eden-emu
/
eden
mirror of https://git.eden-emu.dev/eden-emu/eden.git
1
0
Fork 0
Code Releases Wiki Activity
Browse Source

Revert "[TEST] Force checks on subgroups"

lsfg-android
CamilleLaVey 4 weeks ago
parent
ace66343ce
commit
ca9ff1642f
8 changed files with 7 additions and 84 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 2
      src/shader_recompiler/backend/spirv/emit_spirv.cpp
  2. 42
      src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp
  3. 9
      src/shader_recompiler/backend/spirv/spirv_emit_context.cpp
  4. 7
      src/shader_recompiler/profile.h
  5. 15
      src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
  6. 9
      src/video_core/renderer_vulkan/vk_texture_cache.cpp
  7. 2
      src/video_core/vulkan_common/vulkan_device.cpp
  8. 5
      src/video_core/vulkan_common/vulkan_device.h

2
src/shader_recompiler/backend/spirv/emit_spirv.cpp
View File

@ -435,7 +435,7 @@ void SetupCapabilities(const Profile& profile, const Info& info, EmitContext& ct
}
if ((info.uses_subgroup_vote || info.uses_subgroup_invocation_id ||
info.uses_subgroup_shuffles) &&
profile.support_vote && profile.SupportsSubgroupStage(ctx.stage)) {
profile.support_vote) {
ctx.AddCapability(spv::Capability::GroupNonUniformBallot);
ctx.AddCapability(spv::Capability::GroupNonUniformShuffle);
if (!profile.warp_size_potentially_larger_than_guest) {

42
src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp
View File

@ -1,6 +1,3 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
@ -13,14 +10,7 @@ Id SubgroupScope(EmitContext& ctx) {
return ctx.Const(static_cast<u32>(spv::Scope::Subgroup));
}
bool StageSupportsSubgroups(EmitContext& ctx) {
return ctx.profile.SupportsSubgroupStage(ctx.stage);
}
Id GetThreadId(EmitContext& ctx) {
if (!StageSupportsSubgroups(ctx)) {
return ctx.u32_zero_value;
}
return ctx.OpLoad(ctx.U32[1], ctx.subgroup_local_invocation_id);
}
@ -78,9 +68,6 @@ Id GetMaxThreadId(EmitContext& ctx, Id thread_id, Id clamp, Id segmentation_mask
}
Id SelectValue(EmitContext& ctx, Id in_range, Id value, Id src_thread_id) {
if (!StageSupportsSubgroups(ctx)) {
return value;
}
return ctx.OpSelect(
ctx.U32[1], in_range,
ctx.OpGroupNonUniformShuffle(ctx.U32[1], SubgroupScope(ctx), value, src_thread_id), value);
@ -102,9 +89,6 @@ Id EmitLaneId(EmitContext& ctx) {
}
Id EmitVoteAll(EmitContext& ctx, Id pred) {
if (!StageSupportsSubgroups(ctx)) {
return pred;
}
if (!ctx.profile.warp_size_potentially_larger_than_guest) {
return ctx.OpGroupNonUniformAll(ctx.U1, SubgroupScope(ctx), pred);
}
@ -118,9 +102,6 @@ Id EmitVoteAll(EmitContext& ctx, Id pred) {
}
Id EmitVoteAny(EmitContext& ctx, Id pred) {
if (!StageSupportsSubgroups(ctx)) {
return pred;
}
if (!ctx.profile.warp_size_potentially_larger_than_guest) {
return ctx.OpGroupNonUniformAny(ctx.U1, SubgroupScope(ctx), pred);
}
@ -134,9 +115,6 @@ Id EmitVoteAny(EmitContext& ctx, Id pred) {
}
Id EmitVoteEqual(EmitContext& ctx, Id pred) {
if (!StageSupportsSubgroups(ctx)) {
return ctx.true_value;
}
if (!ctx.profile.warp_size_potentially_larger_than_guest) {
return ctx.OpGroupNonUniformAllEqual(ctx.U1, SubgroupScope(ctx), pred);
}
@ -151,9 +129,6 @@ Id EmitVoteEqual(EmitContext& ctx, Id pred) {
}
Id EmitSubgroupBallot(EmitContext& ctx, Id pred) {
if (!StageSupportsSubgroups(ctx)) {
return ctx.OpSelect(ctx.U32[1], pred, ctx.Const(1u), ctx.u32_zero_value);
}
const Id ballot{ctx.OpGroupNonUniformBallot(ctx.U32[4], SubgroupScope(ctx), pred)};
if (!ctx.profile.warp_size_potentially_larger_than_guest) {
return ctx.OpCompositeExtract(ctx.U32[1], ballot, 0U);
@ -162,37 +137,22 @@ Id EmitSubgroupBallot(EmitContext& ctx, Id pred) {
}
Id EmitSubgroupEqMask(EmitContext& ctx) {
if (!StageSupportsSubgroups(ctx)) {
return ctx.Const(1u);
}
return LoadMask(ctx, ctx.subgroup_mask_eq);
}
Id EmitSubgroupLtMask(EmitContext& ctx) {
if (!StageSupportsSubgroups(ctx)) {
return ctx.u32_zero_value;
}
return LoadMask(ctx, ctx.subgroup_mask_lt);
}
Id EmitSubgroupLeMask(EmitContext& ctx) {
if (!StageSupportsSubgroups(ctx)) {
return ctx.Const(1u);
}
return LoadMask(ctx, ctx.subgroup_mask_le);
}
Id EmitSubgroupGtMask(EmitContext& ctx) {
if (!StageSupportsSubgroups(ctx)) {
return ctx.u32_zero_value;
}
return LoadMask(ctx, ctx.subgroup_mask_gt);
}
Id EmitSubgroupGeMask(EmitContext& ctx) {
if (!StageSupportsSubgroups(ctx)) {
return ctx.Const(1u);
}
return LoadMask(ctx, ctx.subgroup_mask_ge);
}
@ -262,7 +222,7 @@ Id EmitShuffleButterfly(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id
Id EmitFSwizzleAdd(EmitContext& ctx, Id op_a, Id op_b, Id swizzle) {
const Id three{ctx.Const(3U)};
Id mask{GetThreadId(ctx)};
Id mask{ctx.OpLoad(ctx.U32[1], ctx.subgroup_local_invocation_id)};
mask = ctx.OpBitwiseAnd(ctx.U32[1], mask, three);
mask = ctx.OpShiftLeftLogical(ctx.U32[1], mask, ctx.Const(1U));
mask = ctx.OpShiftRightLogical(ctx.U32[1], swizzle, mask);

