glsl: SHFL fix and prefer shift operations over divide in glsl shader

src/shader_recompiler/backend/glsl/emit_glsl_atomic.cpp

@@ -20,7 +20,7 @@ static constexpr std::string_view cas_loop{R"(for (;;){{
 void SharedCasFunction(EmitContext& ctx, IR::Inst& inst, std::string_view offset,
                        std::string_view value, std::string_view function) {
     const auto ret{ctx.var_alloc.Define(inst, GlslVarType::U32)};
-    const std::string smem{fmt::format("smem[{}/4]", offset)};
+    const std::string smem{fmt::format("smem[{}>>2]", offset)};
     ctx.Add(cas_loop.data(), smem, ret, smem, function, smem, value, ret);
 }
 
@@ -45,7 +45,7 @@ void SsboCasFunctionF32(EmitContext& ctx, IR::Inst& inst, const IR::Value& bindi
 
 void EmitSharedAtomicIAdd32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
                             std::string_view value) {
-    ctx.AddU32("{}=atomicAdd(smem[{}/4],{});", inst, pointer_offset, value);
+    ctx.AddU32("{}=atomicAdd(smem[{}>>2],{});", inst, pointer_offset, value);
 }
 
 void EmitSharedAtomicSMin32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
@@ -56,7 +56,7 @@ void EmitSharedAtomicSMin32(EmitContext& ctx, IR::Inst& inst, std::string_view p
 
 void EmitSharedAtomicUMin32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
                             std::string_view value) {
-    ctx.AddU32("{}=atomicMin(smem[{}/4],{});", inst, pointer_offset, value);
+    ctx.AddU32("{}=atomicMin(smem[{}>>2],{});", inst, pointer_offset, value);
 }
 
 void EmitSharedAtomicSMax32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
@@ -67,7 +67,7 @@ void EmitSharedAtomicSMax32(EmitContext& ctx, IR::Inst& inst, std::string_view p
 
 void EmitSharedAtomicUMax32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
                             std::string_view value) {
-    ctx.AddU32("{}=atomicMax(smem[{}/4],{});", inst, pointer_offset, value);
+    ctx.AddU32("{}=atomicMax(smem[{}>>2],{});", inst, pointer_offset, value);
 }
 
 void EmitSharedAtomicInc32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
@@ -82,31 +82,31 @@ void EmitSharedAtomicDec32(EmitContext& ctx, IR::Inst& inst, std::string_view po
 
 void EmitSharedAtomicAnd32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
                            std::string_view value) {
-    ctx.AddU32("{}=atomicAnd(smem[{}/4],{});", inst, pointer_offset, value);
+    ctx.AddU32("{}=atomicAnd(smem[{}>>2],{});", inst, pointer_offset, value);
 }
 
 void EmitSharedAtomicOr32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
                           std::string_view value) {
-    ctx.AddU32("{}=atomicOr(smem[{}/4],{});", inst, pointer_offset, value);
+    ctx.AddU32("{}=atomicOr(smem[{}>>2],{});", inst, pointer_offset, value);
 }
 
 void EmitSharedAtomicXor32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
                            std::string_view value) {
-    ctx.AddU32("{}=atomicXor(smem[{}/4],{});", inst, pointer_offset, value);
+    ctx.AddU32("{}=atomicXor(smem[{}>>2],{});", inst, pointer_offset, value);
 }
 
 void EmitSharedAtomicExchange32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
                                 std::string_view value) {
-    ctx.AddU32("{}=atomicExchange(smem[{}/4],{});", inst, pointer_offset, value);
+    ctx.AddU32("{}=atomicExchange(smem[{}>>2],{});", inst, pointer_offset, value);
 }
 
