radeon_set_context_reg(cmd_buffer->cs, R_028804_DB_EQAA, ms->db_eqaa);
radeon_set_context_reg(cmd_buffer->cs, R_028A4C_PA_SC_MODE_CNTL_1, ms->pa_sc_mode_cntl_1);
- if (old_pipeline && num_samples == old_pipeline->graphics.ms.num_samples)
+ if (old_pipeline && num_samples == old_pipeline->graphics.ms.num_samples &&
+ old_pipeline->shaders[MESA_SHADER_FRAGMENT]->info.info.ps.needs_sample_positions == pipeline->shaders[MESA_SHADER_FRAGMENT]->info.info.ps.needs_sample_positions)
return;
radeon_set_context_reg_seq(cmd_buffer->cs, R_028BDC_PA_SC_LINE_CNTL, 2);
va += image->offset + image->clear_value_offset;
unsigned reg_offset = 0, reg_count = 0;
- if (!image->surface.htile_size)
- return;
+ assert(image->surface.htile_size);
if (aspects & VK_IMAGE_ASPECT_STENCIL_BIT) {
++reg_count;
uint64_t va = radv_buffer_get_va(image->bo);
va += image->offset + image->dcc_pred_offset;
- if (!image->surface.dcc_size)
- return;
+ assert(image->surface.dcc_size);
radeon_emit(cmd_buffer->cs, PKT3(PKT3_WRITE_DATA, 4, 0));
radeon_emit(cmd_buffer->cs, S_370_DST_SEL(V_370_MEM_ASYNC) |
uint64_t va = radv_buffer_get_va(image->bo);
va += image->offset + image->clear_value_offset;
- if (!image->cmask.size && !image->surface.dcc_size)
- return;
+ assert(image->cmask.size || image->surface.dcc_size);
radeon_emit(cmd_buffer->cs, PKT3(PKT3_WRITE_DATA, 4, 0));
radeon_emit(cmd_buffer->cs, S_370_DST_SEL(V_370_MEM_ASYNC) |
uint64_t va;
stages &= cmd_buffer->push_constant_stages;
- if (!stages || !layout || (!layout->push_constant_size && !layout->dynamic_offset_count))
+ if (!stages ||
+ (!layout->push_constant_size && !layout->dynamic_offset_count))
return;
if (!radv_cmd_buffer_upload_alloc(cmd_buffer, layout->push_constant_size +
{
struct radv_shader_variant *compute_shader;
struct radv_pipeline *pipeline = cmd_buffer->state.compute_pipeline;
+ struct radv_device *device = cmd_buffer->device;
+ unsigned compute_resource_limits;
+ unsigned waves_per_threadgroup;
uint64_t va;
if (!pipeline || pipeline == cmd_buffer->state.emitted_compute_pipeline)
va = radv_buffer_get_va(compute_shader->bo) + compute_shader->bo_offset;
MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws,
- cmd_buffer->cs, 16);
+ cmd_buffer->cs, 19);
radeon_set_sh_reg_seq(cmd_buffer->cs, R_00B830_COMPUTE_PGM_LO, 2);
radeon_emit(cmd_buffer->cs, va >> 8);
S_00B860_WAVES(pipeline->max_waves) |
S_00B860_WAVESIZE(pipeline->scratch_bytes_per_wave >> 10));
+ /* Calculate best compute resource limits. */
+ waves_per_threadgroup =
+ DIV_ROUND_UP(compute_shader->info.cs.block_size[0] *
+ compute_shader->info.cs.block_size[1] *
+ compute_shader->info.cs.block_size[2], 64);
+ compute_resource_limits =
+ S_00B854_SIMD_DEST_CNTL(waves_per_threadgroup % 4 == 0);
+
+ if (device->physical_device->rad_info.chip_class >= CIK) {
+ unsigned num_cu_per_se =
+ device->physical_device->rad_info.num_good_compute_units /
+ device->physical_device->rad_info.max_se;
+
+ /* Force even distribution on all SIMDs in CU if the workgroup
+ * size is 64. This has shown some good improvements if # of
+ * CUs per SE is not a multiple of 4.
+ */
+ if (num_cu_per_se % 4 && waves_per_threadgroup == 1)
+ compute_resource_limits |= S_00B854_FORCE_SIMD_DIST(1);
+ }
+
+ radeon_set_sh_reg(cmd_buffer->cs, R_00B854_COMPUTE_RESOURCE_LIMITS,
+ compute_resource_limits);
+
radeon_set_sh_reg_seq(cmd_buffer->cs, R_00B81C_COMPUTE_NUM_THREAD_X, 3);
radeon_emit(cmd_buffer->cs,
S_00B81C_NUM_THREAD_FULL(compute_shader->info.cs.block_size[0]));
{
struct radv_pipeline *pipeline = cmd_buffer->state.compute_pipeline;
struct radv_shader_variant *compute_shader = pipeline->shaders[MESA_SHADER_COMPUTE];
+ unsigned dispatch_initiator = cmd_buffer->device->dispatch_initiator;
struct radeon_winsys *ws = cmd_buffer->device->ws;
struct radeon_winsys_cs *cs = cmd_buffer->cs;
struct ac_userdata_info *loc;
- unsigned dispatch_initiator;
- uint8_t grid_used;
-
- grid_used = compute_shader->info.info.cs.grid_components_used;
loc = radv_lookup_user_sgpr(pipeline, MESA_SHADER_COMPUTE,
AC_UD_CS_GRID_SIZE);
MAYBE_UNUSED unsigned cdw_max = radeon_check_space(ws, cs, 25);
- dispatch_initiator = S_00B800_COMPUTE_SHADER_EN(1) |
- S_00B800_FORCE_START_AT_000(1);
-
- if (cmd_buffer->device->physical_device->rad_info.chip_class >= CIK) {
- /* If the KMD allows it (there is a KMD hw register for it),
- * allow launching waves out-of-order.
- */
- dispatch_initiator |= S_00B800_ORDER_MODE(1);
- }
-
if (info->indirect) {
uint64_t va = radv_buffer_get_va(info->indirect->bo);
radv_cs_add_buffer(ws, cs, info->indirect->bo, 8);
if (loc->sgpr_idx != -1) {
- for (unsigned i = 0; i < grid_used; ++i) {
+ for (unsigned i = 0; i < 3; ++i) {
radeon_emit(cs, PKT3(PKT3_COPY_DATA, 4, 0));
radeon_emit(cs, COPY_DATA_SRC_SEL(COPY_DATA_MEM) |
COPY_DATA_DST_SEL(COPY_DATA_REG));
if (loc->sgpr_idx != -1) {
assert(!loc->indirect);
- assert(loc->num_sgprs == grid_used);
+ assert(loc->num_sgprs == 3);
radeon_set_sh_reg_seq(cs, R_00B900_COMPUTE_USER_DATA_0 +
- loc->sgpr_idx * 4, grid_used);
+ loc->sgpr_idx * 4, 3);
radeon_emit(cs, blocks[0]);
- if (grid_used > 1)
- radeon_emit(cs, blocks[1]);
- if (grid_used > 2)
- radeon_emit(cs, blocks[2]);
+ radeon_emit(cs, blocks[1]);
+ radeon_emit(cs, blocks[2]);
}
radeon_emit(cs, PKT3(PKT3_DISPATCH_DIRECT, 3, 0) |
projects / mesa.git / blobdiff