shader_recompiler: Fix cube sampling coordinates. (#2266)

shadps4-emu · Jan 30, 2025 · 929e152 · 929e152
1 parent 4bb578f
commit 929e152
Show file tree

Hide file tree

Showing 2 changed files with 21 additions and 6 deletions.
diff --git a/src/shader_recompiler/ir/passes/resource_tracking_pass.cpp b/src/shader_recompiler/ir/passes/resource_tracking_pass.cpp
@@ -583,6 +583,18 @@ void PatchTextureBufferArgs(IR::Block& block, IR::Inst& inst, Info& info) {
     }
 }
 
+IR::Value FixCubeCoords(IR::IREmitter& ir, const AmdGpu::Image& image, const IR::Value& x,
+                        const IR::Value& y, const IR::Value& face) {
+    if (!image.IsCube()) {
+        return ir.CompositeConstruct(x, y, face);
+    }
+    // AMD cube math results in coordinates in the range [1.0, 2.0]. We need
+    // to convert this to the range [0.0, 1.0] to get correct results.
+    const auto fixed_x = ir.FPSub(IR::F32{x}, ir.Imm32(1.f));
+    const auto fixed_y = ir.FPSub(IR::F32{y}, ir.Imm32(1.f));
+    return ir.CompositeConstruct(fixed_x, fixed_y, face);
+}
+
 void PatchImageSampleArgs(IR::Block& block, IR::Inst& inst, Info& info,
                           const ImageResource& image_res, const AmdGpu::Image& image) {
     const auto handle = inst.Arg(0);
@@ -643,8 +655,8 @@ void PatchImageSampleArgs(IR::Block& block, IR::Inst& inst, Info& info,
         case AmdGpu::ImageType::Color1DArray:
             return read(0);
         case AmdGpu::ImageType::Color2D:
-        case AmdGpu::ImageType::Color2DArray:
         case AmdGpu::ImageType::Color2DMsaa:
+        case AmdGpu::ImageType::Color2DArray:
             return ir.CompositeConstruct(read(0), read(8));
         case AmdGpu::ImageType::Color3D:
             return ir.CompositeConstruct(read(0), read(8), read(16));
@@ -665,8 +677,8 @@ void PatchImageSampleArgs(IR::Block& block, IR::Inst& inst, Info& info,
             addr_reg = addr_reg + 2;
             return {get_addr_reg(addr_reg - 2), get_addr_reg(addr_reg - 1)};
         case AmdGpu::ImageType::Color2D:
-        case AmdGpu::ImageType::Color2DArray:
         case AmdGpu::ImageType::Color2DMsaa:
+        case AmdGpu::ImageType::Color2DArray:
             // (du/dx, dv/dx), (du/dy, dv/dy)
             addr_reg = addr_reg + 4;
             return {ir.CompositeConstruct(get_addr_reg(addr_reg - 4), get_addr_reg(addr_reg - 3)),
@@ -711,12 +723,13 @@ void PatchImageSampleArgs(IR::Block& block, IR::Inst& inst, Info& info,
         case AmdGpu::ImageType::Color2D: // x, y
             addr_reg = addr_reg + 2;
             return ir.CompositeConstruct(get_coord(addr_reg - 2, 0), get_coord(addr_reg - 1, 1));
-        case AmdGpu::ImageType::Color2DArray: // x, y, slice
-            [[fallthrough]];
         case AmdGpu::ImageType::Color2DMsaa: // x, y, frag
+            [[fallthrough]];
+        case AmdGpu::ImageType::Color2DArray: // x, y, slice
             addr_reg = addr_reg + 3;
-            return ir.CompositeConstruct(get_coord(addr_reg - 3, 0), get_coord(addr_reg - 2, 1),
-                                         get_addr_reg(addr_reg - 1));
+            // Note we can use FixCubeCoords with fallthrough cases since it checks for image type.
+            return FixCubeCoords(ir, image, get_coord(addr_reg - 3, 0), get_coord(addr_reg - 2, 1),
+                                 get_addr_reg(addr_reg - 1));
         case AmdGpu::ImageType::Color3D: // x, y, z
             addr_reg = addr_reg + 3;
             return ir.CompositeConstruct(get_coord(addr_reg - 3, 0), get_coord(addr_reg - 2, 1),

diff --git a/src/shader_recompiler/specialization.h b/src/shader_recompiler/specialization.h
@@ -47,6 +47,7 @@ struct ImageSpecialization {
     AmdGpu::ImageType type = AmdGpu::ImageType::Color2D;
     bool is_integer = false;
     bool is_storage = false;
+    bool is_cube = false;
     AmdGpu::CompMapping dst_select{};
     AmdGpu::NumberConversion num_conversion{};
 
@@ -127,6 +128,7 @@ struct StageSpecialization {
                          spec.type = sharp.GetViewType(desc.is_array);
                          spec.is_integer = AmdGpu::IsInteger(sharp.GetNumberFmt());
                          spec.is_storage = desc.is_written;
+                         spec.is_cube = sharp.IsCube();
                          if (spec.is_storage) {
                              spec.dst_select = sharp.DstSelect();
                          }