#include "radeon/radeon_uvd.h"
#include "util/u_memory.h"
#include "util/u_suballoc.h"
+#include "util/u_tests.h"
#include "vl/vl_decoder.h"
#include "../ddebug/dd_util.h"
return LLVMCreateTargetMachine(si_llvm_get_amdgpu_target(triple), triple,
r600_get_llvm_processor_name(sscreen->b.family),
-#if HAVE_LLVM >= 0x0308
sscreen->b.debug_flags & DBG_SI_SCHED ?
SI_LLVM_DEFAULT_FEATURES ",+si-scheduler" :
-#endif
SI_LLVM_DEFAULT_FEATURES,
LLVMCodeGenLevelDefault,
LLVMRelocDefault,
sscreen->b.chip_class != SI &&
/* These can't use CE due to a power gating bug in the kernel. */
sscreen->b.family != CHIP_CARRIZO &&
- sscreen->b.family != CHIP_STONEY) {
+ sscreen->b.family != CHIP_STONEY &&
+ /* Some CE bug is causing green screen corruption w/ MPV video
+ * playback and occasional corruption w/ 3D. */
+ sscreen->b.chip_class != GFX9) {
sctx->ce_ib = ws->cs_add_const_ib(sctx->b.gfx.cs);
if (!sctx->ce_ib)
goto fail;
}
sctx->ce_suballocator =
- u_suballocator_create(&sctx->b.b, 1024 * 1024,
- 0, PIPE_USAGE_DEFAULT, 0, false);
+ u_suballocator_create(&sctx->b.b, 1024 * 1024, 0,
+ PIPE_USAGE_DEFAULT,
+ R600_RESOURCE_FLAG_UNMAPPABLE, false);
if (!sctx->ce_suballocator)
goto fail;
}
/* CIK cannot unbind a constant buffer (S_BUFFER_LOAD doesn't skip loads
* if NUM_RECORDS == 0). We need to use a dummy buffer instead. */
if (sctx->b.chip_class == CIK) {
- sctx->null_const_buf.buffer = pipe_buffer_create(screen, PIPE_BIND_CONSTANT_BUFFER,
- PIPE_USAGE_DEFAULT, 16);
+ sctx->null_const_buf.buffer =
+ r600_aligned_buffer_create(screen,
+ R600_RESOURCE_FLAG_UNMAPPABLE,
+ PIPE_USAGE_DEFAULT, 16,
+ sctx->screen->b.info.tcc_cache_line_size);
if (!sctx->null_const_buf.buffer)
goto fail;
sctx->null_const_buf.buffer_size = sctx->null_const_buf.buffer->width0;
{
/* Old kernels disallowed some register writes for SI
* that are used for indirect dispatches. */
- return HAVE_LLVM >= 0x309 &&
- (sscreen->b.chip_class >= CIK ||
+ return (sscreen->b.chip_class >= CIK ||
sscreen->b.info.drm_major == 3 ||
(sscreen->b.info.drm_major == 2 &&
sscreen->b.info.drm_minor >= 45));
case PIPE_CAP_INDEP_BLEND_FUNC:
case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
- case PIPE_CAP_USER_INDEX_BUFFERS:
case PIPE_CAP_USER_CONSTANT_BUFFERS:
case PIPE_CAP_START_INSTANCE:
case PIPE_CAP_NPOT_TEXTURES:
case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS:
- return 1;
-
case PIPE_CAP_DOUBLES:
- return HAVE_LLVM >= 0x0307;
+ case PIPE_CAP_TGSI_TEX_TXF_LZ:
+ case PIPE_CAP_TGSI_TES_LAYER_VIEWPORT:
case PIPE_CAP_INT64:
case PIPE_CAP_INT64_DIVMOD:
- return HAVE_LLVM >= 0x0309;
+ case PIPE_CAP_TGSI_CLOCK:
+ return 1;
+
+ case PIPE_CAP_TGSI_VOTE:
+ return HAVE_LLVM >= 0x0400;
+
+ case PIPE_CAP_TGSI_BALLOT:
+ return HAVE_LLVM >= 0x0500;
case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
return !SI_BIG_ENDIAN && sscreen->b.info.has_userptr;
case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
case PIPE_CAP_MAX_VERTEX_STREAMS:
- return 4;
-
case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
- return HAVE_LLVM >= 0x0309 ? 4 : 0;
+ return 4;
case PIPE_CAP_GLSL_FEATURE_LEVEL:
if (si_have_tgsi_compute(sscreen))
return 450;
- return HAVE_LLVM >= 0x0309 ? 420 :
- HAVE_LLVM >= 0x0307 ? 410 : 330;
+ return 420;
case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
return MIN2(sscreen->b.info.max_alloc_size, INT_MAX);
(sscreen->b.info.drm_major == 2 &&
sscreen->b.info.drm_minor < 50);
+ case PIPE_CAP_SPARSE_BUFFER_PAGE_SIZE:
+ /* Disable on SI due to VM faults in CP DMA. Enable once these
+ * faults are mitigated in software.