 void EmitSharedAtomicExchange64(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
                                 std::string_view value) {
     // LOG_WARNING("Int64 Atomics not supported, fallback to non-atomic");
-    ctx.AddU64("{}=packUint2x32(uvec2(smem[{}/4],smem[({}+4)/4]));", inst, pointer_offset,
+    ctx.AddU64("{}=packUint2x32(uvec2(smem[{}>>2],smem[({}+4)>>2]));", inst, pointer_offset,
                pointer_offset);
-    ctx.Add("smem[{}/4]=unpackUint2x32({}).x;smem[({}+4)/4]=unpackUint2x32({}).y;", pointer_offset,
+    ctx.Add("smem[{}>>2]=unpackUint2x32({}).x;smem[({}+4)>>2]=unpackUint2x32({}).y;",
-            value, pointer_offset, value);
+            pointer_offset, value, pointer_offset, value);
 }
 
 void EmitStorageAtomicIAdd32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,

src/shader_recompiler/backend/glsl/emit_glsl_context_get_set.cpp

@@ -31,7 +31,7 @@ void EmitGetCbufU8([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst&
     } else {
         const auto offset_var{ctx.var_alloc.Consume(offset)};
         ctx.AddU32(
-            "{}=bitfieldExtract(floatBitsToUint({}_cbuf{}[{}/16][({}/4)%4]),int(({}%4)*8),8);",
+            "{}=bitfieldExtract(floatBitsToUint({}_cbuf{}[{}/16][({}>>2)%4]),int(({}%4)*8),8);",
             inst, ctx.stage_name, binding.U32(), offset_var, offset_var, offset_var);
     }
 }
@@ -46,8 +46,8 @@ void EmitGetCbufS8([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst&
     } else {
         const auto offset_var{ctx.var_alloc.Consume(offset)};
         ctx.AddU32(
-            "{}=bitfieldExtract(floatBitsToInt({}_cbuf{}[{}/16][({}/4)%4]),int(({}%4)*8),8);", inst,
+            "{}=bitfieldExtract(floatBitsToInt({}_cbuf{}[{}/16][({}>>2)%4]),int(({}%4)*8),8);",
-            ctx.stage_name, binding.U32(), offset_var, offset_var, offset_var);
+            inst, ctx.stage_name, binding.U32(), offset_var, offset_var, offset_var);
     }
 }
 
@@ -60,7 +60,7 @@ void EmitGetCbufU16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst
                    ((offset.U32() / 2) % 2) * 16);
     } else {
         const auto offset_var{ctx.var_alloc.Consume(offset)};
-        ctx.AddU32("{}=bitfieldExtract(floatBitsToUint({}_cbuf{}[{}/16][({}/4)%4]),int((({}/"
+        ctx.AddU32("{}=bitfieldExtract(floatBitsToUint({}_cbuf{}[{}/16][({}>>2)%4]),int((({}/"
                    "2)%2)*16),16);",
                    inst, ctx.stage_name, binding.U32(), offset_var, offset_var, offset_var);
     }
@@ -75,9 +75,9 @@ void EmitGetCbufS16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst
                    ((offset.U32() / 2) % 2) * 16);
     } else {
         const auto offset_var{ctx.var_alloc.Consume(offset)};
-        ctx.AddU32(
+        ctx.AddU32("{}=bitfieldExtract(floatBitsToInt({}_cbuf{}[{}/16][({}>>2)%4]),int((({}/"
-            "{}=bitfieldExtract(floatBitsToInt({}_cbuf{}[{}/16][({}/4)%4]),int((({}/2)%2)*16),16);",
+                   "2)%2)*16),16);",
-            inst, ctx.stage_name, binding.U32(), offset_var, offset_var, offset_var);
+                   inst, ctx.stage_name, binding.U32(), offset_var, offset_var, offset_var);
     }
 }
 
