[vulkan, qcom] Fix shader float controls on QCOM driver (#4297)

Finishes my torture to find the culprit behind the broken behavior with float controls, even if device does support flush denorm on fp32 it doesn't really support it and provokes bad rounding modes by not flushing correctly denorms; this also returns the other working rounding modes on QCOM drivers.

_Special Thanks_

1.- Smoly The Big (@Gidoly)

Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4297
Reviewed-by: Samuel <lizzie@eden-emu.dev>
Reviewed-by: MaranBr <maranbr@eden-emu.dev>

src/common/settings.h

@@ -352,7 +352,7 @@ struct Values {
                                                   true};
 
     SwitchableSetting<ScalingFilter> scaling_filter{linkage,
-                                                    ScalingFilter::Bilinear,
+                                                    ScalingFilter::NearestNeighbor,
                                                     "scaling_filter",
                                                     Category::Renderer,
                                                     Specialization::Default,

src/shader_recompiler/backend/spirv/emit_spirv.cpp

@@ -335,7 +335,7 @@ void SetupDenormControl(const Profile& profile, const IR::Program& program, Emit
     if (info.uses_fp32_denorms_flush && info.uses_fp32_denorms_preserve) {
         LOG_DEBUG(Shader_SPIRV, "Fp32 denorm flush and preserve on the same shader");
     } else if (info.uses_fp32_denorms_flush) {
-        if (profile.support_fp32_denorm_flush) {
+        if (profile.support_fp32_denorm_flush && !profile.has_broken_fp32_denorm_flush) {
             ctx.AddCapability(spv::Capability::DenormFlushToZero);
             ctx.AddExecutionMode(main_func, spv::ExecutionMode::DenormFlushToZero, 32U);
         } else {

src/shader_recompiler/profile.h

@@ -86,6 +86,8 @@ struct Profile {
     bool has_broken_signed_operations{};
     /// Float controls break when fp16 is enabled
     bool has_broken_fp16_float_controls{};
+    /// Declaring fp32 denorm flush to zero miscompiles on some drivers
+    bool has_broken_fp32_denorm_flush{};
     /// Dynamic vec4 indexing is broken on some OpenGL drivers
     bool has_gl_component_indexing_bug{};
     /// The precise type qualifier is broken in the fragment stage of some drivers

src/video_core/renderer_opengl/gl_shader_cache.cpp

@@ -231,6 +231,7 @@ ShaderCache::ShaderCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
           .has_broken_unsigned_image_offsets = true,
           .has_broken_signed_operations = true,
           .has_broken_fp16_float_controls = false,
+          .has_broken_fp32_denorm_flush = false,
           .has_gl_component_indexing_bug = device.HasComponentIndexingBug(),
           .has_gl_precise_bug = device.HasPreciseBug(),
           .has_gl_cbuf_ftou_bug = device.HasCbufFtouBug(),

src/video_core/renderer_vulkan/renderer_vulkan.cpp

@@ -160,7 +160,12 @@ try
                    present_manager,
                    scheduler,
                    PresentFiltersForDisplay)
-    // Captures will use the same filtering as applets as well
+    , blit_capture(device_memory,
+                   device,
+                   memory_allocator,
+                   present_manager,
+                   scheduler,
+                   PresentFiltersForDisplay)
     , blit_applet(device_memory,
                   device,
                   memory_allocator,
@@ -261,12 +266,12 @@ vk::Buffer RendererVulkan::RenderToBuffer(std::span<const Tegra::FramebufferConf
         f.image =
             CreateWrappedImage(memory_allocator, VkExtent2D{layout.width, layout.height}, format);
         f.image_view = CreateWrappedImageView(device, f.image, format);
-        f.framebuffer = blit_applet.CreateFramebuffer(device, layout, *f.image_view, format);
+        f.framebuffer = blit_capture.CreateFramebuffer(device, layout, *f.image_view, format);
         return f;
     }();
 
     auto dst_buffer = CreateWrappedBuffer(memory_allocator, buffer_size, MemoryUsage::Download);
-    blit_applet.DrawToFrame(device, rasterizer, &frame, framebuffers, layout, 1, format);
+    blit_capture.DrawToFrame(device, rasterizer, &frame, framebuffers, layout, 1, format);
 
     scheduler.RequestOutsideRenderPassOperationContext();
     scheduler.Record([&](vk::CommandBuffer cmdbuf) {

src/video_core/renderer_vulkan/renderer_vulkan.h

@@ -95,6 +95,7 @@ private:
     Swapchain swapchain;
     PresentManager present_manager;
     BlitScreen blit_swapchain;
+    BlitScreen blit_capture;
     BlitScreen blit_applet;
     RasterizerVulkan rasterizer;
 #ifdef HAS_LSFG

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -444,6 +444,7 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
         .has_broken_unsigned_image_offsets = false,
         .has_broken_signed_operations = false,
         .has_broken_fp16_float_controls = driver_id == VK_DRIVER_ID_NVIDIA_PROPRIETARY,
+        .has_broken_fp32_denorm_flush = driver_id == VK_DRIVER_ID_QUALCOMM_PROPRIETARY,
         .ignore_nan_fp_comparisons = false,
         .has_broken_spirv_subgroup_mask_vector_extract_dynamic = false,
         .has_broken_robust =

src/video_core/vulkan_common/vulkan_device.cpp

@@ -517,8 +517,6 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
         LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken color write enable.");
         RemoveExtensionFeature(extensions.color_write_enable, features.color_write_enable,
                                VK_EXT_COLOR_WRITE_ENABLE_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.");
         RemoveExtensionFeature(extensions.shader_atomic_int64, features.shader_atomic_int64,
                                VK_KHR_SHADER_ATOMIC_INT64_EXTENSION_NAME);