diff --git a/src/video_core/host_shaders/sgsr1_shader.vert b/src/video_core/host_shaders/sgsr1_shader.vert index 0d2a935a9d..820beb26de 100644 --- a/src/video_core/host_shaders/sgsr1_shader.vert +++ b/src/video_core/host_shaders/sgsr1_shader.vert @@ -3,17 +3,16 @@ #version 450 -layout(push_constant) uniform constants { - vec2 scale; - vec2 size; - vec2 resize_factor; - float edge_sharpness; +layout( push_constant ) uniform constants { + highp vec4 ViewportInfo[1]; + highp vec2 ResizeFactor; + highp float EdgeSharpness; }; layout(location = 0) out highp vec2 texcoord; void main() { float x = float((gl_VertexIndex & 1) << 2); float y = float((gl_VertexIndex & 2) << 1); - gl_Position = vec4(x - 1.0f, y - 1.0f, 0.0, 1.0f) * vec4(sign(resize_factor), 1.f, 1.f); - texcoord = vec2(x, y) * abs(resize_factor) * 0.5; + gl_Position = vec4(x - 1.0f, y - 1.0f, 0.0, 1.0f); + texcoord = vec2(x, y) * ResizeFactor * 0.5; } diff --git a/src/video_core/host_shaders/sgsr1_shader_mobile.frag b/src/video_core/host_shaders/sgsr1_shader_mobile.frag index 2e62d60af3..31c94851da 100644 --- a/src/video_core/host_shaders/sgsr1_shader_mobile.frag +++ b/src/video_core/host_shaders/sgsr1_shader_mobile.frag @@ -3,80 +3,107 @@ #version 460 core -precision highp float; -precision highp int; +//precision mediump float; +//precision highp int; +#define highp +#define mediump // Operation modes: RGBA -> 1, RGBY -> 3, LERP -> 4 -#define OPERATION_MODE 1 -#define EDGE_THRESHOLD (8.0 / 255.0) +#define OperationMode 1 +#define EdgeThreshold 8.0/255.0 +// #define EdgeSharpness 2.0 -layout(push_constant) uniform constants { - vec2 scale; - vec2 size; - vec2 resize_factor; - float edge_sharpness; +layout( push_constant ) uniform constants { + highp vec4 ViewportInfo[1]; + highp vec2 ResizeFactor; + highp float EdgeSharpness; }; -layout(set = 0, binding = 0) uniform sampler2D sampler0; -layout(location=0) in vec2 texcoord; -layout(location=0) out vec4 frag_color; +layout(set = 0, binding = 0) uniform mediump sampler2D ps0; +layout(location=0) in highp vec2 in_TEXCOORD0; +layout(location=0) out vec4 out_Target0; -vec4 weightY(vec4 dx, vec4 dy, vec4 std) { - vec4 x = ((dx * dx) + (dy * dy)) * 0.55f + std; - return (x - 1.f) * (x - 4.f) * 3.8125f; // approx. of (x - 1) * (x - 4)^3 +float fastLanczos2(float x) { + float wA = x-4.0; + float wB = x*wA-wA; + wA *= wA; + return wB*wA; +} + +vec2 weightY(float dx, float dy,float c, float std) { + float x = ((dx*dx)+(dy* dy))* 0.55 + clamp(abs(c)*std, 0.0, 1.0); + float w = fastLanczos2(x); + return vec2(w, w * c); } void main() { - vec4 color = textureLod(sampler0, texcoord.xy, 0.0f); - // image coord - vec2 icoord = (texcoord * size + vec2(-0.5f, 0.5f)); - vec2 icoord_pixel = floor(icoord); - vec2 coord = icoord_pixel * scale; - vec2 pl = icoord - icoord_pixel; - // left: 0, right: 1, upDown: 2 - mat3x4 dg = mat3x4( - textureGather(sampler0, coord, 1), - textureGather(sampler0, coord + vec2(2.f * scale.x, 0.0f), 1), - vec4( - textureGather(sampler0, coord + vec2(scale.x, -scale.y), 1).wz, - textureGather(sampler0, coord + vec2(scale.x, +scale.y), 1).yx - ) - ); - float