@@ -88,7 +88,7 @@ void EmitGetCbufU32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
                    offset.U32() / 16, OffsetSwizzle(offset.U32()));
     } else {
         const auto offset_var{ctx.var_alloc.Consume(offset)};
-        ctx.AddU32("{}=floatBitsToUint({}_cbuf{}[{}/16][({}/4)%4]);", inst, ctx.stage_name,
+        ctx.AddU32("{}=floatBitsToUint({}_cbuf{}[{}/16][({}>>2)%4]);", inst, ctx.stage_name,
                    binding.U32(), offset_var, offset_var);
     }
 }
@@ -100,7 +100,7 @@ void EmitGetCbufF32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
                    OffsetSwizzle(offset.U32()));
     } else {
         const auto offset_var{ctx.var_alloc.Consume(offset)};
-        ctx.AddF32("{}={}_cbuf{}[{}/16][({}/4)%4];", inst, ctx.stage_name, binding.U32(),
+        ctx.AddF32("{}={}_cbuf{}[{}/16][({}>>2)%4];", inst, ctx.stage_name, binding.U32(),
                    offset_var, offset_var);
     }
 }
@@ -116,7 +116,7 @@ void EmitGetCbufU32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding
     } else {
         const auto offset_var{ctx.var_alloc.Consume(offset)};
         ctx.AddU32x2("{}=uvec2(floatBitsToUint({}_cbuf{}[{}/16][({}/"
-                     "4)%4]),floatBitsToUint({}_cbuf{}[({}+4)/16][(({}+4)/4)%4]));",
+                     "4)%4]),floatBitsToUint({}_cbuf{}[({}+4)/16][(({}+4)>>2)%4]));",
                      inst, ctx.stage_name, binding.U32(), offset_var, offset_var, ctx.stage_name,
                      binding.U32(), offset_var, offset_var);
     }

src/shader_recompiler/backend/glsl/emit_glsl_memory.cpp

@@ -13,7 +13,7 @@ void EmitLoadStorageU8([[maybe_unused]] EmitContext& ctx, IR::Inst& inst,
                        [[maybe_unused]] const IR::Value& binding,
                        [[maybe_unused]] const IR::Value& offset) {
     const auto offset_var{ctx.var_alloc.Consume(offset)};
-    ctx.AddU32("{}=bitfieldExtract({}_ssbo{}[{}/4],int({}%4)*8,8);", inst, ctx.stage_name,
+    ctx.AddU32("{}=bitfieldExtract({}_ssbo{}[{}>>2],int({}%4)*8,8);", inst, ctx.stage_name,
                binding.U32(), offset_var, offset_var);
 }
 
@@ -21,7 +21,7 @@ void EmitLoadStorageS8([[maybe_unused]] EmitContext& ctx, IR::Inst& inst,
                        [[maybe_unused]] const IR::Value& binding,
                        [[maybe_unused]] const IR::Value& offset) {
     const auto offset_var{ctx.var_alloc.Consume(offset)};
-    ctx.AddS32("{}=bitfieldExtract(int({}_ssbo{}[{}/4]),int({}%4)*8,8);", inst, ctx.stage_name,
+    ctx.AddS32("{}=bitfieldExtract(int({}_ssbo{}[{}>>2]),int({}%4)*8,8);", inst, ctx.stage_name,
                binding.U32(), offset_var, offset_var);
 }
 
@@ -29,7 +29,7 @@ void EmitLoadStorageU16([[maybe_unused]] EmitContext& ctx, IR::Inst& inst,
                         [[maybe_unused]] const IR::Value& binding,
                         [[maybe_unused]] const IR::Value& offset) {
     const auto offset_var{ctx.var_alloc.Consume(offset)};
-    ctx.AddU32("{}=bitfieldExtract({}_ssbo{}[{}/4],int(({}/2)%2)*16,16);", inst, ctx.stage_name,
+    ctx.AddU32("{}=bitfieldExtract({}_ssbo{}[{}>>2],int(({}>>1)%2)*16,16);", inst, ctx.stage_name,
                binding.U32(), offset_var, offset_var);
 }
 
@@ -37,30 +37,31 @@ void EmitLoadStorageS16([[maybe_unused]] EmitContext& ctx, IR::Inst& inst,
                         [[maybe_unused]] const IR::Value& binding,
                         [[maybe_unused]] const IR::Value& offset) {
     const auto offset_var{ctx.var_alloc.Consume(offset)};
-    ctx.AddS32("{}=bitfieldExtract(int({}_ssbo{}[{}/4]),int(({}/2)%2)*16,16);", inst,
+    ctx.AddS32("{}=bitfieldExtract(int({}_ssbo{}[{}>>2]),int(({}>>1)%2)*16,16);", inst,
                ctx.stage_name, binding.U32(), offset_var, offset_var);
 }
 
