static struct pipe_context *si_create_context(struct pipe_screen *screen, void *priv)
{
struct si_context *sctx = CALLOC_STRUCT(si_context);
struct si_screen* sscreen = (struct si_screen *)screen;
struct radeon_winsys *ws = sscreen->b.ws;
int shader, i;
if (sctx == NULL)
return NULL;
sctx->b.b.screen = screen; /* this must be set first */
sctx->b.b.priv = priv;
sctx->b.b.destroy = si_destroy_context;
sctx->screen = sscreen; /* Easy accessing of screen/winsys. */
if (!r600_common_context_init(&sctx->b, &sscreen->b))
goto fail;
si_init_blit_functions(sctx);
si_init_compute_functions(sctx);
if (sscreen->b.info.has_uvd) {
sctx->b.b.create_video_codec = si_uvd_create_decoder;
sctx->b.b.create_video_buffer = si_video_buffer_create;
} else {
sctx->b.b.create_video_codec = vl_create_decoder;
sctx->b.b.create_video_buffer = vl_video_buffer_create;
}
sctx->b.rings.gfx.cs = ws->cs_create(ws, RING_GFX, si_context_gfx_flush,
sctx, NULL);
sctx->b.rings.gfx.flush = si_context_gfx_flush;
si_init_all_descriptors(sctx);
/* Initialize cache_flush. */
sctx->cache_flush = si_atom_cache_flush;
sctx->atoms.s.cache_flush = &sctx->cache_flush;
sctx->msaa_config = si_atom_msaa_config;
sctx->atoms.s.msaa_config = &sctx->msaa_config;
sctx->atoms.s.streamout_begin = &sctx->b.streamout.begin_atom;
sctx->atoms.s.streamout_enable = &sctx->b.streamout.enable_atom;
switch (sctx->b.chip_class) {
case SI:
case CIK:
si_init_state_functions(sctx);
si_init_config(sctx);
break;
default:
R600_ERR("Unsupported chip class %d.\n", sctx->b.chip_class);
goto fail;
}
sctx->blitter = util_blitter_create(&sctx->b.b);
if (sctx->blitter == NULL)
goto fail;
sctx->blitter->draw_rectangle = r600_draw_rectangle;
sctx->dummy_pixel_shader =
util_make_fragment_cloneinput_shader(&sctx->b.b, 0,
TGSI_SEMANTIC_GENERIC,
TGSI_INTERPOLATE_CONSTANT);
sctx->b.b.bind_fs_state(&sctx->b.b, sctx->dummy_pixel_shader);
/* these must be last */
si_begin_new_cs(sctx);
r600_query_init_backend_mask(&sctx->b); /* this emits commands and must be last */
/* CIK cannot unbind a constant buffer (S_BUFFER_LOAD is buggy
* with a NULL buffer). 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_size = sctx->null_const_buf.buffer->width0;
for (shader = 0; shader < SI_NUM_SHADERS; shader++) {
for (i = 0; i < SI_NUM_CONST_BUFFERS; i++) {
sctx->b.b.set_constant_buffer(&sctx->b.b, shader, i,
&sctx->null_const_buf);
}
}
/* Clear the NULL constant buffer, because loads should return zeros. */
sctx->b.clear_buffer(&sctx->b.b, sctx->null_const_buf.buffer, 0,
sctx->null_const_buf.buffer->width0, 0);
}
return &sctx->b.b;
fail:
si_destroy_context(&sctx->b.b);
return NULL;
}
static struct pipe_context *si_create_context(struct pipe_screen *screen,
void *priv, unsigned flags)
{
struct si_context *sctx = CALLOC_STRUCT(si_context);
struct si_screen* sscreen = (struct si_screen *)screen;
struct radeon_winsys *ws = sscreen->b.ws;
LLVMTargetRef r600_target;
const char *triple = "amdgcn--";
int shader, i;
if (!sctx)
return NULL;
if (sscreen->b.debug_flags & DBG_CHECK_VM)
flags |= PIPE_CONTEXT_DEBUG;
sctx->b.b.screen = screen; /* this must be set first */
sctx->b.b.priv = priv;
sctx->b.b.destroy = si_destroy_context;
sctx->b.set_atom_dirty = (void *)si_set_atom_dirty;
