[PATCH vkd3d 7/7] vkd3d-shader: Decorate SPIR-V for nonuniform access where flagged in dxbc.
Conor McCarthy
cmccarthy at codeweavers.com
Fri Aug 13 09:55:43 CDT 2021
Based on vkd3d-proton patches by Philip Rebohle and Hans-Kristian Arntzen.
Signed-off-by: Conor McCarthy <cmccarthy at codeweavers.com>
---
libs/vkd3d-shader/dxbc.c | 7 ++++
libs/vkd3d-shader/sm4.h | 1 +
libs/vkd3d-shader/spirv.c | 42 ++++++++++++++++++++++--
libs/vkd3d-shader/vkd3d_shader_private.h | 7 ++++
tests/d3d12.c | 3 +-
5 files changed, 55 insertions(+), 5 deletions(-)
diff --git a/libs/vkd3d-shader/dxbc.c b/libs/vkd3d-shader/dxbc.c
index 412b9267..9371e262 100644
--- a/libs/vkd3d-shader/dxbc.c
+++ b/libs/vkd3d-shader/dxbc.c
@@ -1121,6 +1121,7 @@ static bool shader_sm4_read_param(struct vkd3d_sm4_data *priv, const DWORD **ptr
{
param->type = register_type_table[register_type];
}
+ param->modifier = VKD3DSPRM_NONE;
param->data_type = data_type;
if (token & VKD3D_SM4_REGISTER_MODIFIER)
@@ -1155,6 +1156,12 @@ static bool shader_sm4_read_param(struct vkd3d_sm4_data *priv, const DWORD **ptr
}
m &= ~(VKD3D_SM4_REGISTER_MODIFIER_ABS | VKD3D_SM4_REGISTER_MODIFIER_NEGATE);
+ if (m & VKD3D_SM4_REGISTER_MODIFIER_NONUNIFORM)
+ {
+ param->modifier = VKD3DSPRM_NONUNIFORM;
+ m &= ~VKD3D_SM4_REGISTER_MODIFIER_NONUNIFORM;
+ }
+
if (m)
FIXME("Skipping modifier flags 0x%08x.\n", m);
diff --git a/libs/vkd3d-shader/sm4.h b/libs/vkd3d-shader/sm4.h
index e8dbeb4b..8dbbeb3d 100644
--- a/libs/vkd3d-shader/sm4.h
+++ b/libs/vkd3d-shader/sm4.h
@@ -387,6 +387,7 @@ enum vkd3d_sm4_register_modifier
{
VKD3D_SM4_REGISTER_MODIFIER_NEGATE = 0x20,
VKD3D_SM4_REGISTER_MODIFIER_ABS = 0x40,
+ VKD3D_SM4_REGISTER_MODIFIER_NONUNIFORM = 0x10000,
};
enum vkd3d_sm4_output_primitive_type
diff --git a/libs/vkd3d-shader/spirv.c b/libs/vkd3d-shader/spirv.c
index 4e67af49..a0c60972 100644
--- a/libs/vkd3d-shader/spirv.c
+++ b/libs/vkd3d-shader/spirv.c
@@ -1842,7 +1842,8 @@ static bool vkd3d_spirv_compile_module(struct vkd3d_spirv_builder *builder,
|| vkd3d_spirv_capability_is_enabled(builder, SpvCapabilitySampledImageArrayDynamicIndexing)
|| vkd3d_spirv_capability_is_enabled(builder, SpvCapabilityStorageBufferArrayDynamicIndexing)
|| vkd3d_spirv_capability_is_enabled(builder, SpvCapabilityStorageTexelBufferArrayDynamicIndexingEXT)
- || vkd3d_spirv_capability_is_enabled(builder, SpvCapabilityStorageImageArrayDynamicIndexing))
+ || vkd3d_spirv_capability_is_enabled(builder, SpvCapabilityStorageImageArrayDynamicIndexing)
+ || vkd3d_spirv_capability_is_enabled(builder, SpvCapabilityShaderNonUniformEXT))
vkd3d_spirv_build_op_extension(&stream, "SPV_EXT_descriptor_indexing");
if (builder->ext_instr_set_glsl_450)
@@ -2675,6 +2676,15 @@ static void vkd3d_dxbc_compiler_emit_descriptor_binding_for_reg(struct vkd3d_dxb
vkd3d_dxbc_compiler_emit_descriptor_binding(compiler, variable_id, &binding);
}
+static void vkd3d_dxbc_compiler_decorate_nonuniform(struct vkd3d_dxbc_compiler *compiler,
+ uint32_t expression_id)
+{
+ struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
+
+ vkd3d_spirv_enable_capability(builder, SpvCapabilityShaderNonUniformEXT);
+ vkd3d_spirv_build_op_decorate(builder, expression_id, SpvDecorationNonUniformEXT, NULL, 0);
+}
+
static const struct vkd3d_symbol *vkd3d_dxbc_compiler_put_symbol(struct vkd3d_dxbc_compiler *compiler,
const struct vkd3d_symbol *symbol)
{
@@ -3242,6 +3252,9 @@ static uint32_t vkd3d_dxbc_compiler_get_descriptor_index(struct vkd3d_dxbc_compi
index.offset -= binding_base_idx;
index_id = vkd3d_dxbc_compiler_emit_register_addressing(compiler, &index);
+ /* AMD drivers rely on the index being marked as nonuniform */
+ if (reg->modifier == VKD3DSPRM_NONUNIFORM)
+ vkd3d_dxbc_compiler_decorate_nonuniform(compiler, index_id);
return index_id;
}
@@ -3319,6 +3332,8 @@ static void vkd3d_dxbc_compiler_emit_dereference_register(struct vkd3d_dxbc_comp
ptr_type_id = vkd3d_spirv_get_op_type_pointer(builder, register_info->storage_class, type_id);
register_info->id = vkd3d_spirv_build_op_access_chain(builder, ptr_type_id,
register_info->id, indexes, index_count);
+ if (reg->modifier == VKD3DSPRM_NONUNIFORM)
+ vkd3d_dxbc_compiler_decorate_nonuniform(compiler, register_info->id);
}
}
@@ -7992,8 +8007,16 @@ static void vkd3d_dxbc_compiler_prepare_image(struct vkd3d_dxbc_compiler *compil
return;
}
- image->image_id = load ? vkd3d_spirv_build_op_load(builder,
- image->image_type_id, image->id, SpvMemoryAccessMaskNone) : 0;
+ if (load)
+ {
+ image->image_id = vkd3d_spirv_build_op_load(builder, image->image_type_id, image->id, SpvMemoryAccessMaskNone);
+ if (resource_reg->modifier == VKD3DSPRM_NONUNIFORM)
+ vkd3d_dxbc_compiler_decorate_nonuniform(compiler, image->image_id);
+ }
+ else
+ {
+ image->image_id = 0;
+ }
image->image_type_id = vkd3d_dxbc_compiler_get_image_type_id(compiler, resource_reg,
&symbol->info.resource.range, image->resource_type_info,
@@ -8022,9 +8045,14 @@ static void vkd3d_dxbc_compiler_prepare_image(struct vkd3d_dxbc_compiler *compil
sampler_id = vkd3d_spirv_build_op_load(builder,
vkd3d_spirv_get_op_type_sampler(builder), sampler_var_id, SpvMemoryAccessMaskNone);
+ if (sampler_reg->modifier == VKD3DSPRM_NONUNIFORM)
+ vkd3d_dxbc_compiler_decorate_nonuniform(compiler, sampler_id);
+
sampled_image_type_id = vkd3d_spirv_get_op_type_sampled_image(builder, image->image_type_id);
image->sampled_image_id = vkd3d_spirv_build_op_sampled_image(builder,
sampled_image_type_id, image->image_id, sampler_id);
+ if (resource_reg->modifier == VKD3DSPRM_NONUNIFORM)
+ vkd3d_dxbc_compiler_decorate_nonuniform(compiler, image->sampled_image_id);
}
else
{
@@ -8375,6 +8403,8 @@ static void vkd3d_dxbc_compiler_emit_ld_raw_structured_srv_uav(struct vkd3d_dxbc
indices[1] = coordinate_id;
ptr_id = vkd3d_spirv_build_op_access_chain(builder, ptr_type_id, resource_symbol->id, indices, 2);
+ if (resource->reg.modifier == VKD3DSPRM_NONUNIFORM)
+ vkd3d_dxbc_compiler_decorate_nonuniform(compiler, ptr_id);
constituents[j++] = vkd3d_spirv_build_op_load(builder, texel_type_id, ptr_id, SpvMemoryAccessMaskNone);
