From dcc5b4f6b005a2c89bb4e77bca4cfe8705734021 Mon Sep 17 00:00:00 2001
From: Marshall Mohror <mohror64@gmail.com>
Date: Fri, 22 Oct 2021 23:09:29 -0500
Subject: Presentation: Only use FP16 in scaling shaders on supported devices
 in Vulkan

---
 src/video_core/host_shaders/fidelityfx_fsr.comp | 106 ++++++++++++------------
 1 file changed, 54 insertions(+), 52 deletions(-)

(limited to 'src/video_core/host_shaders/fidelityfx_fsr.comp')

diff --git a/src/video_core/host_shaders/fidelityfx_fsr.comp b/src/video_core/host_shaders/fidelityfx_fsr.comp
index cbb601580..6b97f789d 100644
--- a/src/video_core/host_shaders/fidelityfx_fsr.comp
+++ b/src/video_core/host_shaders/fidelityfx_fsr.comp
@@ -28,80 +28,82 @@
 // THE SOFTWARE.
 
 layout( push_constant ) uniform constants {
-    u32vec2 input_size;
+    uvec4 Const0;
+    uvec4 Const1;
+    uvec4 Const2;
+    uvec4 Const3;
 };
 
-uvec4 Const0;
-uvec4 Const1;
-uvec4 Const2;
-uvec4 Const3;
+layout(set=0,binding=0) uniform sampler2D InputTexture;
+layout(set=0,binding=1,rgba16f) uniform image2D OutputTexture;
 
 #define A_GPU 1
 #define A_GLSL 1
-#define A_HALF
 
-#include "ffx_a.h"
+#ifndef YUZU_USE_FP16
+    #include "ffx_a.h"
 
-f16vec4 LinearToSRGB(f16vec4 linear) {
-    bvec4 selector = greaterThan(linear, f16vec4(0.00313066844250063));
-    f16vec4 low = linear * float16_t(12.92);
-    f16vec4 high = float16_t(1.055) * pow(linear, f16vec4(1 / 2.4)) - float16_t(0.055);
-    return mix(low, high, selector);
-}
-
-f16vec4 SRGBToLinear(f16vec4 srgb) {
-    bvec4 selector = greaterThan(srgb, f16vec4(0.0404482362771082));
-    f16vec4 low = srgb * float16_t(1.0 / 12.92);
-    f16vec4 high = pow((srgb + float16_t(0.055)) * float16_t(1.0 / 1.055), f16vec4(2.4));
-    return mix(low, high, selector);
-}
+    #if USE_EASU
+        #define FSR_EASU_F 1
+        AF4 FsrEasuRF(AF2 p) { AF4 res = textureGather(InputTexture, p, 0); return res; }
+        AF4 FsrEasuGF(AF2 p) { AF4 res = textureGather(InputTexture, p, 1); return res; }
+        AF4 FsrEasuBF(AF2 p) { AF4 res = textureGather(InputTexture, p, 2); return res; }
+    #endif
+    #if USE_RCAS
+        #define FSR_RCAS_F 1
+        AF4 FsrRcasLoadF(ASU2 p) { return texelFetch(InputTexture, ASU2(p), 0); }
+        void FsrRcasInputF(inout AF1 r, inout AF1 g, inout AF1 b) {}
+    #endif
+#else
+    #define A_HALF
+    #include "ffx_a.h"
 