9
src/shader_recompiler/backend/spirv/spirv_emit_context.cpp
View File

@ -1447,7 +1447,7 @@ void EmitContext::DefineInputs(const IR::Program& program) {
if (info.uses_is_helper_invocation) {
is_helper_invocation = DefineInput(*this, U1, false, spv::BuiltIn::HelperInvocation);
}
if (info.uses_subgroup_mask && profile.SupportsSubgroupStage(stage)) {
if (info.uses_subgroup_mask) {
subgroup_mask_eq = DefineInput(*this, U32[4], false, spv::BuiltIn::SubgroupEqMaskKHR);
subgroup_mask_lt = DefineInput(*this, U32[4], false, spv::BuiltIn::SubgroupLtMaskKHR);
subgroup_mask_le = DefineInput(*this, U32[4], false, spv::BuiltIn::SubgroupLeMaskKHR);
@ -1461,10 +1461,9 @@ void EmitContext::DefineInputs(const IR::Program& program) {
Decorate(subgroup_mask_ge, spv::Decoration::Flat);
}
}
if ((info.uses_fswzadd || info.uses_subgroup_invocation_id || info.uses_subgroup_shuffles ||
(profile.warp_size_potentially_larger_than_guest &&
(info.uses_subgroup_vote || info.uses_subgroup_mask))) &&
profile.SupportsSubgroupStage(stage)) {
if (info.uses_fswzadd || info.uses_subgroup_invocation_id || info.uses_subgroup_shuffles ||
(profile.warp_size_potentially_larger_than_guest &&
(info.uses_subgroup_vote || info.uses_subgroup_mask))) {
AddCapability(spv::Capability::GroupNonUniform);
subgroup_local_invocation_id =
DefineInput(*this, U32[1], false, spv::BuiltIn::SubgroupLocalInvocationId);

7
src/shader_recompiler/profile.h
View File

@ -10,8 +10,6 @@
namespace Shader {
enum class Stage : u32;
struct Profile {
u32 supported_spirv{0x00010000};
bool unified_descriptor_binding{};
@ -34,7 +32,6 @@ struct Profile {
bool support_fp64_signed_zero_nan_preserve{};
bool support_explicit_workgroup_layout{};
bool support_vote{};
u32 supported_subgroup_stages{0x7F};
bool support_viewport_index_layer_non_geometry{};
bool support_viewport_mask{};
bool support_typeless_image_loads{};
@ -98,10 +95,6 @@ struct Profile {
u64 min_ssbo_alignment{};
u32 max_user_clip_distances{};
bool SupportsSubgroupStage(Stage stage) const {
return (supported_subgroup_stages & (1u << static_cast<u32>(stage))) != 0;
}
};
} // namespace Shader