edgeVote = abs(dg[0].z - dg[0].y) + abs(color.y - dg[0].y) + abs(color.y - dg[0].z); - if (edgeVote > EDGE_THRESHOLD) { - float mean = (dg[0].y + dg[0].z + dg[1].x + dg[1].w) * 0.25f; - dg = dg - mean; - vec4 sum = abs(dg[0]) + abs(dg[1]) + abs(dg[2]); - float std = 2.181818f / (sum.x + sum.y + sum.z + sum.w); - mat2x4 w = mat2x4( - weightY( - pl.xxxx + vec4(+1.0f, +0.0f, +0.0f, +1.0f), - pl.yyyy + vec4(-1.0f, -1.0f, +0.0f, +0.0f), - clamp(abs(dg[0]) * std, 0.0f, 1.0f) - ) + weightY( - pl.xxxx + vec4(-1.0f, -2.0f, -2.0f, -1.0f), - pl.yyyy + vec4(-1.0f, -1.0f, +0.0f, +0.0f), - clamp(abs(dg[1]) * std, 0.0f, 1.0f) - ) + weightY( - pl.xxxx + vec4(+0.0f, -1.0f, -1.0f, +0.0f), - pl.yyyy + vec4(+1.0f, +1.0f, -2.0f, -2.0f), - clamp(abs(dg[2]) * std, 0.0f, 1.0f) - ), - dg[0] + dg[1] + dg[2] - ); - // compute final y with bounds - vec2 yb = vec2( - min(min(dg[0].y, dg[0].z), min(dg[1].x, dg[1].w)), // min - max(max(dg[0].y, dg[0].z), max(dg[1].x, dg[1].w)) // max - ); - vec2 fvy = vec2( - w[0].x + w[0].y + w[0].z + w[0].w, - w[1].x + w[1].y + w[1].z + w[1].w - ); - float fy = clamp((fvy.y / fvy.x) * edge_sharpness, yb[0], yb[1]); - // Smooth high contrast input - float dy = clamp(fy - color.y + mean, -23.0f / 255.0f, 23.0f / 255.0f); - color = clamp(color + dy, 0.0f, 1.0f); + vec4 color; + if(OperationMode == 1) + color.xyz = textureLod(ps0, in_TEXCOORD0.xy, 0.0).xyz; + else + color.xyzw = textureLod(ps0, in_TEXCOORD0.xy, 0.0).xyzw; + + highp float xCenter = abs(in_TEXCOORD0.x + -0.5); + highp float yCenter = abs(in_TEXCOORD0.y + -0.5); + + //todo: config the SR region based on needs + //if ( OperationMode!=4 && xCenter*xCenter+yCenter*yCenter<=0.4 * 0.4) + if ( OperationMode!=4) { + highp vec2 imgCoord = ((in_TEXCOORD0.xy * ViewportInfo[0].zw) + vec2(-0.5, 0.5)); + highp vec2 imgCoordPixel = floor(imgCoord); + highp vec2 coord = imgCoordPixel * ViewportInfo[0].xy; + vec2 pl = imgCoord - imgCoordPixel; + vec4 left = textureGather(ps0, coord, OperationMode); + float edgeVote = abs(left.z - left.y) + abs(color[OperationMode] - left.y) + abs(color[OperationMode] - left.z) ; + if(edgeVote > EdgeThreshold) { + coord.x += ViewportInfo[0].x; + + highp vec2 IR_highp_vec2_0 = coord + vec2(ViewportInfo[0].x, 0.0); + vec4 right = textureGather(ps0, IR_highp_vec2_0, OperationMode); + vec4 upDown; + highp vec2 IR_highp_vec2_1 = coord + vec2(0.0, -ViewportInfo[0].y); + upDown.xy = textureGather(ps0, IR_highp_vec2_1, OperationMode).wz; + highp vec2 IR_highp_vec2_2 = coord + vec2(0.0, ViewportInfo[0].y); + upDown.zw = textureGather(ps0, IR_highp_vec2_2, OperationMode).yx; + + float mean = (left.y+left.z+right.x+right.w)*0.25; + left = left - vec4(mean); + right = right - vec4(mean); + upDown = upDown - vec4(mean); + color.w =color[OperationMode] - mean; + + float sum = (((((abs(left.x)+abs(left.y))+abs(left.z))+abs(left.w))+(((abs(right.x)+abs(right.y))+abs(right.z))+abs(right.w)))+(((abs(upDown.x)+abs(upDown.y))+abs(upDown.z))+abs(upDown.w))); + float std = 2.181818/sum; + + vec2 aWY = weightY(pl.x, pl.y+1.0, upDown.x,std); + aWY += weightY(pl.x-1.0, pl.y+1.0, upDown.y,std); + aWY += weightY(pl.x-1.0, pl.y-2.0, upDown.z,std); + aWY += weightY(pl.x, pl.y-2.0, upDown.w,std); + aWY += weightY(pl.x+1.0, pl.y-1.0, left.x,std); + aWY += weightY(pl.x, pl.y-1.0, left.y,std); + aWY += weightY(pl.x, pl.y, left.z,std); + aWY += weightY(pl.x+1.0, pl.y, left.w,std); + aWY += weightY(pl.x-1.0, pl.y-1.0, right.x,std); + aWY += weightY(pl.x-2.0, pl.y-1.0, right.y,std); + aWY += weightY(pl.x-2.0, pl.y, right.z,std); + aWY += weightY(pl.x-1.0, pl.y, right.w,std); + + float finalY = aWY.y/aWY.x; + + float maxY = max(max(left.y,left.z),max(right.x,right.w)); + float minY = min(min(left.y,left.z),min(right.x,right.w)); + finalY = clamp(EdgeSharpness*finalY, minY, maxY); + + float deltaY = finalY -color.w; + + //smooth high contrast input + deltaY = clamp(deltaY, -23.0 / 255.0, 23.0 / 255.0); + + color.x = clamp((color.x+deltaY),0.0,1.0); + color.y = clamp((color.y+deltaY),0.0,1.0); + color.z = clamp((color.z+deltaY),0.0,1.0); + } } - color.w = 1.0f; //assume alpha channel is not used - frag_color.xyzw = color; + + color.w = 1.0; //assume alpha channel is not used + out_Target0.xyzw = color; } \ No newline at end of file diff --git a/src/video_core/host_shaders/sgsr1_shader_mobile_edge_direction.frag b/src/video_core/host_shaders/sgsr1_shader_mobile_edge_direction.frag index 10a8ff8e00..477b73fe63 100644 --- a/src/video_core/host_shaders/sgsr1_shader_mobile_edge_direction.frag +++ b/src/video_core/host_shaders/sgsr1_shader_mobile_edge_direction.frag @@ -3,20 +3,26 @@ #version 460 core -//precision float; -//precision int; +//precision mediump float; +//precision highp int; +#define highp +#define mediump // Operation modes: RGBA -> 1, RGBY -> 3, LERP -> 4 #define OperationMode 1 +// If set, will use edge direction to improve visual quality +// Expect a minimal cost increase +#define UseEdgeDirection 1 #define EdgeThreshold 8.0/255.0 +// #define EdgeSharpness 2.0 layout( push_constant ) uniform constants { - vec4 ViewportInfo[1]; - vec2 ResizeFactor; - float EdgeSharpness; + highp vec4 ViewportInfo[1]; + highp vec2 ResizeFactor; + highp float EdgeSharpness; }; -layout(set = 0, binding = 0) uniform sampler2D ps0; -layout(location=0) in vec2 in_TEXCOORD0; +layout(set = 0, binding = 0) uniform mediump sampler2D ps0; +layout(location=0) in highp vec2 in_TEXCOORD0; layout(location=0) out vec4 out_Target0; float fastLanczos2(float x) { @@ -26,11 +32,23 @@ float fastLanczos2(float x) { return wB*wA; } -vec2 weightY(float dx, float dy, float c, vec3 data) { +#if defined(UseEdgeDirection) +vec2 weightY(float dx, float dy, float c, vec3 data) +#else +vec2 weightY(float dx, float dy, float c, float data) +#endif +{ +#if defined(UseEdgeDirection) float std = data.x; vec2 dir = data.yz; + float edgeDis = ((dx*dir.y)+(dy*dir.x)); float x = (((dx*dx)+(dy*dy))+((edgeDis*edgeDis)*((clamp(((c*c)*std),0.0,1.0)*0.7)+-1.0))); +#else + float std = data; + float x = ((dx*dx)+(dy* dy))* 0.55 + clamp(abs(c)*std, 0.0, 1.0); +#endif + float w = fastLanczos2(x); return vec2(w, w * c); } @@ -55,22 +73,27 @@ void main() { else color.xyzw = textureLod(ps0, in_TEXCOORD0.xy, 0.0).xyzw; + highp float xCenter = abs(in_TEXCOORD0.x + -0.5); + highp float yCenter = abs(in_TEXCOORD0.y + -0.5); + + //todo: config the SR region based on needs + //if ( OperationMode!