 void EmitLoadStorage32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
                        const IR::Value& offset) {
     const auto offset_var{ctx.var_alloc.Consume(offset)};
-    ctx.AddU32("{}={}_ssbo{}[{}/4];", inst, ctx.stage_name, binding.U32(), offset_var);
+    ctx.AddU32("{}={}_ssbo{}[{}>>2];", inst, ctx.stage_name, binding.U32(), offset_var);
 }
 
 void EmitLoadStorage64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
                        const IR::Value& offset) {
     const auto offset_var{ctx.var_alloc.Consume(offset)};
-    ctx.AddU32x2("{}=uvec2({}_ssbo{}[{}/4],{}_ssbo{}[({}+4)/4]);", inst, ctx.stage_name,
+    ctx.AddU32x2("{}=uvec2({}_ssbo{}[{}>>2],{}_ssbo{}[({}+4)>>2]);", inst, ctx.stage_name,
                  binding.U32(), offset_var, ctx.stage_name, binding.U32(), offset_var);
 }
 
 void EmitLoadStorage128(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
                         const IR::Value& offset) {
     const auto offset_var{ctx.var_alloc.Consume(offset)};
-    ctx.AddU32x4(
+    ctx.AddU32x4("{}=uvec4({}_ssbo{}[{}>>2],{}_ssbo{}[({}+4)>>2],{}_ssbo{}[({}+8)>>2],{}_ssbo{}[({}"
-        "{}=uvec4({}_ssbo{}[{}/4],{}_ssbo{}[({}+4)/4],{}_ssbo{}[({}+8)/4],{}_ssbo{}[({}+12)/4]);",
+                 "+12)>>2]);",
-        inst, ctx.stage_name, binding.U32(), offset_var, ctx.stage_name, binding.U32(), offset_var,
+                 inst, ctx.stage_name, binding.U32(), offset_var, ctx.stage_name, binding.U32(),
-        ctx.stage_name, binding.U32(), offset_var, ctx.stage_name, binding.U32(), offset_var);
+                 offset_var, ctx.stage_name, binding.U32(), offset_var, ctx.stage_name,
+                 binding.U32(), offset_var);
 }
 
 void EmitWriteStorageU8([[maybe_unused]] EmitContext& ctx,
@@ -68,7 +69,7 @@ void EmitWriteStorageU8([[maybe_unused]] EmitContext& ctx,
                         [[maybe_unused]] const IR::Value& offset,
                         [[maybe_unused]] std::string_view value) {
     const auto offset_var{ctx.var_alloc.Consume(offset)};
-    ctx.Add("{}_ssbo{}[{}/4]=bitfieldInsert({}_ssbo{}[{}/4],{},int({}%4)*8,8);", ctx.stage_name,
+    ctx.Add("{}_ssbo{}[{}>>2]=bitfieldInsert({}_ssbo{}[{}>>2],{},int({}%4)*8,8);", ctx.stage_name,
             binding.U32(), offset_var, ctx.stage_name, binding.U32(), offset_var, value,
             offset_var);
 }
@@ -78,7 +79,7 @@ void EmitWriteStorageS8([[maybe_unused]] EmitContext& ctx,
                         [[maybe_unused]] const IR::Value& offset,
                         [[maybe_unused]] std::string_view value) {
     const auto offset_var{ctx.var_alloc.Consume(offset)};
-    ctx.Add("{}_ssbo{}[{}/4]=bitfieldInsert({}_ssbo{}[{}/4],{},int({}%4)*8,8);", ctx.stage_name,
+    ctx.Add("{}_ssbo{}[{}>>2]=bitfieldInsert({}_ssbo{}[{}>>2],{},int({}%4)*8,8);", ctx.stage_name,
             binding.U32(), offset_var, ctx.stage_name, binding.U32(), offset_var, value,
             offset_var);
 }
@@ -88,7 +89,7 @@ void EmitWriteStorageU16([[maybe_unused]] EmitContext& ctx,
                          [[maybe_unused]] const IR::Value& offset,
                          [[maybe_unused]] std::string_view value) {
     const auto offset_var{ctx.var_alloc.Consume(offset)};
-    ctx.Add("{}_ssbo{}[{}/4]=bitfieldInsert({}_ssbo{}[{}/4],{},int(({}/2)%2)*16,16);",
+    ctx.Add("{}_ssbo{}[{}>>2]=bitfieldInsert({}_ssbo{}[{}>>2],{},int(({}>>1)%2)*16,16);",
             ctx.stage_name, binding.U32(), offset_var, ctx.stage_name, binding.U32(), offset_var,
             value, offset_var);
 }
@@ -98,7 +99,7 @@ void EmitWriteStorageS16([[maybe_unused]] EmitContext& ctx,
                          [[maybe_unused]] const IR::Value& offset,
                          [[maybe_unused]] std::string_view value) {
     const auto offset_var{ctx.var_alloc.Consume(offset)};
-    ctx.Add("{}_ssbo{}[{}/4]=bitfieldInsert({}_ssbo{}[{}/4],{},int(({}/2)%2)*16,16);",
+    ctx.Add("{}_ssbo{}[{}>>2]=bitfieldInsert({}_ssbo{}[{}>>2],{},int(({}>>1)%2)*16,16);",
             ctx.stage_name, binding.U32(), offset_var, ctx.stage_name, binding.U32(), offset_var,
             value, offset_var);
 }
@@ -106,14 +107,14 @@ void EmitWriteStorageS16([[maybe_unused]] EmitContext& ctx,
 void EmitWriteStorage32(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
                         std::string_view value) {
     const auto offset_var{ctx.var_alloc.Consume(offset)};
-    ctx.Add("{}_ssbo{}[{}/4]={};", ctx.stage_name, binding.U32(), offset_var, value);
+    ctx.Add("{}_ssbo{}[{}>>2]={};", ctx.stage_name, binding.U32(), offset_var, value);
 }
 