sctx->screen = sscreen; /* Easy accessing of screen/winsys. */
sctx->is_debug = (flags & PIPE_CONTEXT_DEBUG) != 0;
if (!r600_common_context_init(&sctx->b, &sscreen->b))
goto fail;
if (sscreen->b.info.drm_major == 3)
sctx->b.b.get_device_reset_status = si_amdgpu_get_reset_status;
si_init_blit_functions(sctx);
si_init_compute_functions(sctx);
si_init_cp_dma_functions(sctx);
si_init_debug_functions(sctx);
if (sscreen->b.info.has_uvd) {
sctx->b.b.create_video_codec = si_uvd_create_decoder;
sctx->b.b.create_video_buffer = si_video_buffer_create;
} else {
sctx->b.b.create_video_codec = vl_create_decoder;
sctx->b.b.create_video_buffer = vl_video_buffer_create;
}
sctx->b.gfx.cs = ws->cs_create(sctx->b.ctx, RING_GFX,
si_context_gfx_flush, sctx);
if (!(sscreen->b.debug_flags & DBG_NO_CE) && ws->cs_add_const_ib) {
sctx->ce_ib = ws->cs_add_const_ib(sctx->b.gfx.cs);
if (!sctx->ce_ib)
goto fail;
if (ws->cs_add_const_preamble_ib) {
sctx->ce_preamble_ib =
ws->cs_add_const_preamble_ib(sctx->b.gfx.cs);
if (!sctx->ce_preamble_ib)
goto fail;
}
sctx->ce_suballocator =
u_suballocator_create(&sctx->b.b, 1024 * 1024,
64, PIPE_BIND_CUSTOM,
PIPE_USAGE_DEFAULT, FALSE);
if (!sctx->ce_suballocator)
goto fail;
}
sctx->b.gfx.flush = si_context_gfx_flush;
/* Border colors. */
sctx->border_color_table = malloc(SI_MAX_BORDER_COLORS *
sizeof(*sctx->border_color_table));
if (!sctx->border_color_table)
goto fail;
sctx->border_color_buffer = (struct r600_resource*)
pipe_buffer_create(screen, PIPE_BIND_CUSTOM, PIPE_USAGE_DEFAULT,
SI_MAX_BORDER_COLORS *
sizeof(*sctx->border_color_table));
if (!sctx->border_color_buffer)
goto fail;
sctx->border_color_map =
ws->buffer_map(sctx->border_color_buffer->buf,
NULL, PIPE_TRANSFER_WRITE);
if (!sctx->border_color_map)
goto fail;
si_init_all_descriptors(sctx);
si_init_state_functions(sctx);
si_init_shader_functions(sctx);
if (sctx->b.chip_class >= CIK)
cik_init_sdma_functions(sctx);
else
si_init_dma_functions(sctx);
if (sscreen->b.debug_flags & DBG_FORCE_DMA)
sctx->b.b.resource_copy_region = sctx->b.dma_copy;
sctx->blitter = util_blitter_create(&sctx->b.b);
if (sctx->blitter == NULL)
goto fail;
sctx->blitter->draw_rectangle = r600_draw_rectangle;
sctx->sample_mask.sample_mask = 0xffff;
/* these must be last */
si_begin_new_cs(sctx);
r600_query_init_backend_mask(&sctx->b); /* this emits commands and must be last */
/* CIK cannot unbind a constant buffer (S_BUFFER_LOAD is buggy
* with a NULL buffer). 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);
if (!sctx->null_const_buf.buffer)
goto fail;
sctx->null_const_buf.buffer_size = sctx->null_const_buf.buffer->width0;
for (shader = 0; shader < SI_NUM_SHADERS; shader++) {
for (i = 0; i < SI_NUM_CONST_BUFFERS; i++) {
sctx->b.b.set_constant_buffer(&sctx->b.b, shader, i,
&sctx->null_const_buf);
}
}
/* Clear the NULL constant buffer, because loads should return zeros. */
sctx->b.clear_buffer(&sctx->b.b, sctx->null_const_buf.buffer, 0,
sctx->null_const_buf.buffer->width0, 0,
R600_COHERENCY_SHADER);
}
/* XXX: This is the maximum value allowed. I'm not sure how to compute
* this for non-cs shaders. Using the wrong value here can result in
* GPU lockups, but the maximum value seems to always work.