=4 && xCenter*xCenter+yCenter*yCenter<=0.4 * 0.4) if ( OperationMode!=4) { - vec2 imgCoord = ((in_TEXCOORD0.xy*ViewportInfo[0].zw)+vec2(-0.5,0.5)); - vec2 imgCoordPixel = floor(imgCoord); - vec2 coord = (imgCoordPixel*ViewportInfo[0].xy); + highp vec2 imgCoord = ((in_TEXCOORD0.xy*ViewportInfo[0].zw)+vec2(-0.5,0.5)); + highp vec2 imgCoordPixel = floor(imgCoord); + highp vec2 coord = (imgCoordPixel*ViewportInfo[0].xy); vec2 pl = imgCoord - imgCoordPixel; vec4 left = textureGather(ps0, coord, OperationMode); float edgeVote = abs(left.z - left.y) + abs(color[OperationMode] - left.y) + abs(color[OperationMode] - left.z) ; if(edgeVote > EdgeThreshold) { coord.x += ViewportInfo[0].x; - vec2 IR_highp_vec2_0 = coord + vec2(ViewportInfo[0].x, 0.0); + highp vec2 IR_highp_vec2_0 = coord + vec2(ViewportInfo[0].x, 0.0); vec4 right = textureGather(ps0, IR_highp_vec2_0, OperationMode); vec4 upDown; - vec2 IR_highp_vec2_1 = coord + vec2(0.0, -ViewportInfo[0].y); + highp vec2 IR_highp_vec2_1 = coord + vec2(0.0, -ViewportInfo[0].y); upDown.xy = textureGather(ps0, IR_highp_vec2_1, OperationMode).wz; - vec2 IR_highp_vec2_2 = coord + vec2(0.0, ViewportInfo[0].y); + highp vec2 IR_highp_vec2_2 = coord + vec2(0.0, ViewportInfo[0].y); upDown.zw = textureGather(ps0, IR_highp_vec2_2, OperationMode).yx; float mean = (left.y+left.z+right.x+right.w)*0.25; @@ -83,7 +106,11 @@ void main() { float sumMean = 1.014185e+01/sum; float std = (sumMean*sumMean); +#if defined(UseEdgeDirection) vec3 data = vec3(std, edgeDirection(left, right)); +#else + float data = std; +#endif vec2 aWY = weightY(pl.x, pl.y+1.0, upDown.x,data); aWY += weightY(pl.x-1.0, pl.y+1.0, upDown.y,data); aWY += weightY(pl.x-1.0, pl.y-2.0, upDown.z,data); diff --git a/src/video_core/renderer_vulkan/present/sgsr.cpp b/src/video_core/renderer_vulkan/present/sgsr.cpp index 4175190690..301a23c45f 100644 --- a/src/video_core/renderer_vulkan/present/sgsr.cpp +++ b/src/video_core/renderer_vulkan/present/sgsr.cpp @@ -20,35 +20,33 @@ namespace Vulkan { using PushConstants = std::array; SGSR::SGSR(const Device& device, MemoryAllocator& memory_allocator, size_t image_count, VkExtent2D extent, bool edge_dir) - : m_device{device} - , m_memory_allocator{memory_allocator} - , m_image_count{image_count} - , m_extent{extent} + : m_device{device}, m_memory_allocator{memory_allocator} + , m_image_count{image_count}, m_extent{extent} , m_edge_dir{edge_dir} { // Not finished yet initializing at ctor time? m_dynamic_images.resize(m_image_count); for (auto& images : m_dynamic_images) { - images.image = CreateWrappedImage(m_memory_allocator, m_extent, VK_FORMAT_R16G16B16A16_SFLOAT); - images.image_view = CreateWrappedImageView(m_device, images.image, VK_FORMAT_R16G16B16A16_SFLOAT); + images.images[0] = CreateWrappedImage(m_memory_allocator, m_extent, VK_FORMAT_B8G8R8A8_UNORM); + images.image_views[0] = CreateWrappedImageView(m_device, images.images[0], VK_FORMAT_B8G8R8A8_UNORM); } - m_renderpass = CreateWrappedRenderPass(m_device, VK_FORMAT_R16G16B16A16_SFLOAT); + m_renderpass = CreateWrappedRenderPass(m_device, VK_FORMAT_B8G8R8A8_UNORM); for (auto& images : m_dynamic_images) - images.framebuffer = CreateWrappedFramebuffer(m_device, m_renderpass, images.image_view, m_extent); + images.framebuffers[0] = CreateWrappedFramebuffer(m_device, m_renderpass, images.image_views[0], m_extent); m_sampler = CreateBilinearSampler(m_device); m_vert_shader = BuildShader(m_device, SGSR1_SHADER_VERT_SPV); - m_stage_shader = m_edge_dir + m_stage_shader[0] = m_edge_dir ? BuildShader(m_device, SGSR1_SHADER_MOBILE_EDGE_DIRECTION_FRAG_SPV) : BuildShader(m_device, SGSR1_SHADER_MOBILE_FRAG_SPV); // 2 descriptors, 2 descriptor sets per invocation - m_descriptor_pool = CreateWrappedDescriptorPool(m_device, m_image_count, m_image_count); + m_descriptor_pool = CreateWrappedDescriptorPool(m_device, SGSR_STAGE_COUNT * m_image_count, SGSR_STAGE_COUNT * m_image_count); m_descriptor_set_layout = CreateWrappedDescriptorSetLayout(m_device, {VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER}); - VkDescriptorSetLayout layout = *m_descriptor_set_layout; + std::vector layouts(SGSR_STAGE_COUNT, *m_descriptor_set_layout); for (auto& images : m_dynamic_images) - images.descriptor_sets = CreateWrappedDescriptorSets(m_descriptor_pool, layout); + images.descriptor_sets = CreateWrappedDescriptorSets(m_descriptor_pool, layouts); const VkPushConstantRange range{ .stageFlags = VK_SHADER_STAGE_VERTEX_BIT | VK_SHADER_STAGE_FRAGMENT_BIT, @@ -65,7 +63,7 @@ SGSR::SGSR(const Device& device, MemoryAllocator& memory_allocator, size_t image .pPushConstantRanges = &range, }; m_pipeline_layout = m_device.GetLogical().CreatePipelineLayout(ci); - m_stage_pipeline = CreateWrappedPipeline(m_device, m_renderpass, m_pipeline_layout, std::tie(m_vert_shader, m_stage_shader)); + m_stage_pipeline[0] = CreateWrappedPipeline(m_device, m_renderpass, m_pipeline_layout, std::tie(m_vert_shader, m_stage_shader[0])); } void SGSR::UpdateDescriptorSets(VkImageView image_view, size_t image_index) { @@ -81,7 +79,7 @@ void SGSR::UploadImages(Scheduler& scheduler) { if (!m_images_ready) { scheduler.Record([&](vk::CommandBuffer cmdbuf) { for (auto& image : m_dynamic_images) - ClearColorImage(cmdbuf, *image.image); + ClearColorImage(cmdbuf, *image.images[0]); }); scheduler.Finish(); m_images_ready = true; @@ -90,12 +88,12 @@ void SGSR::UploadImages(Scheduler& scheduler) { VkImageView SGSR::Draw(Scheduler& scheduler, size_t image_index, VkImage source_image, VkImageView source_image_view, VkExtent2D input_image_extent, const Common::Rectangle& crop_rect) { Images& images = m_dynamic_images[image_index]; - auto const output_image = *images.image; - auto const descriptor_set = images.descriptor_sets[0]; - auto const framebuffer = *images.framebuffer; - auto const pipeline = *m_stage_pipeline; + auto const stage0_image = *images.images[0]; + auto const stage0_descriptor_set = images.descriptor_sets[0]; + auto const stage0_framebuffer = *images.framebuffers[0]; + auto const stage0_pipeline = *m_stage_pipeline[0]; - VkPipelineLayout layout = *m_pipeline_layout; + VkPipelineLayout