-#if USE_EASU
-    #define FSR_EASU_H 1
-    f16vec4 FsrEasuRH(vec2 p) { f16vec4 res = f16vec4(textureGather(InputTexture, p, 0)); return res; }
-    f16vec4 FsrEasuGH(vec2 p) { f16vec4 res = f16vec4(textureGather(InputTexture, p, 1)); return res; }
-    f16vec4 FsrEasuBH(vec2 p) { f16vec4 res = f16vec4(textureGather(InputTexture, p, 2)); return res; }
-#endif
-#if USE_RCAS
-    #define FSR_RCAS_H 1
-    f16vec4 FsrRcasLoadH(ASW2 p) { return f16vec4(texelFetch(InputTexture, ASU2(p), 0)); }
-    void FsrRcasInputH(inout float16_t r, inout float16_t g, inout float16_t b) {}
+    #if USE_EASU
+        #define FSR_EASU_H 1
+        AH4 FsrEasuRH(AF2 p) { AH4 res = AH4(textureGather(InputTexture, p, 0)); return res; }
+        AH4 FsrEasuGH(AF2 p) { AH4 res = AH4(textureGather(InputTexture, p, 1)); return res; }
+        AH4 FsrEasuBH(AF2 p) { AH4 res = AH4(textureGather(InputTexture, p, 2)); return res; }
+    #endif
+    #if USE_RCAS
+        #define FSR_RCAS_H 1
+        AH4 FsrRcasLoadH(ASW2 p) { return AH4(texelFetch(InputTexture, ASU2(p), 0)); }
+        void FsrRcasInputH(inout AH1 r,inout AH1 g,inout AH1 b){}
+    #endif
 #endif
 
 #include "ffx_fsr1.h"
 
-void CurrFilter(u32vec2 pos) {
-    // For debugging
+void CurrFilter(AU2 pos) {
 #if USE_BILINEAR
-    vec2 pp = (vec2(pos) * vec2_AU2(Const0.xy) + vec2_AU2(Const0.zw)) * vec2_AU2(Const1.xy) + vec2(0.5, -0.5) * vec2_AU2(Const1.zw);
-    imageStore(OutputTexture, ivec2(pos), textureLod(InputTexture, pp, 0.0));
+    AF2 pp = (AF2(pos) * AF2_AU2(Const0.xy) + AF2_AU2(Const0.zw)) * AF2_AU2(Const1.xy) + AF2(0.5, -0.5) * AF2_AU2(Const1.zw);
+    imageStore(OutputTexture, ASU2(pos), textureLod(InputTexture, pp, 0.0));
 #endif
 #if USE_EASU
-    f16vec3 c;
-    FsrEasuH(c, pos, Const0, Const1, Const2, Const3);
-    imageStore(OutputTexture, ivec2(pos), f16vec4(c, 1));
+    #ifndef YUZU_USE_FP16
+        AF3 c;
+        FsrEasuF(c, pos, Const0, Const1, Const2, Const3);
+        imageStore(OutputTexture, ASU2(pos), AF4(c, 1));
+    #else
+        AH3 c;
+        FsrEasuH(c, pos, Const0, Const1, Const2, Const3);
+        imageStore(OutputTexture, ASU2(pos), AH4(c, 1));
+    #endif
 #endif
 #if USE_RCAS
-    f16vec3 c;
-    FsrRcasH(c.r, c.g, c.b, pos, Const0);
-    imageStore(OutputTexture, ivec2(pos), f16vec4(c, 1));
+    #ifndef YUZU_USE_FP16
+        AF3 c;
+        FsrRcasF(c.r, c.g, c.b, pos, Const0);
+        imageStore(OutputTexture, ASU2(pos), AF4(c, 1));
+    #else
+        AH3 c;
+        FsrRcasH(c.r, c.g, c.b, pos, Const0);
+        imageStore(OutputTexture, ASU2(pos), AH4(c, 1));
+    #endif
 #endif
-
 }
 
 layout(local_size_x=64) in;
 void main() {
-
-#if USE_EASU || USE_BILINEAR
-    vec2 ires = vec2(input_size);
-    vec2 tres = textureSize(InputTexture, 0);
-    vec2 ores = imageSize(OutputTexture);
-    FsrEasuCon(Const0, Const1, Const2, Const3, ires.x, ires.y, tres.x, tres.y, ores.x, ores.y);
-#endif
-#if USE_RCAS
-    FsrRcasCon(Const0, 0.25f);
-#endif
-
     // Do remapping of local xy in workgroup for a more PS-like swizzle pattern.
     AU2 gxy = ARmp8x8(gl_LocalInvocationID.x) + AU2(gl_WorkGroupID.x << 4u, gl_WorkGroupID.y << 4u);
     CurrFilter(gxy);
-- 
cgit v1.2.3