr, job_no);
+ pandecode_tiler_job_bfr(h, mem, payload_ptr, job_no, gpu_id);
else
- pandecode_vertex_job_bfr(h, mem, payload_ptr, job_no);
+ 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);
+ pandecode_vertex_or_tiler_job_mdg(h, mem, payload_ptr, job_no, gpu_id);
break;
projects / mesa.git / blobdiff