pipeline_layout = *m_pipeline_layout; VkRenderPass renderpass = *m_renderpass; VkExtent2D extent = m_extent; @@ -114,10 +112,10 @@ VkImageView SGSR::Draw(Scheduler& scheduler, size_t image_index, VkImage source_ // highp float EdgeSharpness; // }; PushConstants viewport_con{}; - viewport_con[0] = std::bit_cast(std::abs(1.f / viewport_width)); - viewport_con[1] = std::bit_cast(std::abs(1.f / viewport_height)); - viewport_con[2] = std::bit_cast(std::abs(viewport_width)); - viewport_con[3] = std::bit_cast(std::abs(viewport_height)); + viewport_con[0] = std::bit_cast(1.f / viewport_width); + viewport_con[1] = std::bit_cast(1.f / viewport_height); + viewport_con[2] = std::bit_cast(viewport_width); + viewport_con[3] = std::bit_cast(viewport_height); viewport_con[4] = std::bit_cast(viewport_width / input_image_width); viewport_con[5] = std::bit_cast(viewport_height / input_image_height); viewport_con[6] = std::bit_cast(sharpening); @@ -128,16 +126,16 @@ VkImageView SGSR::Draw(Scheduler& scheduler, size_t image_index, VkImage source_ scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([=](vk::CommandBuffer cmdbuf) { TransitionImageLayout(cmdbuf, source_image, VK_IMAGE_LAYOUT_GENERAL); - TransitionImageLayout(cmdbuf, output_image, VK_IMAGE_LAYOUT_GENERAL); - BeginRenderPass(cmdbuf, renderpass, framebuffer, extent); - cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline); - cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, layout, 0, descriptor_set, {}); - cmdbuf.PushConstants(layout, VK_SHADER_STAGE_VERTEX_BIT | VK_SHADER_STAGE_FRAGMENT_BIT, viewport_con); + TransitionImageLayout(cmdbuf, stage0_image, VK_IMAGE_LAYOUT_GENERAL); + BeginRenderPass(cmdbuf, renderpass, stage0_framebuffer, extent); + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, stage0_pipeline); + cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline_layout, 0, stage0_descriptor_set, {}); + cmdbuf.PushConstants(pipeline_layout, VK_SHADER_STAGE_VERTEX_BIT | VK_SHADER_STAGE_FRAGMENT_BIT, viewport_con); cmdbuf.Draw(3, 1, 0, 0); cmdbuf.EndRenderPass(); - TransitionImageLayout(cmdbuf, output_image, VK_IMAGE_LAYOUT_GENERAL); + TransitionImageLayout(cmdbuf, stage0_image, VK_IMAGE_LAYOUT_GENERAL); }); - return *images.image_view; + return *images.image_views[0]; } } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/present/sgsr.h b/src/video_core/renderer_vulkan/present/sgsr.h index 67b25be75d..d6362c361d 100644 --- a/src/video_core/renderer_vulkan/present/sgsr.h +++ b/src/video_core/renderer_vulkan/present/sgsr.h @@ -31,16 +31,16 @@ private: vk::DescriptorSetLayout m_descriptor_set_layout; vk::PipelineLayout m_pipeline_layout; vk::ShaderModule m_vert_shader; - vk::ShaderModule m_stage_shader; - vk::Pipeline m_stage_pipeline; + std::array m_stage_shader; + std::array m_stage_pipeline; vk::RenderPass m_renderpass; vk::Sampler m_sampler; struct Images { vk::DescriptorSets descriptor_sets; - vk::Image image; - vk::ImageView image_view; - vk::Framebuffer framebuffer; + std::array images; + std::array image_views; + std::array framebuffers; }; std::vector m_dynamic_images; bool m_images_ready{};