Merge pull request #10699 from liamwhite/conditional-barrier

shader_recompiler: remove barriers in conditional control flow when device lacks support
This commit is contained in:
Matías Locatti 2023-06-12 16:50:59 -03:00 committed by GitHub
commit 42b2bc204f
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
10 changed files with 65 additions and 0 deletions

View File

@ -216,6 +216,7 @@ add_library(shader_recompiler STATIC
frontend/maxwell/translate_program.h
host_translate_info.h
ir_opt/collect_shader_info_pass.cpp
ir_opt/conditional_barrier_pass.cpp
ir_opt/constant_propagation_pass.cpp
ir_opt/dead_code_elimination_pass.cpp
ir_opt/dual_vertex_pass.cpp

View File

@ -289,6 +289,9 @@ IR::Program TranslateProgram(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Blo
if (!host_info.support_int64) {
Optimization::LowerInt64ToInt32(program);
}
if (!host_info.support_conditional_barrier) {
Optimization::ConditionalBarrierPass(program);
}
Optimization::SsaRewritePass(program);
Optimization::ConstantPropagationPass(env, program);

View File

@ -18,6 +18,8 @@ struct HostTranslateInfo {
bool support_viewport_index_layer{}; ///< True when the device supports gl_Layer in VS
bool support_geometry_shader_passthrough{}; ///< True when the device supports geometry
///< passthrough shaders
bool support_conditional_barrier{}; ///< True when the device supports barriers in conditional
///< control flow
};
} // namespace Shader

View File

@ -0,0 +1,44 @@
// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "shader_recompiler/frontend/ir/program.h"
#include "shader_recompiler/ir_opt/passes.h"
namespace Shader::Optimization {
void ConditionalBarrierPass(IR::Program& program) {
s32 conditional_control_flow_count{0};
s32 conditional_return_count{0};
for (IR::AbstractSyntaxNode& node : program.syntax_list) {
switch (node.type) {
case IR::AbstractSyntaxNode::Type::If:
case IR::AbstractSyntaxNode::Type::Loop:
conditional_control_flow_count++;
break;
case IR::AbstractSyntaxNode::Type::EndIf:
case IR::AbstractSyntaxNode::Type::Repeat:
conditional_control_flow_count--;
break;
case IR::AbstractSyntaxNode::Type::Unreachable:
case IR::AbstractSyntaxNode::Type::Return:
if (conditional_control_flow_count > 0) {
conditional_return_count++;
}
break;
case IR::AbstractSyntaxNode::Type::Block:
for (IR::Inst& inst : node.data.block->Instructions()) {
if ((conditional_control_flow_count > 0 || conditional_return_count > 0) &&
inst.GetOpcode() == IR::Opcode::Barrier) {
LOG_WARNING(Shader, "Barrier within conditional control flow");
inst.ReplaceOpcode(IR::Opcode::Identity);
}
}
break;
default:
break;
}
}
ASSERT(conditional_control_flow_count == 0);
}
} // namespace Shader::Optimization

View File

@ -13,6 +13,7 @@ struct HostTranslateInfo;
namespace Shader::Optimization {
void CollectShaderInfoPass(Environment& env, IR::Program& program);
void ConditionalBarrierPass(IR::Program& program);
void ConstantPropagationPass(Environment& env, IR::Program& program);
void DeadCodeEliminationPass(IR::Program& program);
void GlobalMemoryToStorageBufferPass(IR::Program& program);

View File

@ -201,6 +201,7 @@ Device::Device(Core::Frontend::EmuWindow& emu_window) {
use_asynchronous_shaders = Settings::values.use_asynchronous_shaders.GetValue() &&
!(is_amd || (is_intel && !is_linux)) && !strict_context_required;
use_driver_cache = is_nvidia;
supports_conditional_barriers = !is_intel;
LOG_INFO(Render_OpenGL, "Renderer_VariableAOFFI: {}", has_variable_aoffi);
LOG_INFO(Render_OpenGL, "Renderer_ComponentIndexingBug: {}", has_component_indexing_bug);

View File

@ -188,6 +188,10 @@ public:
return strict_context_required;
}
bool SupportsConditionalBarriers() const {
return supports_conditional_barriers;
}
private:
static bool TestVariableAoffi();
static bool TestPreciseBug();
@ -233,6 +237,7 @@ private:
bool has_bool_ref_bug{};
bool can_report_memory{};
bool strict_context_required{};
bool supports_conditional_barriers{};
std::string vendor_name;
};

View File

@ -239,6 +239,7 @@ ShaderCache::ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindo
.support_snorm_render_buffer = false,
.support_viewport_index_layer = device.HasVertexViewportLayer(),
.support_geometry_shader_passthrough = device.HasGeometryShaderPassthrough(),
.support_conditional_barrier = device.SupportsConditionalBarriers(),
} {
if (use_asynchronous_shaders) {
workers = CreateWorkers();

View File

@ -386,6 +386,8 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
IsFormatSupported(VK_FORMAT_D24_UNORM_S8_UINT,
VK_FORMAT_FEATURE_DEPTH_STENCIL_ATTACHMENT_BIT, FormatType::Optimal);
supports_conditional_barriers = !(is_intel_anv || is_intel_windows);
CollectPhysicalMemoryInfo();
CollectToolingInfo();

View File

@ -585,6 +585,10 @@ public:
return properties.properties.limits.maxVertexInputBindings;
}
bool SupportsConditionalBarriers() const {
return supports_conditional_barriers;
}
private:
/// Checks if the physical device is suitable and configures the object state
/// with all necessary info about its properties.
@ -688,6 +692,7 @@ private:
bool must_emulate_bgr565{}; ///< Emulates BGR565 by swizzling RGB565 format.
bool dynamic_state3_blending{}; ///< Has all blending features of dynamic_state3.
bool dynamic_state3_enables{}; ///< Has all enables features of dynamic_state3.
bool supports_conditional_barriers{}; ///< Allows barriers in conditional control flow.
u64 device_access_memory{}; ///< Total size of device local memory in bytes.
u32 sets_per_pool{}; ///< Sets per Description Pool