*/
sctx->scratch_waves = 32 * sscreen->b.info.num_good_compute_units;
/* Initialize LLVM TargetMachine */
r600_target = radeon_llvm_get_r600_target(triple);
sctx->tm = LLVMCreateTargetMachine(r600_target, triple,
r600_get_llvm_processor_name(sscreen->b.family),
#if HAVE_LLVM >= 0x0308
sscreen->b.debug_flags & DBG_SI_SCHED ?
"+DumpCode,+vgpr-spilling,+si-scheduler" :
#endif
"+DumpCode,+vgpr-spilling",
LLVMCodeGenLevelDefault,
LLVMRelocDefault,
LLVMCodeModelDefault);
return &sctx->b.b;
fail:
fprintf(stderr, "radeonsi: Failed to create a context.\n");
si_destroy_context(&sctx->b.b);
return NULL;
}
static struct pipe_context *si_create_context(struct pipe_screen *screen,
unsigned flags)
{
struct si_context *sctx = CALLOC_STRUCT(si_context);
struct si_screen* sscreen = (struct si_screen *)screen;
struct radeon_winsys *ws = sscreen->ws;
int shader, i;
bool stop_exec_on_failure = (flags & PIPE_CONTEXT_LOSE_CONTEXT_ON_RESET) != 0;
if (!sctx)
return NULL;
sctx->has_graphics = sscreen->info.chip_class == SI ||
!(flags & PIPE_CONTEXT_COMPUTE_ONLY);
if (flags & PIPE_CONTEXT_DEBUG)
sscreen->record_llvm_ir = true; /* racy but not critical */
sctx->b.screen = screen; /* this must be set first */
sctx->b.priv = NULL;
sctx->b.destroy = si_destroy_context;
sctx->screen = sscreen; /* Easy accessing of screen/winsys. */
sctx->is_debug = (flags & PIPE_CONTEXT_DEBUG) != 0;
slab_create_child(&sctx->pool_transfers, &sscreen->pool_transfers);
slab_create_child(&sctx->pool_transfers_unsync, &sscreen->pool_transfers);
sctx->ws = sscreen->ws;
sctx->family = sscreen->info.family;
sctx->chip_class = sscreen->info.chip_class;
if (sscreen->info.has_gpu_reset_counter_query) {
sctx->gpu_reset_counter =
sctx->ws->query_value(sctx->ws, RADEON_GPU_RESET_COUNTER);
}
if (sctx->chip_class == CIK ||
sctx->chip_class == VI ||
sctx->chip_class == GFX9) {
sctx->eop_bug_scratch = si_resource(
pipe_buffer_create(&sscreen->b, 0, PIPE_USAGE_DEFAULT,
16 * sscreen->info.num_render_backends));
if (!sctx->eop_bug_scratch)
goto fail;
}
/* Initialize context allocators. */
sctx->allocator_zeroed_memory =
u_suballocator_create(&sctx->b, 128 * 1024,
0, PIPE_USAGE_DEFAULT,
SI_RESOURCE_FLAG_UNMAPPABLE |
SI_RESOURCE_FLAG_CLEAR, false);
if (!sctx->allocator_zeroed_memory)
goto fail;
sctx->b.stream_uploader = u_upload_create(&sctx->b, 1024 * 1024,
0, PIPE_USAGE_STREAM,
SI_RESOURCE_FLAG_READ_ONLY);
if (!sctx->b.stream_uploader)
goto fail;
sctx->cached_gtt_allocator = u_upload_create(&sctx->b, 16 * 1024,
0, PIPE_USAGE_STAGING, 0);
if (!sctx->cached_gtt_allocator)
goto fail;
sctx->ctx = sctx->ws->ctx_create(sctx->ws);
if (!sctx->ctx)
goto fail;
if (sscreen->info.num_sdma_rings && !(sscreen->debug_flags & DBG(NO_ASYNC_DMA))) {
sctx->dma_cs = sctx->ws->cs_create(sctx->ctx, RING_DMA,
(void*)si_flush_dma_cs,
sctx, stop_exec_on_failure);
}
