From fe91066f4673f7a3ee87235f08b72db4910eb01c Mon Sep 17 00:00:00 2001
From: Wollnashorn <Wollnashorn@users.noreply.github.com>
Date: Wed, 5 Apr 2023 03:02:24 +0200
Subject: [PATCH] video_core: Enable ImageGather with subpixel offset on Intel

---
 src/shader_recompiler/profile.h                      | 6 +++---
 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 -----
 7 files changed, 11 insertions(+), 17 deletions(-)

diff --git a/src/shader_recompiler/profile.h b/src/shader_recompiler/profile.h
index 31390e869b..9f88fb4407 100644
--- a/src/shader_recompiler/profile.h
+++ b/src/shader_recompiler/profile.h
@@ -52,9 +52,9 @@ struct Profile {
     bool need_declared_frag_colors{};
     /// Prevents fast math optimizations that may cause inaccuracies
     bool need_fastmath_off{};
-    /// Some GPU vendors use a lower fixed point format of 16.8 when calculating pixel coordinates
-    /// in the ImageGather instruction than the Maxwell architecture does. Applying an offset does
-    /// fix this mismatching rounding behaviour.
+    /// Some GPU vendors use a different rounding precision when calculating texture pixel
+    /// coordinates with the 16.8 format in the ImageGather instruction than the Maxwell
+    /// architecture. Applying an offset does fix this mismatching rounding behaviour.
     bool need_gather_subpixel_offset{};
 
     /// OpFClamp is broken and OpFMax + OpFMin should be used instead
diff --git a/src/video_core/renderer_opengl/gl_device.cpp b/src/video_core/renderer_opengl/gl_device.cpp
index d36a0a7a1d..22ed16ebfb 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 e8104c4def..cc0b95f1a5 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 b40aa6f5e6..6ecda29842 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 f512572673..8963b6a666 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 0939b62c92..6f288b3f8a 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 50e95bccab..41b5da18ac 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