 void EmitWriteStorage64(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
                         std::string_view value) {
     const auto offset_var{ctx.var_alloc.Consume(offset)};
-    ctx.Add("{}_ssbo{}[{}/4]={}.x;", ctx.stage_name, binding.U32(), offset_var, value);
+    ctx.Add("{}_ssbo{}[{}>>2]={}.x;", ctx.stage_name, binding.U32(), offset_var, value);
-    ctx.Add("{}_ssbo{}[({}+4)/4]={}.y;", ctx.stage_name, binding.U32(), offset_var, value);
+    ctx.Add("{}_ssbo{}[({}+4)>>2]={}.y;", ctx.stage_name, binding.U32(), offset_var, value);
 }
 
 void EmitWriteStorage128([[maybe_unused]] EmitContext& ctx,
@@ -121,9 +122,9 @@ void EmitWriteStorage128([[maybe_unused]] EmitContext& ctx,
                          [[maybe_unused]] const IR::Value& offset,
                          [[maybe_unused]] std::string_view value) {
     const auto offset_var{ctx.var_alloc.Consume(offset)};
-    ctx.Add("{}_ssbo{}[{}/4]={}.x;", ctx.stage_name, binding.U32(), offset_var, value);
+    ctx.Add("{}_ssbo{}[{}>>2]={}.x;", ctx.stage_name, binding.U32(), offset_var, value);
-    ctx.Add("{}_ssbo{}[({}+4)/4]={}.y;", ctx.stage_name, binding.U32(), offset_var, value);
+    ctx.Add("{}_ssbo{}[({}+4)>>2]={}.y;", ctx.stage_name, binding.U32(), offset_var, value);
-    ctx.Add("{}_ssbo{}[({}+8)/4]={}.z;", ctx.stage_name, binding.U32(), offset_var, value);
+    ctx.Add("{}_ssbo{}[({}+8)>>2]={}.z;", ctx.stage_name, binding.U32(), offset_var, value);
-    ctx.Add("{}_ssbo{}[({}+12)/4]={}.w;", ctx.stage_name, binding.U32(), offset_var, value);
+    ctx.Add("{}_ssbo{}[({}+12)>>2]={}.w;", ctx.stage_name, binding.U32(), offset_var, value);
 }
 } // namespace Shader::Backend::GLSL