+ */
+ if (sscreen->b.chip_class >= CIK &&
+ sscreen->b.info.drm_major == 3 &&
+ sscreen->b.info.drm_minor >= 13)
+ return RADEON_SPARSE_PAGE_SIZE;
+ return 0;
+
/* Unsupported features. */
case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
case PIPE_CAP_VERTEXID_NOBASE:
case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES:
- case PIPE_CAP_TGSI_VOTE:
case PIPE_CAP_MAX_WINDOW_RECTANGLES:
case PIPE_CAP_NATIVE_FENCE_FD:
case PIPE_CAP_TGSI_FS_FBFETCH:
case PIPE_CAP_TGSI_MUL_ZERO_WINS:
case PIPE_CAP_UMA:
+ case PIPE_CAP_POLYGON_MODE_FILL_RECTANGLE:
return 0;
case PIPE_CAP_QUERY_BUFFER_OBJECT:
return 30;
case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
- return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_R600;
+ return sscreen->b.chip_class <= VI ?
+ PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_R600 : 0;
/* Stream output. */
case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
return 0;
}
-static int si_get_shader_param(struct pipe_screen* pscreen, unsigned shader, enum pipe_shader_cap param)
+static int si_get_shader_param(struct pipe_screen* pscreen,
+ enum pipe_shader_type shader,
+ enum pipe_shader_cap param)
{
struct si_screen *sscreen = (struct si_screen *)pscreen;
case PIPE_SHADER_FRAGMENT:
case PIPE_SHADER_VERTEX:
case PIPE_SHADER_GEOMETRY:
- break;
case PIPE_SHADER_TESS_CTRL:
case PIPE_SHADER_TESS_EVAL:
- /* LLVM 3.6.2 is required for tessellation because of bug fixes there */
- if (HAVE_LLVM == 0x0306 && MESA_LLVM_VERSION_PATCH < 2)
- return 0;
break;
case PIPE_SHADER_COMPUTE:
switch (param) {
case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
return SI_NUM_SAMPLERS;
case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
- return HAVE_LLVM >= 0x0309 ? SI_NUM_SHADER_BUFFERS : 0;
+ return SI_NUM_SHADER_BUFFERS;
case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
- return HAVE_LLVM >= 0x0309 ? SI_NUM_IMAGES : 0;
+ return SI_NUM_IMAGES;
case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
return 32;
case PIPE_SHADER_CAP_PREFERRED_IR:
case PIPE_SHADER_CAP_INTEGERS:
case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
+ case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS:
return 1;
case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
return shader != PIPE_SHADER_GEOMETRY;
/* Unsupported boolean features. */
- case PIPE_SHADER_CAP_MAX_PREDS:
case PIPE_SHADER_CAP_SUBROUTINES:
case PIPE_SHADER_CAP_SUPPORTED_IRS:
case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
struct si_screen *sscreen = (struct si_screen *)pscreen;
struct si_shader_part *parts[] = {
sscreen->vs_prologs,
- sscreen->vs_epilogs,
sscreen->tcs_epilogs,
sscreen->gs_prologs,
sscreen->ps_prologs,
};
unsigned i;
- if (!sscreen)
- return;
-
if (!sscreen->b.ws->unref(sscreen->b.ws))
return;
- if (util_queue_is_initialized(&sscreen->shader_compiler_queue))
- util_queue_destroy(&sscreen->shader_compiler_queue);
+ util_queue_destroy(&sscreen->shader_compiler_queue);
for (i = 0; i < ARRAY_SIZE(sscreen->tm); i++)
if (sscreen->tm[i])
FREE(part);
}
}
- pipe_mutex_destroy(sscreen->shader_parts_mutex);
+ mtx_destroy(&sscreen->shader_parts_mutex);
si_destroy_shader_cache(sscreen);
r600_destroy_common_screen(&sscreen->b);
}
case CHIP_POLARIS10:
case CHIP_POLARIS11:
case CHIP_POLARIS12:
+ case CHIP_VEGA10:
sscreen->gs_table_depth = 32;
return true;
default:
/* Override family and chip_class. */
sscreen->b.family = sscreen->b.info.family = i;
- if (i >= CHIP_TONGA)
+ if (i >= CHIP_VEGA10)
+ sscreen->b.chip_class = sscreen->b.info.chip_class = GFX9;
+ else if (i >= CHIP_TONGA)