bool use_sdma_upload = sscreen->info.has_dedicated_vram && sctx->dma_cs;
sctx->b.const_uploader = u_upload_create(&sctx->b, 256 * 1024,
0, PIPE_USAGE_DEFAULT,
SI_RESOURCE_FLAG_32BIT |
(use_sdma_upload ?
SI_RESOURCE_FLAG_UPLOAD_FLUSH_EXPLICIT_VIA_SDMA :
(sscreen->cpdma_prefetch_writes_memory ?
0 : SI_RESOURCE_FLAG_READ_ONLY)));
if (!sctx->b.const_uploader)
goto fail;
if (use_sdma_upload)
u_upload_enable_flush_explicit(sctx->b.const_uploader);
sctx->gfx_cs = ws->cs_create(sctx->ctx,
sctx->has_graphics ? RING_GFX : RING_COMPUTE,
(void*)si_flush_gfx_cs, sctx, stop_exec_on_failure);
/* Border colors. */
sctx->border_color_table = malloc(SI_MAX_BORDER_COLORS *
sizeof(*sctx->border_color_table));
if (!sctx->border_color_table)
goto fail;
sctx->border_color_buffer = si_resource(
pipe_buffer_create(screen, 0, PIPE_USAGE_DEFAULT,
SI_MAX_BORDER_COLORS *
sizeof(*sctx->border_color_table)));
if (!sctx->border_color_buffer)
goto fail;
sctx->border_color_map =
ws->buffer_map(sctx->border_color_buffer->buf,
NULL, PIPE_TRANSFER_WRITE);
if (!sctx->border_color_map)
goto fail;
/* Initialize context functions used by graphics and compute. */
sctx->b.emit_string_marker = si_emit_string_marker;
sctx->b.set_debug_callback = si_set_debug_callback;
sctx->b.set_log_context = si_set_log_context;
sctx->b.set_context_param = si_set_context_param;
sctx->b.get_device_reset_status = si_get_reset_status;
sctx->b.set_device_reset_callback = si_set_device_reset_callback;
si_init_all_descriptors(sctx);
si_init_buffer_functions(sctx);
si_init_clear_functions(sctx);
si_init_blit_functions(sctx);
si_init_compute_functions(sctx);
si_init_compute_blit_functions(sctx);
si_init_debug_functions(sctx);
si_init_fence_functions(sctx);
si_init_state_compute_functions(sctx);
if (sscreen->debug_flags & DBG(FORCE_DMA))
sctx->b.resource_copy_region = sctx->dma_copy;
/* Initialize graphics-only context functions. */
if (sctx->has_graphics) {
si_init_context_texture_functions(sctx);
si_init_query_functions(sctx);
si_init_msaa_functions(sctx);
si_init_shader_functions(sctx);
si_init_state_functions(sctx);
si_init_streamout_functions(sctx);
si_init_viewport_functions(sctx);
sctx->blitter = util_blitter_create(&sctx->b);
if (sctx->blitter == NULL)
goto fail;
sctx->blitter->skip_viewport_restore = true;
si_init_draw_functions(sctx);
}
/* Initialize SDMA functions. */
if (sctx->chip_class >= CIK)
cik_init_sdma_functions(sctx);
else
si_init_dma_functions(sctx);
sctx->sample_mask = 0xffff;
/* Initialize multimedia functions. */
if (sscreen->info.has_hw_decode) {
sctx->b.create_video_codec = si_uvd_create_decoder;
sctx->b.create_video_buffer = si_video_buffer_create;
} else {
sctx->b.create_video_codec = vl_create_decoder;
sctx->b.create_video_buffer = vl_video_buffer_create;
}
if (sctx->chip_class >= GFX9) {
sctx->wait_mem_scratch = si_resource(
pipe_buffer_create(screen, 0, PIPE_USAGE_DEFAULT, 4));
