summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
-rw-r--r--src/video_core/buffer_cache/buffer_cache.h19
-rw-r--r--src/video_core/renderer_opengl/gl_buffer_cache.h12
2 files changed, 22 insertions, 9 deletions
diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h
index ec64f2293..47cb0a47d 100644
--- a/src/video_core/buffer_cache/buffer_cache.h
+++ b/src/video_core/buffer_cache/buffer_cache.h
@@ -680,6 +680,9 @@ void BufferCache<P>::SetUniformBuffersState(const std::array<u32, NUM_STAGES>& m
680 const UniformBufferSizes* sizes) { 680 const UniformBufferSizes* sizes) {
681 if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) { 681 if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) {
682 if (enabled_uniform_buffer_masks != mask) { 682 if (enabled_uniform_buffer_masks != mask) {
683 if constexpr (IS_OPENGL) {
684 fast_bound_uniform_buffers.fill(0);
685 }
683 dirty_uniform_buffers.fill(~u32{0}); 686 dirty_uniform_buffers.fill(~u32{0});
684 } 687 }
685 } 688 }
@@ -1020,6 +1023,7 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
1020 // Fast path for Nvidia 1023 // Fast path for Nvidia
1021 if (!HasFastUniformBufferBound(stage, binding_index)) { 1024 if (!HasFastUniformBufferBound(stage, binding_index)) {
1022 // We only have to bind when the currently bound buffer is not the fast version 1025 // We only have to bind when the currently bound buffer is not the fast version
1026 fast_bound_uniform_buffers[stage] |= 1U << binding_index;
1023 runtime.BindFastUniformBuffer(stage, binding_index, size); 1027 runtime.BindFastUniformBuffer(stage, binding_index, size);
1024 } 1028 }
1025 const auto span = ImmediateBufferWithData(cpu_addr, size); 1029 const auto span = ImmediateBufferWithData(cpu_addr, size);
@@ -1027,8 +1031,9 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
1027 return; 1031 return;
1028 } 1032 }
1029 } 1033 }
1030 fast_bound_uniform_buffers[stage] |= 1U << binding_index; 1034 if constexpr (IS_OPENGL) {
1031 1035 fast_bound_uniform_buffers[stage] |= 1U << binding_index;
1036 }
1032 // Stream buffer path to avoid stalling on non-Nvidia drivers or Vulkan 1037 // Stream buffer path to avoid stalling on non-Nvidia drivers or Vulkan
1033 const std::span<u8> span = runtime.BindMappedUniformBuffer(stage, binding_index, size); 1038 const std::span<u8> span = runtime.BindMappedUniformBuffer(stage, binding_index, size);
1034 cpu_memory.ReadBlockUnsafe(cpu_addr, span.data(), size); 1039 cpu_memory.ReadBlockUnsafe(cpu_addr, span.data(), size);
@@ -1046,9 +1051,15 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
1046 // This exists to avoid instances where the fast buffer is bound and a GPU write happens 1051 // This exists to avoid instances where the fast buffer is bound and a GPU write happens
1047 return; 1052 return;
1048 } 1053 }
1049 fast_bound_uniform_buffers[stage] &= ~(1U << binding_index);
1050
1051 const u32 offset = buffer.Offset(cpu_addr); 1054 const u32 offset = buffer.Offset(cpu_addr);
1055 if constexpr (IS_OPENGL) {
1056 // Fast buffer will be unbound
1057 fast_bound_uniform_buffers[stage] &= ~(1U << binding_index);
1058
1059 // Mark the index as dirty if offset doesn't match
1060 const bool is_copy_bind = offset != 0 && !runtime.SupportsNonZeroUniformOffset();
1061 dirty_uniform_buffers[stage] |= (is_copy_bind ? 1U : 0U) << index;
1062 }
1052 if constexpr (NEEDS_BIND_UNIFORM_INDEX) { 1063 if constexpr (NEEDS_BIND_UNIFORM_INDEX) {
1053 runtime.BindUniformBuffer(stage, binding_index, buffer, offset, size); 1064 runtime.BindUniformBuffer(stage, binding_index, buffer, offset, size);
1054 } else { 1065 } else {
diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.h b/src/video_core/renderer_opengl/gl_buffer_cache.h
index bc16abafb..060d36427 100644
--- a/src/video_core/renderer_opengl/gl_buffer_cache.h
+++ b/src/video_core/renderer_opengl/gl_buffer_cache.h
@@ -92,16 +92,14 @@ public:
92 VideoCore::Surface::PixelFormat format); 92 VideoCore::Surface::PixelFormat format);
93 93
94 void BindFastUniformBuffer(size_t stage, u32 binding_index, u32 size) { 94 void BindFastUniformBuffer(size_t stage, u32 binding_index, u32 size) {
95 const GLuint handle = fast_uniforms[stage][binding_index].handle;
96 const GLsizeiptr gl_size = static_cast<GLsizeiptr>(size);
95 if (use_assembly_shaders) { 97 if (use_assembly_shaders) {
96 const GLuint handle = fast_uniforms[stage][binding_index].handle;
97 const GLsizeiptr gl_size = static_cast<GLsizeiptr>(size);
98 glBindBufferRangeNV(PABO_LUT[stage], binding_index, handle, 0, gl_size); 98 glBindBufferRangeNV(PABO_LUT[stage], binding_index, handle, 0, gl_size);
99 } else { 99 } else {
100 const GLuint base_binding = graphics_base_uniform_bindings[stage]; 100 const GLuint base_binding = graphics_base_uniform_bindings[stage];
101 const GLuint binding = base_binding + binding_index; 101 const GLuint binding = base_binding + binding_index;
102 glBindBufferRange(GL_UNIFORM_BUFFER, binding, 102 glBindBufferRange(GL_UNIFORM_BUFFER, binding, handle, 0, gl_size);
103 fast_uniforms[stage][binding_index].handle, 0,
104 static_cast<GLsizeiptr>(size));
105 } 103 }
106 } 104 }
107 105
@@ -134,6 +132,10 @@ public:
134 return has_fast_buffer_sub_data; 132 return has_fast_buffer_sub_data;
135 } 133 }
136 134
135 [[nodiscard]] bool SupportsNonZeroUniformOffset() const noexcept {
136 return !use_assembly_shaders;
137 }
138
137 void SetBaseUniformBindings(const std::array<GLuint, 5>& bindings) { 139 void SetBaseUniformBindings(const std::array<GLuint, 5>& bindings) {
138 graphics_base_uniform_bindings = bindings; 140 graphics_base_uniform_bindings = bindings;
139 } 141 }