assert(program->ir_type == PIPE_SHADER_IR_NIR);
si_nir_scan_shader(sel->nir, &sel->info);
- /* Store the declared LDS size into si_shader_info for the shader
- * cache to include it.
- */
- sel->info.properties[TGSI_PROPERTY_CS_LOCAL_SIZE] = program->local_size;
-
+ sel->info.base.cs.shared_size = program->local_size;
si_get_active_slot_masks(&sel->info, &sel->active_const_and_shader_buffers,
&sel->active_samplers_and_images);
program->shader.is_monolithic = true;
program->reads_variable_block_size =
- sel->info.uses_block_size && sel->info.properties[TGSI_PROPERTY_CS_FIXED_BLOCK_WIDTH] == 0;
+ sel->info.uses_block_size && sel->info.base.cs.local_size[0] == 0;
program->num_cs_user_data_dwords =
- sel->info.properties[TGSI_PROPERTY_CS_USER_DATA_COMPONENTS_AMD];
+ sel->info.base.cs.user_data_components_amd;
unsigned user_sgprs = SI_NUM_RESOURCE_SGPRS + (sel->info.uses_grid_size ? 3 : 0) +
(program->reads_variable_block_size ? 3 : 0) +
/* Fast path for compute shaders - some descriptors passed via user SGPRs. */
/* Shader buffers in user SGPRs. */
- for (unsigned i = 0; i < 3 && user_sgprs <= 12 && sel->info.shader_buffers_declared & (1 << i); i++) {
+ for (unsigned i = 0; i < MIN2(3, sel->info.base.num_ssbos) && user_sgprs <= 12; i++) {
user_sgprs = align(user_sgprs, 4);
if (i == 0)
sel->cs_shaderbufs_sgpr_index = user_sgprs;
}
/* Images in user SGPRs. */
- unsigned non_msaa_images = sel->info.images_declared & ~sel->info.msaa_images_declared;
+ unsigned non_msaa_images = u_bit_consecutive(0, sel->info.base.num_images) &
+ ~sel->info.base.msaa_images;
for (unsigned i = 0; i < 3 && non_msaa_images & (1 << i); i++) {
- unsigned num_sgprs = sel->info.image_buffers & (1 << i) ? 4 : 8;
+ unsigned num_sgprs = sel->info.base.image_buffers & (1 << i) ? 4 : 8;
if (align(user_sgprs, num_sgprs) + num_sgprs > 16)
break;
struct si_shader_selector *sel = &program->sel;
pipe_reference_init(&sel->base.reference, 1);
- sel->type = PIPE_SHADER_COMPUTE;
+ sel->info.stage = MESA_SHADER_COMPUTE;
sel->screen = sscreen;
+ sel->const_and_shader_buf_descriptors_index =
+ si_const_and_shader_buffer_descriptors_idx(PIPE_SHADER_COMPUTE);
+ sel->sampler_and_images_descriptors_index =
+ si_sampler_and_image_descriptors_idx(PIPE_SHADER_COMPUTE);
program->shader.selector = &program->sel;
program->ir_type = cso->ir_type;
program->local_size = cso->req_local_mem;
sel->compiler_ctx_state.is_debug_context = sctx->is_debug;
p_atomic_inc(&sscreen->num_shaders_created);
- si_schedule_initial_compile(sctx, PIPE_SHADER_COMPUTE, &sel->ready, &sel->compiler_ctx_state,
+ si_schedule_initial_compile(sctx, MESA_SHADER_COMPUTE, &sel->ready, &sel->compiler_ctx_state,
program, si_create_compute_state_async);
} else {
const struct pipe_binary_program_header *header;
void si_emit_initial_compute_regs(struct si_context *sctx, struct radeon_cmdbuf *cs)
{
- uint64_t bc_va;
+ uint64_t bc_va = sctx->border_color_buffer->gpu_address;
radeon_set_sh_reg_seq(cs, R_00B858_COMPUTE_STATIC_THREAD_MGMT_SE0, 2);
/* R_00B858_COMPUTE_STATIC_THREAD_MGMT_SE0 / SE1,
radeon_emit(cs, S_00B858_SH0_CU_EN(0xffff) | S_00B858_SH1_CU_EN(0xffff));
radeon_emit(cs, S_00B858_SH0_CU_EN(0xffff) | S_00B858_SH1_CU_EN(0xffff));
+ if (sctx->chip_class == GFX6) {
+ /* This register has been moved to R_00CD20_COMPUTE_MAX_WAVE_ID
+ * and is now per pipe, so it should be handled in the
+ * kernel if we want to use something other than the default value.