if (!sctx->wait_mem_scratch)
goto fail;
/* Initialize the memory. */
si_cp_write_data(sctx, sctx->wait_mem_scratch, 0, 4,
V_370_MEM, V_370_ME, &sctx->wait_mem_number);
}
/* 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->chip_class == CIK) {
sctx->null_const_buf.buffer =
pipe_aligned_buffer_create(screen,
SI_RESOURCE_FLAG_32BIT,
PIPE_USAGE_DEFAULT, 16,
sctx->screen->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;
unsigned start_shader = sctx->has_graphics ? 0 : PIPE_SHADER_COMPUTE;
for (shader = start_shader; shader < SI_NUM_SHADERS; shader++) {
for (i = 0; i < SI_NUM_CONST_BUFFERS; i++) {
sctx->b.set_constant_buffer(&sctx->b, shader, i,
&sctx->null_const_buf);
}
}
si_set_rw_buffer(sctx, SI_HS_CONST_DEFAULT_TESS_LEVELS,
&sctx->null_const_buf);
si_set_rw_buffer(sctx, SI_VS_CONST_INSTANCE_DIVISORS,
&sctx->null_const_buf);
si_set_rw_buffer(sctx, SI_VS_CONST_CLIP_PLANES,
&sctx->null_const_buf);
si_set_rw_buffer(sctx, SI_PS_CONST_POLY_STIPPLE,
&sctx->null_const_buf);
si_set_rw_buffer(sctx, SI_PS_CONST_SAMPLE_POSITIONS,
&sctx->null_const_buf);
}
uint64_t max_threads_per_block;
screen->get_compute_param(screen, PIPE_SHADER_IR_TGSI,
PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK,
&max_threads_per_block);
/* The maximum number of scratch waves. Scratch space isn't divided
* evenly between CUs. The number is only a function of the number of CUs.
* We can decrease the constant to decrease the scratch buffer size.
*
* sctx->scratch_waves must be >= the maximum posible size of
* 1 threadgroup, so that the hw doesn't hang from being unable
* to start any.
*
* The recommended value is 4 per CU at most. Higher numbers don't
* bring much benefit, but they still occupy chip resources (think
* async compute). I've seen ~2% performance difference between 4 and 32.
*/
sctx->scratch_waves = MAX2(32 * sscreen->info.num_good_compute_units,
max_threads_per_block / 64);
si_init_compiler(sscreen, &sctx->compiler);
/* Bindless handles. */
sctx->tex_handles = _mesa_hash_table_create(NULL, _mesa_hash_pointer,
_mesa_key_pointer_equal);
sctx->img_handles = _mesa_hash_table_create(NULL, _mesa_hash_pointer,
_mesa_key_pointer_equal);
util_dynarray_init(&sctx->resident_tex_handles, NULL);
util_dynarray_init(&sctx->resident_img_handles, NULL);
util_dynarray_init(&sctx->resident_tex_needs_color_decompress, NULL);
util_dynarray_init(&sctx->resident_img_needs_color_decompress, NULL);
util_dynarray_init(&sctx->resident_tex_needs_depth_decompress, NULL);
sctx->sample_pos_buffer =
pipe_buffer_create(sctx->b.screen, 0, PIPE_USAGE_DEFAULT,
sizeof(sctx->sample_positions));
pipe_buffer_write(&sctx->b, sctx->sample_pos_buffer, 0,
sizeof(sctx->sample_positions), &sctx->sample_positions);
/* this must be last */
si_begin_new_gfx_cs(sctx);
if (sctx->chip_class == CIK) {
/* Clear the NULL constant buffer, because loads should return zeros.