}
}
@@ -8517,6 +8547,8 @@ static void vkd3d_dxbc_compiler_emit_store_uav_raw_structured(struct vkd3d_dxbc_
indices[1] = coordinate_id;
ptr_id = vkd3d_spirv_build_op_access_chain(builder, ptr_type_id, resource_symbol->id, indices, 2);
+ if (dst->reg.modifier == VKD3DSPRM_NONUNIFORM)
+ vkd3d_dxbc_compiler_decorate_nonuniform(compiler, ptr_id);
vkd3d_spirv_build_op_store(builder, ptr_id, data_id, SpvMemoryAccessMaskNone);
}
}
@@ -8894,6 +8926,8 @@ static void vkd3d_dxbc_compiler_emit_atomic_instruction(struct vkd3d_dxbc_compil
pointer_id = vkd3d_spirv_build_op_image_texel_pointer(builder,
ptr_type_id, image.id, coordinate_id, sample_id);
}
+ if (resource->reg.modifier == VKD3DSPRM_NONUNIFORM)
+ vkd3d_dxbc_compiler_decorate_nonuniform(compiler, pointer_id);
}
val_id = vkd3d_dxbc_compiler_emit_load_src_with_type(compiler, &src[1], VKD3DSP_WRITEMASK_0, component_type);
@@ -8942,6 +8976,8 @@ static void vkd3d_dxbc_compiler_emit_bufinfo(struct vkd3d_dxbc_compiler *compile
type_id = vkd3d_spirv_get_type_id(builder, VKD3D_SHADER_COMPONENT_UINT, 1);
val_id = vkd3d_spirv_build_op_image_query_size(builder, type_id, image.image_id);
+ if (src->reg.modifier == VKD3DSPRM_NONUNIFORM)
+ vkd3d_dxbc_compiler_decorate_nonuniform(compiler, image.id);
write_mask = VKD3DSP_WRITEMASK_0;
}
diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h
index 9948045e..a70cab59 100644
--- a/libs/vkd3d-shader/vkd3d_shader_private.h
+++ b/libs/vkd3d-shader/vkd3d_shader_private.h
@@ -455,6 +455,12 @@ enum vkd3d_immconst_type
VKD3D_IMMCONST_VEC4,
};
+enum vkd3d_shader_register_modifier
+{
+ VKD3DSPRM_NONE = 0,
+ VKD3DSPRM_NONUNIFORM = 1,
+};
+
enum vkd3d_shader_src_modifier
{
VKD3DSPSM_NONE = 0,
@@ -612,6 +618,7 @@ struct vkd3d_shader_register_index
struct vkd3d_shader_register
{
enum vkd3d_shader_register_type type;
+ enum vkd3d_shader_register_modifier modifier;
enum vkd3d_data_type data_type;
struct vkd3d_shader_register_index idx[3];
enum vkd3d_immconst_type immconst_type;
diff --git a/tests/d3d12.c b/tests/d3d12.c
index bd3ef624..ecda1272 100644
--- a/tests/d3d12.c
+++ b/tests/d3d12.c
@@ -34960,7 +34960,7 @@ static void test_unbounded_resource_arrays(void)
D3D12_RESOURCE_STATE_UNORDERED_ACCESS, D3D12_RESOURCE_STATE_COPY_SOURCE);
get_buffer_readback_with_command_list(output_buffers[i], DXGI_FORMAT_R32_UINT, &rb, queue, command_list);
/* Buffers at index >= 64 are aliased. */
- todo_if(i != 10 && i != 74)
+ todo_if(i != 74)
check_readback_data_uint(&rb, NULL, (i < 64 ? 63 - i : 127 - i) ^ 0x35, 0);
release_resource_readback(&rb);
reset_command_list(command_list, context.allocator);
@@ -35107,7 +35107,6 @@ static void test_unbounded_samplers(void)
{
unsigned int value = get_readback_uint(&rb, i, 0, 0);
unsigned int expected = (i & 1) ? 100 : 10;
- todo_if(i & 1)
ok(value == expected, "Got %u, expected %u at %u.\n", value, expected, i);
}
release_resource_readback(&rb);
--
2.32.0
More information about the wine-devel
mailing list