NIR_PASS(progress, shader, nir_remove_dead_variables,
nir_var_function_temp);
- NIR_PASS_V(shader, nir_lower_alu_to_scalar, NULL);
+ NIR_PASS_V(shader, nir_lower_alu_to_scalar, NULL, NULL);
NIR_PASS_V(shader, nir_lower_phis_to_scalar);
NIR_PASS(progress, shader, nir_copy_prop);
NIR_PASS_V(nir, nir_remove_dead_variables,
nir_var_shader_in | nir_var_shader_out | nir_var_system_value);
+ NIR_PASS_V(nir, nir_propagate_invariant);
+
NIR_PASS_V(nir, nir_lower_system_values);
NIR_PASS_V(nir, nir_lower_clip_cull_distance_arrays);
NIR_PASS_V(nir, radv_nir_lower_ycbcr_textures, layout);
*/
nir_lower_var_copies(nir);
+ /* Lower large variables that are always constant with load_constant
+ * intrinsics, which get turned into PC-relative loads from a data
+ * section next to the shader.
+ */
+ NIR_PASS_V(nir, nir_opt_large_constants,
+ glsl_get_natural_size_align_bytes, 16);
+
/* Indirect lowering must be called after the radv_optimize_nir() loop
* has been called at least once. Otherwise indirect lowering can
* bloat the instruction count of the loop and cause it to be
static void radv_postprocess_config(const struct radv_physical_device *pdevice,
const struct ac_shader_config *config_in,
- const struct radv_shader_variant_info *info,
+ const struct radv_shader_info *info,
gl_shader_stage stage,
struct ac_shader_config *config_out)
{
config_out->rsrc2 = S_00B12C_USER_SGPR(info->num_user_sgprs) |
S_00B12C_SCRATCH_EN(scratch_enabled) |
- S_00B12C_SO_BASE0_EN(!!info->info.so.strides[0]) |
- S_00B12C_SO_BASE1_EN(!!info->info.so.strides[1]) |
- S_00B12C_SO_BASE2_EN(!!info->info.so.strides[2]) |
- S_00B12C_SO_BASE3_EN(!!info->info.so.strides[3]) |
- S_00B12C_SO_EN(!!info->info.so.num_outputs);
+ S_00B12C_SO_BASE0_EN(!!info->so.strides[0]) |
+ S_00B12C_SO_BASE1_EN(!!info->so.strides[1]) |
+ S_00B12C_SO_BASE2_EN(!!info->so.strides[2]) |
+ S_00B12C_SO_BASE3_EN(!!info->so.strides[3]) |
+ S_00B12C_SO_EN(!!info->so.num_outputs);
config_out->rsrc1 = S_00B848_VGPRS((num_vgprs - 1) /
- (info->info.wave_size == 32 ? 8 : 4)) |
+ (info->wave_size == 32 ? 8 : 4)) |
S_00B848_DX10_CLAMP(1) |
S_00B848_FLOAT_MODE(config_out->float_mode);
config_out->rsrc2 |= S_00B22C_OC_LDS_EN(1);
} else if (info->tes.as_es) {
assert(pdevice->rad_info.chip_class <= GFX8);
- vgpr_comp_cnt = info->info.uses_prim_id ? 3 : 2;
+ vgpr_comp_cnt = info->uses_prim_id ? 3 : 2;
config_out->rsrc2 |= S_00B12C_OC_LDS_EN(1);
} else {
- bool enable_prim_id = info->tes.export_prim_id || info->info.uses_prim_id;
+ bool enable_prim_id = info->tes.export_prim_id || info->uses_prim_id;
vgpr_comp_cnt = enable_prim_id ? 3 : 2;
config_out->rsrc1 |= S_00B128_MEM_ORDERED(pdevice->rad_info.chip_class >= GFX10);
* StepRate0 is set to 1. so that VGPR3 doesn't have to be loaded.