* Note that this forces CP DMA to be used, because clover deadlocks
* for some reason when the compute codepath is used.
*/
uint32_t clear_value = 0;
si_clear_buffer(sctx, sctx->null_const_buf.buffer, 0,
sctx->null_const_buf.buffer->width0,
&clear_value, 4, SI_COHERENCY_SHADER, true);
}
return &sctx->b;
fail:
fprintf(stderr, "radeonsi: Failed to create a context.\n");
si_destroy_context(&sctx->b);
return NULL;
}
static struct pipe_context *si_create_context(struct pipe_screen *screen,
void *priv, unsigned flags)
{
struct si_context *sctx = CALLOC_STRUCT(si_context);
struct si_screen* sscreen = (struct si_screen *)screen;
struct radeon_winsys *ws = sscreen->b.ws;
int shader, i;
if (!sctx)
return NULL;
if (sscreen->b.debug_flags & DBG_CHECK_VM)
flags |= PIPE_CONTEXT_DEBUG;
if (flags & PIPE_CONTEXT_DEBUG)
sscreen->record_llvm_ir = true; /* racy but not critical */
sctx->b.b.screen = screen; /* this must be set first */
sctx->b.b.priv = priv;
sctx->b.b.destroy = si_destroy_context;
sctx->b.b.emit_string_marker = si_emit_string_marker;
sctx->b.set_atom_dirty = (void *)si_set_atom_dirty;
sctx->screen = sscreen; /* Easy accessing of screen/winsys. */
sctx->is_debug = (flags & PIPE_CONTEXT_DEBUG) != 0;
if (!r600_common_context_init(&sctx->b, &sscreen->b))
goto fail;
if (sscreen->b.info.drm_major == 3)
sctx->b.b.get_device_reset_status = si_amdgpu_get_reset_status;
si_init_blit_functions(sctx);
si_init_compute_functions(sctx);
si_init_cp_dma_functions(sctx);
si_init_debug_functions(sctx);
if (sscreen->b.info.has_uvd) {
sctx->b.b.create_video_codec = si_uvd_create_decoder;
sctx->b.b.create_video_buffer = si_video_buffer_create;
} else {
sctx->b.b.create_video_codec = vl_create_decoder;
sctx->b.b.create_video_buffer = vl_video_buffer_create;
}
sctx->b.gfx.cs = ws->cs_create(sctx->b.ctx, RING_GFX,
si_context_gfx_flush, sctx);
if (!(sscreen->b.debug_flags & DBG_NO_CE) && ws->cs_add_const_ib) {
sctx->ce_ib = ws->cs_add_const_ib(sctx->b.gfx.cs);
if (!sctx->ce_ib)
goto fail;
if (ws->cs_add_const_preamble_ib) {
sctx->ce_preamble_ib =
ws->cs_add_const_preamble_ib(sctx->b.gfx.cs);
if (!sctx->ce_preamble_ib)
goto fail;
}
sctx->ce_suballocator =
u_suballocator_create(&sctx->b.b, 1024 * 1024,
PIPE_BIND_CUSTOM,
PIPE_USAGE_DEFAULT, false);
if (!sctx->ce_suballocator)
goto fail;
}
sctx->b.gfx.flush = si_context_gfx_flush;
/* Border colors. */
sctx->border_color_table = malloc(SI_MAX_BORDER_COLORS *
sizeof(*sctx->border_color_table));
if (!sctx->border_color_table)
goto fail;
sctx->border_color_buffer = (struct r600_resource*)
pipe_buffer_create(screen, PIPE_BIND_CUSTOM, PIPE_USAGE_DEFAULT,
SI_MAX_BORDER_COLORS *
sizeof(*sctx->border_color_table));
if (!sctx->border_color_buffer)
goto fail;
sctx->border_color_map =
ws->buffer_map(sctx->border_color_buffer->buf,
NULL, PIPE_TRANSFER_WRITE);
if (!sctx->border_color_map)
goto fail;
si_init_all_descriptors(sctx);
si_init_state_functions(sctx);
si_init_shader_functions(sctx);
if (sctx->b.chip_class >= CIK)
cik_init_sdma_functions(sctx);
else
si_init_dma_functions(sctx);
if (sscreen->b.debug_flags & DBG_FORCE_DMA)
sctx->b.b.resource_copy_region = sctx->b.dma_copy;
sctx->blitter = util_blitter_create(&sctx->b.b);
