From 780240e6979b198e7bd10feaad5399b8b4b63762 Mon Sep 17 00:00:00 2001 From: Wollnashorn Date: Wed, 5 Apr 2023 01:29:46 +0200 Subject: shader_recompiler: Add subpixel offset for correct rounding at `ImageGather` On AMD a subpixel offset of 1/512 of the texel size is applied to the texture coordinates at a ImageGather call to ensure the rounding at the texel centers is done the same way as in Maxwell or other Nvidia architectures. See https://www.reedbeta.com/blog/texture-gathers-and-coordinate-precision/ for more details why this might be necessary. This should fix shadow artifacts at object edges in Zelda: Breath of the Wild (#9957, #6956). --- src/video_core/renderer_opengl/gl_device.cpp | 1 + src/video_core/renderer_opengl/gl_device.h | 5 +++++ src/video_core/renderer_opengl/gl_shader_cache.cpp | 1 + src/video_core/renderer_vulkan/vk_pipeline_cache.cpp | 1 + src/video_core/vulkan_common/vulkan_device.cpp | 1 + src/video_core/vulkan_common/vulkan_device.h | 5 +++++ 6 files changed, 14 insertions(+) (limited to 'src/video_core') diff --git a/src/video_core/renderer_opengl/gl_device.cpp b/src/video_core/renderer_opengl/gl_device.cpp index 22ed16ebf..d36a0a7a1 100644 --- a/src/video_core/renderer_opengl/gl_device.cpp +++ b/src/video_core/renderer_opengl/gl_device.cpp @@ -169,6 +169,7 @@ Device::Device(Core::Frontend::EmuWindow& emu_window) { has_draw_texture = GLAD_GL_NV_draw_texture; warp_size_potentially_larger_than_guest = !is_nvidia && !is_intel; need_fastmath_off = is_nvidia; + need_gather_subpixel_offset = is_amd; can_report_memory = GLAD_GL_NVX_gpu_memory_info; // At the moment of writing this, only Nvidia's driver optimizes BufferSubData on exclusive diff --git a/src/video_core/renderer_opengl/gl_device.h b/src/video_core/renderer_opengl/gl_device.h index 3ff8cad83..e8104c4de 100644 --- a/src/video_core/renderer_opengl/gl_device.h +++ b/src/video_core/renderer_opengl/gl_device.h @@ -160,6 +160,10 @@ public: return need_fastmath_off; } + bool NeedsGatherSubpixelOffset() const { + return need_gather_subpixel_offset; + } + bool HasCbufFtouBug() const { return has_cbuf_ftou_bug; } @@ -225,6 +229,7 @@ private: bool has_draw_texture{}; bool warp_size_potentially_larger_than_guest{}; bool need_fastmath_off{}; + bool need_gather_subpixel_offset{}; bool has_cbuf_ftou_bug{}; bool has_bool_ref_bug{}; bool can_report_memory{}; diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp index 479bb8ba3..b40aa6f5e 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp @@ -218,6 +218,7 @@ ShaderCache::ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindo .lower_left_origin_mode = true, .need_declared_frag_colors = true, .need_fastmath_off = device.NeedsFastmathOff(), + .need_gather_subpixel_offset = device.NeedsGatherSubpixelOffset(), .has_broken_spirv_clamp = true, .has_broken_unsigned_image_offsets = true, diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp index 0684cceed..f51257267 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp @@ -329,6 +329,7 @@ PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, const Device& device .lower_left_origin_mode = false, .need_declared_frag_colors = false, + .need_gather_subpixel_offset = device.NeedsGatherSubpixelOffset(), .has_broken_spirv_clamp = driver_id == VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS, .has_broken_spirv_position_input = driver_id == VK_DRIVER_ID_QUALCOMM_PROPRIETARY, diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index 6f288b3f8..0939b62c9 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -431,6 +431,7 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR "AMD GCN4 and earlier have broken VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT"); has_broken_cube_compatibility = true; } + need_gather_subpixel_offset = true; } if (extensions.sampler_filter_minmax && is_amd) { // Disable ext_sampler_filter_minmax on AMD GCN4 and lower as it is broken. diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h index 41b5da18a..50e95bcca 100644 --- a/src/video_core/vulkan_common/vulkan_device.h +++ b/src/video_core/vulkan_common/vulkan_device.h @@ -554,6 +554,10 @@ public: return features.robustness2.nullDescriptor; } + bool NeedsGatherSubpixelOffset() const { + return need_gather_subpixel_offset; + } + u32 GetMaxVertexInputAttributes() const { return properties.properties.limits.maxVertexInputAttributes; } @@ -664,6 +668,7 @@ 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 need_gather_subpixel_offset{}; ///< Needs offset at ImageGather for correct rounding. u64 device_access_memory{}; ///< Total size of device local memory in bytes. u32 sets_per_pool{}; ///< Sets per Description Pool -- cgit v1.2.3 From fe91066f4673f7a3ee87235f08b72db4910eb01c Mon Sep 17 00:00:00 2001 From: Wollnashorn Date: Wed, 5 Apr 2023 03:02:24 +0200 Subject: video_core: Enable ImageGather with subpixel offset on Intel --- src/video_core/renderer_opengl/gl_device.cpp | 1 - src/video_core/renderer_opengl/gl_device.h | 9 ++++----- src/video_core/renderer_opengl/gl_shader_cache.cpp | 2 +- src/video_core/renderer_vulkan/vk_pipeline_cache.cpp | 4 +++- src/video_core/vulkan_common/vulkan_device.cpp | 1 - src/video_core/vulkan_common/vulkan_device.h | 5 ----- 6 files changed, 8 insertions(+), 14 deletions(-) (limited to 'src/video_core') diff --git a/src/video_core/renderer_opengl/gl_device.cpp b/src/video_core/renderer_opengl/gl_device.cpp index d36a0a7a1..22ed16ebf 100644 --- a/src/video_core/renderer_opengl/gl_device.cpp +++ b/src/video_core/renderer_opengl/gl_device.cpp @@ -169,7 +169,6 @@ Device::Device(Core::Frontend::EmuWindow& emu_window) { has_draw_texture = GLAD_GL_NV_draw_texture; warp_size_potentially_larger_than_guest = !is_nvidia && !is_intel; need_fastmath_off = is_nvidia; - need_gather_subpixel_offset = is_amd; can_report_memory = GLAD_GL_NVX_gpu_memory_info; // At the moment of writing this, only Nvidia's driver optimizes BufferSubData on exclusive diff --git a/src/video_core/renderer_opengl/gl_device.h b/src/video_core/renderer_opengl/gl_device.h index e8104c4de..cc0b95f1a 100644 --- a/src/video_core/renderer_opengl/gl_device.h +++ b/src/video_core/renderer_opengl/gl_device.h @@ -160,10 +160,6 @@ public: return need_fastmath_off; } - bool NeedsGatherSubpixelOffset() const { - return need_gather_subpixel_offset; - } - bool HasCbufFtouBug() const { return has_cbuf_ftou_bug; } @@ -180,6 +176,10 @@ public: return vendor_name == "ATI Technologies Inc."; } + bool IsIntel() const { + return vendor_name == "Intel"; + } + bool CanReportMemoryUsage() const { return can_report_memory; } @@ -229,7 +229,6 @@ private: bool has_draw_texture{}; bool warp_size_potentially_larger_than_guest{}; bool need_fastmath_off{}; - bool need_gather_subpixel_offset{}; bool has_cbuf_ftou_bug{}; bool has_bool_ref_bug{}; bool can_report_memory{}; diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp index b40aa6f5e..6ecda2984 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp @@ -218,7 +218,7 @@ ShaderCache::ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindo .lower_left_origin_mode = true, .need_declared_frag_colors = true, .need_fastmath_off = device.NeedsFastmathOff(), - .need_gather_subpixel_offset = device.NeedsGatherSubpixelOffset(), + .need_gather_subpixel_offset = device.IsAmd() || device.IsIntel(), .has_broken_spirv_clamp = true, .has_broken_unsigned_image_offsets = true, diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp index f51257267..8963b6a66 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp @@ -329,7 +329,9 @@ PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, const Device& device .lower_left_origin_mode = false, .need_declared_frag_colors = false, - .need_gather_subpixel_offset = device.NeedsGatherSubpixelOffset(), + .need_gather_subpixel_offset = driver_id == VK_DRIVER_ID_AMD_PROPRIETARY || + driver_id == VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS || + driver_id == VK_DRIVER_ID_INTEL_OPEN_SOURCE_MESA, .has_broken_spirv_clamp = driver_id == VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS, .has_broken_spirv_position_input = driver_id == VK_DRIVER_ID_QUALCOMM_PROPRIETARY, diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index 0939b62c9..6f288b3f8 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -431,7 +431,6 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR "AMD GCN4 and earlier have broken VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT"); has_broken_cube_compatibility = true; } - need_gather_subpixel_offset = true; } if (extensions.sampler_filter_minmax && is_amd) { // Disable ext_sampler_filter_minmax on AMD GCN4 and lower as it is broken. diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h index 50e95bcca..41b5da18a 100644 --- a/src/video_core/vulkan_common/vulkan_device.h +++ b/src/video_core/vulkan_common/vulkan_device.h @@ -554,10 +554,6 @@ public: return features.robustness2.nullDescriptor; } - bool NeedsGatherSubpixelOffset() const { - return need_gather_subpixel_offset; - } - u32 GetMaxVertexInputAttributes() const { return properties.properties.limits.maxVertexInputAttributes; } @@ -668,7 +664,6 @@ 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 need_gather_subpixel_offset{}; ///< Needs offset at ImageGather for correct rounding. u64 device_access_memory{}; ///< Total size of device local memory in bytes. u32 sets_per_pool{}; ///< Sets per Description Pool -- cgit v1.2.3 From c0e5ecc399cc08e4cd54b04c8d63b99e1fcbddc7 Mon Sep 17 00:00:00 2001 From: Wollnashorn Date: Wed, 12 Apr 2023 17:11:02 +0200 Subject: video_core: Enable ImageGather rounding fix on AMD open source drivers --- src/video_core/renderer_vulkan/vk_pipeline_cache.cpp | 2 ++ 1 file changed, 2 insertions(+) (limited to 'src/video_core') diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp index 8963b6a66..985cc3203 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp @@ -330,6 +330,8 @@ PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, const Device& device .lower_left_origin_mode = false, .need_declared_frag_colors = false, .need_gather_subpixel_offset = driver_id == VK_DRIVER_ID_AMD_PROPRIETARY || + driver_id == VK_DRIVER_ID_AMD_OPEN_SOURCE || + driver_id == VK_DRIVER_ID_MESA_RADV || driver_id == VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS || driver_id == VK_DRIVER_ID_INTEL_OPEN_SOURCE_MESA, -- cgit v1.2.3