*/
if (pdevice->rad_info.chip_class >= GFX10) {
- vgpr_comp_cnt = info->info.vs.needs_instance_id ? 3 : 1;
+ vgpr_comp_cnt = info->vs.needs_instance_id ? 3 : 1;
} else {
- vgpr_comp_cnt = info->info.vs.needs_instance_id ? 2 : 1;
+ vgpr_comp_cnt = info->vs.needs_instance_id ? 2 : 1;
}
} else {
config_out->rsrc2 |= S_00B12C_OC_LDS_EN(1);
* VGPR0-3: (VertexID, RelAutoindex, InstanceID / StepRate0, InstanceID).
* StepRate0 is set to 1. so that VGPR3 doesn't have to be loaded.
*/
- vgpr_comp_cnt = info->info.vs.needs_instance_id ? 2 : 1;
+ vgpr_comp_cnt = info->vs.needs_instance_id ? 2 : 1;
} else if (info->vs.as_es) {
assert(pdevice->rad_info.chip_class <= GFX8);
/* VGPR0-3: (VertexID, InstanceID / StepRate0, ...) */
- vgpr_comp_cnt = info->info.vs.needs_instance_id ? 1 : 0;
+ vgpr_comp_cnt = info->vs.needs_instance_id ? 1 : 0;
} else {
/* VGPR0-3: (VertexID, InstanceID / StepRate0, PrimID, InstanceID)
* If PrimID is disabled. InstanceID / StepRate1 is loaded instead.
* StepRate0 is set to 1. so that VGPR3 doesn't have to be loaded.
*/
- if (info->info.vs.needs_instance_id && pdevice->rad_info.chip_class >= GFX10) {
+ if (info->vs.needs_instance_id && pdevice->rad_info.chip_class >= GFX10) {
vgpr_comp_cnt = 3;
} else if (info->vs.export_prim_id) {
vgpr_comp_cnt = 2;
- } else if (info->info.vs.needs_instance_id) {
+ } else if (info->vs.needs_instance_id) {
vgpr_comp_cnt = 1;
} else {
vgpr_comp_cnt = 0;
config_out->rsrc1 |= S_00B848_MEM_ORDERED(pdevice->rad_info.chip_class >= GFX10) |
S_00B848_WGP_MODE(pdevice->rad_info.chip_class >= GFX10);
config_out->rsrc2 |=
- S_00B84C_TGID_X_EN(info->info.cs.uses_block_id[0]) |
- S_00B84C_TGID_Y_EN(info->info.cs.uses_block_id[1]) |
- S_00B84C_TGID_Z_EN(info->info.cs.uses_block_id[2]) |
- S_00B84C_TIDIG_COMP_CNT(info->info.cs.uses_thread_id[2] ? 2 :
- info->info.cs.uses_thread_id[1] ? 1 : 0) |
- S_00B84C_TG_SIZE_EN(info->info.cs.uses_local_invocation_idx) |
+ S_00B84C_TGID_X_EN(info->cs.uses_block_id[0]) |
+ S_00B84C_TGID_Y_EN(info->cs.uses_block_id[1]) |
+ S_00B84C_TGID_Z_EN(info->cs.uses_block_id[2]) |
+ S_00B84C_TIDIG_COMP_CNT(info->cs.uses_thread_id[2] ? 2 :
+ info->cs.uses_thread_id[1] ? 1 : 0) |
+ S_00B84C_TG_SIZE_EN(info->cs.uses_local_invocation_idx) |
S_00B84C_LDS_SIZE(config_in->lds_size);
break;
default:
/* VGPR5-8: (VertexID, UserVGPR0, UserVGPR1, UserVGPR2 / InstanceID) */
if (es_stage == MESA_SHADER_VERTEX) {
- es_vgpr_comp_cnt = info->info.vs.needs_instance_id ? 3 : 0;
+ es_vgpr_comp_cnt = info->vs.needs_instance_id ? 3 : 0;
} else if (es_stage == MESA_SHADER_TESS_EVAL) {
- bool enable_prim_id = info->tes.export_prim_id || info->info.uses_prim_id;
+ bool enable_prim_id = info->tes.export_prim_id || info->uses_prim_id;
es_vgpr_comp_cnt = enable_prim_id ? 3 : 2;
} else
unreachable("Unexpected ES shader stage");
bool tes_triangles = stage == MESA_SHADER_TESS_EVAL &&
info->tes.primitive_mode >= 4; /* GL_TRIANGLES */
- if (info->info.uses_invocation_id || stage == MESA_SHADER_VERTEX) {
+ if (info->uses_invocation_id || stage == MESA_SHADER_VERTEX) {