if (sctx->blitter == NULL)
goto fail;
sctx->blitter->draw_rectangle = r600_draw_rectangle;
sctx->sample_mask.sample_mask = 0xffff;
/* these must be last */
si_begin_new_cs(sctx);
r600_query_init_backend_mask(&sctx->b); /* this emits commands and must be last */
/* CIK cannot unbind a constant buffer (S_BUFFER_LOAD is buggy
* with a NULL buffer). 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);
if (!sctx->null_const_buf.buffer)
goto fail;
sctx->null_const_buf.buffer_size = sctx->null_const_buf.buffer->width0;
for (shader = 0; shader < SI_NUM_SHADERS; shader++) {
for (i = 0; i < SI_NUM_CONST_BUFFERS; i++) {
sctx->b.b.set_constant_buffer(&sctx->b.b, shader, i,
&sctx->null_const_buf);
}
}
/* Clear the NULL constant buffer, because loads should return zeros. */
sctx->b.clear_buffer(&sctx->b.b, sctx->null_const_buf.buffer, 0,
sctx->null_const_buf.buffer->width0, 0,
R600_COHERENCY_SHADER);
}
uint64_t max_threads_per_block;
screen->get_compute_param(screen, PIPE_SHADER_IR_TGSI,
PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK,
&max_threads_per_block);
/* The maximum number of scratch waves. Scratch space isn't divided
* evenly between CUs. The number is only a function of the number of CUs.
* We can decrease the constant to decrease the scratch buffer size.
*
* sctx->scratch_waves must be >= the maximum posible size of
* 1 threadgroup, so that the hw doesn't hang from being unable
* to start any.
*
* The recommended value is 4 per CU at most. Higher numbers don't
* bring much benefit, but they still occupy chip resources (think
* async compute). I've seen ~2% performance difference between 4 and 32.
*/
sctx->scratch_waves = MAX2(32 * sscreen->b.info.num_good_compute_units,
max_threads_per_block / 64);
sctx->tm = si_create_llvm_target_machine(sscreen);
return &sctx->b.b;
fail:
fprintf(stderr, "radeonsi: Failed to create a context.\n");
si_destroy_context(&sctx->b.b);
return NULL;
}
static struct pipe_context *si_create_context(struct pipe_screen *screen, void *priv)
{
struct si_context *sctx = CALLOC_STRUCT(si_context);
struct si_screen* sscreen = (struct si_screen *)screen;
struct radeon_winsys *ws = sscreen->b.ws;
LLVMTargetRef r600_target;
#if HAVE_LLVM >= 0x0306
const char *triple = "amdgcn--";
#endif
int shader, i;
if (sctx == NULL)
return NULL;
sctx->b.b.screen = screen; /* this must be set first */
sctx->b.b.priv = priv;
sctx->b.b.destroy = si_destroy_context;
sctx->b.set_atom_dirty = (void *)si_set_atom_dirty;
sctx->screen = sscreen; /* Easy accessing of screen/winsys. */
if (!r600_common_context_init(&sctx->b, &sscreen->b))
goto fail;
if (sscreen->b.info.drm_major == 3)
sctx->b.b.get_device_reset_status = si_amdgpu_get_reset_status;
si_init_blit_functions(sctx);
si_init_compute_functions(sctx);
si_init_cp_dma_functions(sctx);
if (sscreen->b.info.has_uvd) {
sctx->b.b.create_video_codec = si_uvd_create_decoder;
sctx->b.b.create_video_buffer = si_video_buffer_create;
} else {
sctx->b.b.create_video_codec = vl_create_decoder;
sctx->b.b.create_video_buffer = vl_video_buffer_create;
}
sctx->b.rings.gfx.cs = ws->cs_create(sctx->b.ctx, RING_GFX, si_context_gfx_flush,
sctx, sscreen->b.trace_bo ?