src/shader_recompiler/backend/glsl/emit_glsl_shared_memory.cpp

@@ -11,70 +11,70 @@
 namespace Shader::Backend::GLSL {
 void EmitLoadSharedU8([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
                       [[maybe_unused]] std::string_view offset) {
-    ctx.AddU32("{}=bitfieldExtract(smem[{}/4],int({}%4)*8,8);", inst, offset, offset);
+    ctx.AddU32("{}=bitfieldExtract(smem[{}>>2],int({}%4)*8,8);", inst, offset, offset);
 }
 
 void EmitLoadSharedS8([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
                       [[maybe_unused]] std::string_view offset) {
-    ctx.AddS32("{}=bitfieldExtract(int(smem[{}/4]),int({}%4)*8,8);", inst, offset, offset);
+    ctx.AddS32("{}=bitfieldExtract(int(smem[{}>>2]),int({}%4)*8,8);", inst, offset, offset);
 }
 
 void EmitLoadSharedU16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
                        [[maybe_unused]] std::string_view offset) {
-    ctx.AddU32("{}=bitfieldExtract(smem[{}/4],int(({}/2)%2)*16,16);", inst, offset, offset);
+    ctx.AddU32("{}=bitfieldExtract(smem[{}>>2],int(({}>>1)%2)*16,16);", inst, offset, offset);
 }
 
 void EmitLoadSharedS16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
                        [[maybe_unused]] std::string_view offset) {
-    ctx.AddS32("{}=bitfieldExtract(int(smem[{}/4]),int(({}/2)%2)*16,16);", inst, offset, offset);
+    ctx.AddS32("{}=bitfieldExtract(int(smem[{}>>2]),int(({}>>1)%2)*16,16);", inst, offset, offset);
 }
 
 void EmitLoadSharedU32([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
                        [[maybe_unused]] std::string_view offset) {
-    ctx.AddU32("{}=smem[{}/4];", inst, offset);
+    ctx.AddU32("{}=smem[{}>>2];", inst, offset);
 }
 
 void EmitLoadSharedU64([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
                        [[maybe_unused]] std::string_view offset) {
-    ctx.AddU32x2("{}=uvec2(smem[{}/4],smem[({}+4)/4]);", inst, offset, offset);
+    ctx.AddU32x2("{}=uvec2(smem[{}>>2],smem[({}+4)>>2]);", inst, offset, offset);
 }
 
 void EmitLoadSharedU128([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
                         [[maybe_unused]] std::string_view offset) {
-    ctx.AddU32x4("{}=uvec4(smem[{}/4],smem[({}+4)/4],smem[({}+8)/4],smem[({}+12)/4]);", inst,
+    ctx.AddU32x4("{}=uvec4(smem[{}>>2],smem[({}+4)>>2],smem[({}+8)>>2],smem[({}+12)>>2]);", inst,
                  offset, offset, offset, offset);
 }
 
 void EmitWriteSharedU8([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] std::string_view offset,
                        [[maybe_unused]] std::string_view value) {
-    ctx.Add("smem[{}/4]=bitfieldInsert(smem[{}/4],{},int({}%4)*8,8);", offset, offset, value,
+    ctx.Add("smem[{}>>2]=bitfieldInsert(smem[{}>>2],{},int({}%4)*8,8);", offset, offset, value,
             offset);
 }
 
 void EmitWriteSharedU16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] std::string_view offset,
                         [[maybe_unused]] std::string_view value) {
-    ctx.Add("smem[{}/4]=bitfieldInsert(smem[{}/4],{},int(({}/2)%2)*16,16);", offset, offset, value,
+    ctx.Add("smem[{}>>2]=bitfieldInsert(smem[{}>>2],{},int(({}>>1)%2)*16,16);", offset, offset,
-            offset);
+            value, offset);
 }
 
 void EmitWriteSharedU32([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] std::string_view offset,
                         [[maybe_unused]] std::string_view value) {
-    ctx.Add("smem[{}/4]={};", offset, value);
+    ctx.Add("smem[{}>>2]={};", offset, value);
 }
 