gs_vgpr_comp_cnt = 3; /* VGPR3 contains InvocationID. */
- } else if (info->info.uses_prim_id) {
+ } else if (info->uses_prim_id) {
gs_vgpr_comp_cnt = 2; /* VGPR2 contains PrimitiveID. */
} else if (info->gs.vertices_in >= 3 || tes_triangles) {
gs_vgpr_comp_cnt = 1; /* VGPR1 contains offsets 2, 3 */
if (es_type == MESA_SHADER_VERTEX) {
/* VGPR0-3: (VertexID, InstanceID / StepRate0, ...) */
- if (info->info.vs.needs_instance_id) {
+ if (info->vs.needs_instance_id) {
es_vgpr_comp_cnt = pdevice->rad_info.chip_class >= GFX10 ? 3 : 1;
} else {
es_vgpr_comp_cnt = 0;
}
} else if (es_type == MESA_SHADER_TESS_EVAL) {
- es_vgpr_comp_cnt = info->info.uses_prim_id ? 3 : 2;
+ es_vgpr_comp_cnt = info->uses_prim_id ? 3 : 2;
} else {
unreachable("invalid shader ES type");
}
/* If offsets 4, 5 are used, GS_VGPR_COMP_CNT is ignored and
* VGPR[0:4] are always loaded.
*/
- if (info->info.uses_invocation_id) {
+ if (info->uses_invocation_id) {
gs_vgpr_comp_cnt = 3; /* VGPR3 contains InvocationID. */
- } else if (info->info.uses_prim_id) {
+ } else if (info->uses_prim_id) {
gs_vgpr_comp_cnt = 2; /* VGPR2 contains PrimitiveID. */
} else if (info->gs.vertices_in >= 3) {
gs_vgpr_comp_cnt = 1; /* VGPR1 contains offsets 2, 3 */
esgs_ring_size = 32 * 1024;
}
- if (binary->variant_info.is_ngg) {
+ if (binary->info.is_ngg) {
/* GS stores Primitive IDs into LDS at the address
* corresponding to the ES thread of the provoking
* vertex. All ES threads load and export PrimitiveID
* for their thread.
*/
if (binary->stage == MESA_SHADER_VERTEX &&
- binary->variant_info.vs.export_prim_id) {
+ binary->info.vs.export_prim_id) {
/* TODO: Do not harcode this value */
esgs_ring_size = 256 /* max_out_verts */ * 4;
}
/* Make sure to have LDS space for NGG scratch. */
/* TODO: Compute this correctly somehow? */
- if (binary->variant_info.is_ngg)
+ if (binary->info.is_ngg)
sym->size -= 32;
}
struct ac_rtld_open_info open_info = {
.info = &device->physical_device->rad_info,
.shader_type = binary->stage,
- .wave_size = binary->variant_info.info.wave_size,
+ .wave_size = binary->info.wave_size,
.num_parts = 1,
.elf_ptrs = &elf_data,
.elf_sizes = &elf_size,
variant->exec_size = variant->code_size;
}
- variant->info = binary->variant_info;
- radv_postprocess_config(device->physical_device, &config, &binary->variant_info,
+ variant->info = binary->info;
+ radv_postprocess_config(device->physical_device, &config, &binary->info,
binary->stage, &variant->config);
void *dest_ptr = radv_alloc_shader_memory(device, variant);
struct nir_shader * const *shaders,
int shader_count,
gl_shader_stage stage,
+ struct radv_shader_info *info,
struct radv_nir_compiler_options *options,
bool gs_copy_shader,
bool keep_shader_info,
enum ac_target_machine_options tm_options = 0;
struct ac_llvm_compiler ac_llvm;
struct radv_shader_binary *binary = NULL;
- struct radv_shader_variant_info variant_info = {0};
bool thread_compiler;
options->family = chip_family;
if (gs_copy_shader) {
assert(shader_count == 1);