15
src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
View File

@ -353,20 +353,6 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
serialization_thread(1, "VkPipelineSerialization") {
const auto& float_control{device.FloatControlProperties()};
const VkDriverId driver_id{device.GetDriverID()};
const VkShaderStageFlags subgroup_stages{device.GetSubgroupSupportedStages()};
const auto subgroup_stage_bit{[subgroup_stages](VkShaderStageFlags flag, Shader::Stage stage) {
return (subgroup_stages & flag) != 0 ? (1u << static_cast<u32>(stage)) : 0u;
}};
const u32 supported_subgroup_stages{
subgroup_stage_bit(VK_SHADER_STAGE_VERTEX_BIT, Shader::Stage::VertexA) |
subgroup_stage_bit(VK_SHADER_STAGE_VERTEX_BIT, Shader::Stage::VertexB) |
subgroup_stage_bit(VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT,
Shader::Stage::TessellationControl) |
subgroup_stage_bit(VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT,
Shader::Stage::TessellationEval) |
subgroup_stage_bit(VK_SHADER_STAGE_GEOMETRY_BIT, Shader::Stage::Geometry) |
subgroup_stage_bit(VK_SHADER_STAGE_FRAGMENT_BIT, Shader::Stage::Fragment) |
subgroup_stage_bit(VK_SHADER_STAGE_COMPUTE_BIT, Shader::Stage::Compute)};
profile = Shader::Profile{
.supported_spirv = device.SupportedSpirvVersion(),
.unified_descriptor_binding = true,
@ -396,7 +382,6 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
float_control.shaderSignedZeroInfNanPreserveFloat64 != VK_FALSE,
.support_explicit_workgroup_layout = device.IsKhrWorkgroupMemoryExplicitLayoutSupported(),
.support_vote = device.IsSubgroupFeatureSupported(VK_SUBGROUP_FEATURE_VOTE_BIT),
.supported_subgroup_stages = supported_subgroup_stages,
.support_viewport_index_layer_non_geometry =
device.IsExtShaderViewportIndexLayerSupported(),
.support_viewport_mask = device.IsNvViewportArray2Supported(),

9
src/video_core/renderer_vulkan/vk_texture_cache.cpp
View File

@ -998,18 +998,11 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched
compute_pass_descriptor_queue);
}
}
// Log which unswizzle passes are active, for debugging purposes.
LOG_INFO(Render_Vulkan, "GPU compute unswizzle passes active: 2d={}, 3d={}, pitch={}",
bl_unswizzle_2d_pass.has_value(), bl_unswizzle_3d_pass.has_value(),
pitch_unswizzle_pass.has_value());
for (size_t index_a = 0; index_a < VideoCore::Surface::MaxPixelFormat; index_a++) {
const auto image_format = static_cast<PixelFormat>(index_a);
if (IsPixelFormatASTC(image_format) && !device.IsOptimalAstcSupported()) {
view_formats[index_a].push_back(VK_FORMAT_A8B8G8R8_UNORM_PACK32);
} else if (HasImageUnswizzlePasses() && !IsPixelFormatASTC(image_format) &&
!IsPixelFormatBCn(image_format) &&
VideoCore::Surface::DefaultBlockWidth(image_format) == 1 &&
VideoCore::Surface::DefaultBlockHeight(image_format) == 1 &&
} else if (HasImageUnswizzlePasses() && !IsPixelFormatBCn(image_format) &&
VideoCore::Surface::GetFormatType(image_format) ==
VideoCore::Surface::SurfaceType::ColorTexture) {
const u32 bytes_per_block = VideoCore::Surface::BytesPerBlock(image_format);

2
src/video_core/vulkan_common/vulkan_device.cpp
View File

@ -505,8 +505,6 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken CustomBorderColor.");
RemoveExtensionFeature(extensions.custom_border_color, features.custom_border_color,
VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME);
RemoveExtensionFeature(extensions.border_color_swizzle, features.border_color_swizzle,
VK_EXT_BORDER_COLOR_SWIZZLE_EXTENSION_NAME);
LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken shader float controls.");
RemoveExtension(extensions.shader_float_controls, VK_KHR_SHADER_FLOAT_CONTROLS_EXTENSION_NAME);
LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken shader atomic int64.");

5
src/video_core/vulkan_common/vulkan_device.h
View File

@ -430,11 +430,6 @@ FN_MAX_LIMIT_LIST
return properties.subgroup_properties.supportedOperations & feature;
}
/// Returns the shader stages that support subgroup operations.
VkShaderStageFlags GetSubgroupSupportedStages() const {
return properties.subgroup_properties.supportedStages;
}
/// Returns the maximum number of push descriptors.
u32 MaxPushDescriptors() const {
return properties.push_descriptor.maxPushDescriptors;

Write Preview
Loading…
Cancel
Save