sscreen->b.chip_class = sscreen->b.info.chip_class = VI;
else if (i >= CHIP_BONAIRE)
sscreen->b.chip_class = sscreen->b.info.chip_class = CIK;
exit(1);
}
+static void si_test_vmfault(struct si_screen *sscreen)
+{
+ struct pipe_context *ctx = sscreen->b.aux_context;
+ struct si_context *sctx = (struct si_context *)ctx;
+ struct pipe_resource *buf =
+ pipe_buffer_create(&sscreen->b.b, 0, PIPE_USAGE_DEFAULT, 64);
+
+ if (!buf) {
+ puts("Buffer allocation failed.");
+ exit(1);
+ }
+
+ r600_resource(buf)->gpu_address = 0; /* cause a VM fault */
+
+ if (sscreen->b.debug_flags & DBG_TEST_VMFAULT_CP) {
+ si_copy_buffer(sctx, buf, buf, 0, 4, 4, 0);
+ ctx->flush(ctx, NULL, 0);
+ puts("VM fault test: CP - done.");
+ }
+ if (sscreen->b.debug_flags & DBG_TEST_VMFAULT_SDMA) {
+ sctx->b.dma_clear_buffer(ctx, buf, 0, 4, 0);
+ ctx->flush(ctx, NULL, 0);
+ puts("VM fault test: SDMA - done.");
+ }
+ if (sscreen->b.debug_flags & DBG_TEST_VMFAULT_SHADER) {
+ util_test_constant_buffer(ctx, buf);
+ puts("VM fault test: Shader - done.");
+ }
+ exit(0);
+}
+
struct pipe_screen *radeonsi_screen_create(struct radeon_winsys *ws)
{
struct si_screen *sscreen = CALLOC_STRUCT(si_screen);
return NULL;
}
+ /* Only enable as many threads as we have target machines and CPUs. */
+ num_cpus = sysconf(_SC_NPROCESSORS_ONLN);
+ num_compiler_threads = MIN2(num_cpus, ARRAY_SIZE(sscreen->tm));
+
+ if (!util_queue_init(&sscreen->shader_compiler_queue, "si_shader",
+ 32, num_compiler_threads)) {
+ si_destroy_shader_cache(sscreen);
+ FREE(sscreen);
+ return NULL;
+ }
+
si_handle_env_var_force_family(sscreen);
if (!debug_get_bool_option("RADEON_DISABLE_PERFCOUNTERS", false))
sscreen->b.info.pfp_fw_version >= 121 &&
sscreen->b.info.me_fw_version >= 87);
- sscreen->has_ds_bpermute = HAVE_LLVM >= 0x0309 &&
- sscreen->b.chip_class >= VI;
+ sscreen->has_ds_bpermute = sscreen->b.chip_class >= VI;
+ sscreen->has_msaa_sample_loc_bug = (sscreen->b.family >= CHIP_POLARIS10 &&
+ sscreen->b.family <= CHIP_POLARIS12) ||
+ sscreen->b.family == CHIP_VEGA10;
sscreen->b.has_cp_dma = true;
sscreen->b.has_streamout = true;
- pipe_mutex_init(sscreen->shader_parts_mutex);
+
+ /* Some chips have RB+ registers, but don't support RB+. Those must
+ * always disable it.
+ */
+ if (sscreen->b.family == CHIP_STONEY ||
+ sscreen->b.chip_class >= GFX9) {
+ sscreen->b.has_rbplus = true;
+
+ sscreen->b.rbplus_allowed =
+ !(sscreen->b.debug_flags & DBG_NO_RB_PLUS) &&
+ sscreen->b.family == CHIP_STONEY;
+ }
+
+ (void) mtx_init(&sscreen->shader_parts_mutex, mtx_plain);
sscreen->use_monolithic_shaders =
- HAVE_LLVM < 0x0308 ||
(sscreen->b.debug_flags & DBG_MONOLITHIC_SHADERS) != 0;
sscreen->b.barrier_flags.cp_to_L2 = SI_CONTEXT_INV_SMEM_L1 |
if (debug_get_bool_option("RADEON_DUMP_SHADERS", false))
sscreen->b.debug_flags |= DBG_FS | DBG_VS | DBG_GS | DBG_PS | DBG_CS;
- /* Only enable as many threads as we have target machines and CPUs. */
- num_cpus = sysconf(_SC_NPROCESSORS_ONLN);
- num_compiler_threads = MIN2(num_cpus, ARRAY_SIZE(sscreen->tm));
-
for (i = 0; i < num_compiler_threads; i++)
sscreen->tm[i] = si_create_llvm_target_machine(sscreen);
- util_queue_init(&sscreen->shader_compiler_queue, "si_shader",
- 32, num_compiler_threads);
-
/* Create the auxiliary context. This must be done last. */
sscreen->b.aux_context = sscreen->b.b.context_create(&sscreen->b.b, NULL, 0);
if (sscreen->b.debug_flags & DBG_TEST_DMA)
r600_test_dma(&sscreen->b);
+ if (sscreen->b.debug_flags & (DBG_TEST_VMFAULT_CP |
+ DBG_TEST_VMFAULT_SDMA |
+ DBG_TEST_VMFAULT_SHADER))
+ si_test_vmfault(sscreen);
+
return &sscreen->b.b;
}
projects / mesa.git / blobdiff