sscreen->b.trace_bo->cs_buf : NULL);
sctx->b.rings.gfx.flush = si_context_gfx_flush;
si_init_all_descriptors(sctx);
/* Initialize cache_flush. */
sctx->cache_flush = si_atom_cache_flush;
sctx->atoms.s.cache_flush = &sctx->cache_flush;
sctx->msaa_sample_locs = si_atom_msaa_sample_locs;
sctx->atoms.s.msaa_sample_locs = &sctx->msaa_sample_locs;
sctx->msaa_config = si_atom_msaa_config;
sctx->atoms.s.msaa_config = &sctx->msaa_config;
sctx->atoms.s.streamout_begin = &sctx->b.streamout.begin_atom;
sctx->atoms.s.streamout_enable = &sctx->b.streamout.enable_atom;
si_init_state_functions(sctx);
si_init_shader_functions(sctx);
if (sscreen->b.debug_flags & DBG_FORCE_DMA)
sctx->b.b.resource_copy_region = sctx->b.dma_copy;
sctx->blitter = util_blitter_create(&sctx->b.b);
if (sctx->blitter == NULL)
goto fail;
sctx->blitter->draw_rectangle = r600_draw_rectangle;
/* these must be last */
si_begin_new_cs(sctx);
r600_query_init_backend_mask(&sctx->b); /* this emits commands and must be last */
/* CIK cannot unbind a constant buffer (S_BUFFER_LOAD is buggy
* with a NULL buffer). 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);
if (!sctx->null_const_buf.buffer)
goto fail;
sctx->null_const_buf.buffer_size = sctx->null_const_buf.buffer->width0;
for (shader = 0; shader < SI_NUM_SHADERS; shader++) {
for (i = 0; i < SI_NUM_CONST_BUFFERS; i++) {
sctx->b.b.set_constant_buffer(&sctx->b.b, shader, i,
&sctx->null_const_buf);
}
}
/* Clear the NULL constant buffer, because loads should return zeros. */
sctx->b.clear_buffer(&sctx->b.b, sctx->null_const_buf.buffer, 0,
sctx->null_const_buf.buffer->width0, 0, false);
}
/* XXX: This is the maximum value allowed. I'm not sure how to compute
* this for non-cs shaders. Using the wrong value here can result in
* GPU lockups, but the maximum value seems to always work.
*/
sctx->scratch_waves = 32 * sscreen->b.info.max_compute_units;
#if HAVE_LLVM >= 0x0306
/* Initialize LLVM TargetMachine */
r600_target = radeon_llvm_get_r600_target(triple);
sctx->tm = LLVMCreateTargetMachine(r600_target, triple,
r600_get_llvm_processor_name(sscreen->b.family),
"+DumpCode,+vgpr-spilling",
LLVMCodeGenLevelDefault,
LLVMRelocDefault,
LLVMCodeModelDefault);
#endif
return &sctx->b.b;
fail:
si_destroy_context(&sctx->b.b);
return NULL;
}