radv_compile_gs_copy_shader(&ac_llvm, *shaders, &binary,
- &variant_info, options);
+ info, options);
} else {
- radv_compile_nir_shader(&ac_llvm, &binary, &variant_info,
+ radv_compile_nir_shader(&ac_llvm, &binary, info,
shaders, shader_count, options);
}
- binary->variant_info = variant_info;
+ binary->info = *info;
radv_destroy_llvm_compiler(&ac_llvm, thread_compiler);
int shader_count,
struct radv_pipeline_layout *layout,
const struct radv_shader_variant_key *key,
+ struct radv_shader_info *info,
bool keep_shader_info,
struct radv_shader_binary **binary_out)
{
options.supports_spill = true;
options.robust_buffer_access = device->robust_buffer_access;
- return shader_variant_compile(device, module, shaders, shader_count, shaders[shader_count - 1]->info.stage,
+ return shader_variant_compile(device, module, shaders, shader_count, shaders[shader_count - 1]->info.stage, info,
&options, false, keep_shader_info, binary_out);
}
struct radv_shader_variant *
radv_create_gs_copy_shader(struct radv_device *device,
struct nir_shader *shader,
+ struct radv_shader_info *info,
struct radv_shader_binary **binary_out,
bool keep_shader_info,
bool multiview)
options.key.has_multiview_view_index = multiview;
return shader_variant_compile(device, NULL, &shader, 1, MESA_SHADER_VERTEX,
- &options, true, keep_shader_info, binary_out);
+ info, &options, true, keep_shader_info, binary_out);
}
void
}
const char *
-radv_get_shader_name(struct radv_shader_variant_info *info,
+radv_get_shader_name(struct radv_shader_info *info,
gl_shader_stage stage)
{
switch (stage) {
{
enum chip_class chip_class = device->physical_device->rad_info.chip_class;
unsigned lds_increment = chip_class >= GFX7 ? 512 : 256;
- uint8_t wave_size = variant->info.info.wave_size;
+ uint8_t wave_size = variant->info.wave_size;
struct ac_shader_config *conf = &variant->config;
unsigned max_simd_waves;
unsigned lds_per_wave = 0;
- max_simd_waves = ac_get_max_simd_waves(device->physical_device->rad_info.family);
+ max_simd_waves = ac_get_max_wave64_per_simd(device->physical_device->rad_info.family);
if (stage == MESA_SHADER_FRAGMENT) {
lds_per_wave = conf->lds_size * lds_increment +
- align(variant->info.info.ps.num_interp * 48,
+ align(variant->info.ps.num_interp * 48,
lds_increment);
} else if (stage == MESA_SHADER_COMPUTE) {
unsigned max_workgroup_size =
if (conf->num_sgprs)
max_simd_waves =
MIN2(max_simd_waves,
- ac_get_num_physical_sgprs(chip_class) / conf->num_sgprs);
+ ac_get_num_physical_sgprs(&device->physical_device->rad_info) /
+ conf->num_sgprs);
if (conf->num_vgprs)
max_simd_waves =
VkShaderStatisticsInfoAMD statistics = {};
statistics.shaderStageMask = shaderStage;
statistics.numPhysicalVgprs = RADV_NUM_PHYSICAL_VGPRS;
- statistics.numPhysicalSgprs = ac_get_num_physical_sgprs(device->physical_device->rad_info.chip_class);
+ statistics.numPhysicalSgprs = ac_get_num_physical_sgprs(&device->physical_device->rad_info);
statistics.numAvailableSgprs = statistics.numPhysicalSgprs;
if (stage == MESA_SHADER_COMPUTE) {