 void EmitWriteSharedU64([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] std::string_view offset,
                         [[maybe_unused]] std::string_view value) {
-    ctx.Add("smem[{}/4]={}.x;", offset, value);
+    ctx.Add("smem[{}>>2]={}.x;", offset, value);
-    ctx.Add("smem[({}+4)/4]={}.y;", offset, value);
+    ctx.Add("smem[({}+4)>>2]={}.y;", offset, value);
 }
 
 void EmitWriteSharedU128([[maybe_unused]] EmitContext& ctx,
                          [[maybe_unused]] std::string_view offset,
                          [[maybe_unused]] std::string_view value) {
-    ctx.Add("smem[{}/4]={}.x;", offset, value);
+    ctx.Add("smem[{}>>2]={}.x;", offset, value);
-    ctx.Add("smem[({}+4)/4]={}.y;", offset, value);
+    ctx.Add("smem[({}+4)>>2]={}.y;", offset, value);
-    ctx.Add("smem[({}+8)/4]={}.z;", offset, value);
+    ctx.Add("smem[({}+8)>>2]={}.z;", offset, value);
-    ctx.Add("smem[({}+12)/4]={}.w;", offset, value);
+    ctx.Add("smem[({}+12)>>2]={}.w;", offset, value);
 }
 
 } // namespace Shader::Backend::GLSL

src/shader_recompiler/backend/glsl/emit_glsl_warp.cpp

@@ -112,7 +112,7 @@ void EmitShuffleIndex(EmitContext& ctx, IR::Inst& inst, std::string_view value,
     const auto src_thread_id{fmt::format("({})|({})", lhs, min_thread_id)};
     ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id);
     SetInBoundsFlag(ctx, inst);
-    ctx.AddU32("{}=shfl_in_bounds?{}:{};", inst, value, src_thread_id);
+    ctx.AddU32("{}=shfl_in_bounds?readInvocationARB({},{}):{};", inst, value, src_thread_id, value);
 }
 
 void EmitShuffleUp(EmitContext& ctx, IR::Inst& inst, std::string_view value, std::string_view index,
@@ -122,7 +122,7 @@ void EmitShuffleUp(EmitContext& ctx, IR::Inst& inst, std::string_view value, std
     const auto src_thread_id{fmt::format("({}-{})", thread_id, index)};
     ctx.Add("shfl_in_bounds=int({})>=int({});", src_thread_id, max_thread_id);
     SetInBoundsFlag(ctx, inst);
-    ctx.AddU32("{}=shfl_in_bounds?{}:{};", inst, value, src_thread_id);
+    ctx.AddU32("{}=shfl_in_bounds?readInvocationARB({},{}):{};", inst, value, src_thread_id, value);
 }
 
 void EmitShuffleDown(EmitContext& ctx, IR::Inst& inst, std::string_view value,
@@ -133,7 +133,7 @@ void EmitShuffleDown(EmitContext& ctx, IR::Inst& inst, std::string_view value,
     const auto src_thread_id{fmt::format("({}+{})", thread_id, index)};
     ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id);
     SetInBoundsFlag(ctx, inst);
-    ctx.AddU32("{}=shfl_in_bounds?{}:{};", inst, value, src_thread_id);
+    ctx.AddU32("{}=shfl_in_bounds?readInvocationARB({},{}):{};", inst, value, src_thread_id, value);
 }
 
 void EmitShuffleButterfly(EmitContext& ctx, IR::Inst& inst, std::string_view value,
@@ -144,7 +144,7 @@ void EmitShuffleButterfly(EmitContext& ctx, IR::Inst& inst, std::string_view val
     const auto src_thread_id{fmt::format("({}^{})", thread_id, index)};
     ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id);
     SetInBoundsFlag(ctx, inst);
-    ctx.AddU32("{}=shfl_in_bounds?{}:{};", inst, value, src_thread_id);
+    ctx.AddU32("{}=shfl_in_bounds?readInvocationARB({},{}):{};", inst, value, src_thread_id, value);
 }
 
 void EmitFSwizzleAdd([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,