+ *
+ * TODO: This should be:
+ * (number of compute units) * 4 * (waves per simd) - 1
+ */
+ radeon_set_sh_reg(cs, R_00B82C_COMPUTE_MAX_WAVE_ID, 0x190 /* Default value */);
+
+ if (sctx->screen->info.si_TA_CS_BC_BASE_ADDR_allowed)
+ radeon_set_config_reg(cs, R_00950C_TA_CS_BC_BASE_ADDR, bc_va >> 8);
+ }
+
if (sctx->chip_class >= GFX7) {
/* Also set R_00B858_COMPUTE_STATIC_THREAD_MGMT_SE2 / SE3 */
radeon_set_sh_reg_seq(cs, R_00B864_COMPUTE_STATIC_THREAD_MGMT_SE2, 2);
radeon_set_sh_reg(cs, R_00B82C_COMPUTE_PERFCOUNT_ENABLE, 0);
radeon_set_sh_reg(cs, R_00B878_COMPUTE_THREAD_TRACE_ENABLE, 0);
}
- }
- if (sctx->chip_class >= GFX10) {
- radeon_set_sh_reg(cs, R_00B890_COMPUTE_USER_ACCUM_0, 0);
- radeon_set_sh_reg(cs, R_00B894_COMPUTE_USER_ACCUM_1, 0);
- radeon_set_sh_reg(cs, R_00B898_COMPUTE_USER_ACCUM_2, 0);
- radeon_set_sh_reg(cs, R_00B89C_COMPUTE_USER_ACCUM_3, 0);
- radeon_set_sh_reg(cs, R_00B8A0_COMPUTE_PGM_RSRC3, 0);
- radeon_set_sh_reg(cs, R_00B9F4_COMPUTE_DISPATCH_TUNNEL, 0);
- }
-
- /* This register has been moved to R_00CD20_COMPUTE_MAX_WAVE_ID
- * and is now per pipe, so it should be handled in the
- * kernel if we want to use something other than the default value,
- * which is now 0x22f.
- */
- if (sctx->chip_class <= GFX6) {
- /* XXX: This should be:
- * (number of compute units) * 4 * (waves per simd) - 1 */
-
- radeon_set_sh_reg(cs, R_00B82C_COMPUTE_MAX_WAVE_ID, 0x190 /* Default value */);
- }
-
- /* Set the pointer to border colors. */
- bc_va = sctx->border_color_buffer->gpu_address;
-
- if (sctx->chip_class >= GFX7) {
+ /* Set the pointer to border colors. */
radeon_set_uconfig_reg_seq(cs, R_030E00_TA_CS_BC_BASE_ADDR, 2);
radeon_emit(cs, bc_va >> 8); /* R_030E00_TA_CS_BC_BASE_ADDR */
radeon_emit(cs, S_030E04_ADDRESS(bc_va >> 40)); /* R_030E04_TA_CS_BC_BASE_ADDR_HI */
- } else {
- if (sctx->screen->info.si_TA_CS_BC_BASE_ADDR_allowed) {
- radeon_set_config_reg(cs, R_00950C_TA_CS_BC_BASE_ADDR, bc_va >> 8);
- }
}
/* cs_preamble_state initializes this for the gfx queue, so only do this
radeon_set_uconfig_reg(cs, R_0301EC_CP_COHER_START_DELAY,
sctx->chip_class >= GFX10 ? 0x20 : 0);
}
+
+ if (sctx->chip_class >= GFX10) {
+ radeon_set_sh_reg(cs, R_00B890_COMPUTE_USER_ACCUM_0, 0);
+ radeon_set_sh_reg(cs, R_00B894_COMPUTE_USER_ACCUM_1, 0);
+ radeon_set_sh_reg(cs, R_00B898_COMPUTE_USER_ACCUM_2, 0);
+ radeon_set_sh_reg(cs, R_00B89C_COMPUTE_USER_ACCUM_3, 0);
+ radeon_set_sh_reg(cs, R_00B8A0_COMPUTE_PGM_RSRC3, 0);
+ radeon_set_sh_reg(cs, R_00B9F4_COMPUTE_DISPATCH_TUNNEL, 0);
+ }
}
static bool si_setup_compute_scratch_buffer(struct si_context *sctx, struct si_shader *shader,