[hlsl] Update DecomposeStorageAccess to conditionalize Convert creation.

When generating the decomposed indices we only need to create a `Convert` instruction if the types don't match. Bug: 379684039 Change-Id: Iac61a657c55239387b37528327bb76a2d3690a87 Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/215634 Commit-Queue: James Price <[email protected]> Reviewed-by: James Price <[email protected]>
google · Nov 19, 2024 · a1ee883 · a1ee883
1 parent 3d64565
commit a1ee883
Show file tree

Hide file tree

Showing 73 changed files with 866 additions and 609 deletions.
diff --git a/src/tint/lang/hlsl/writer/access_test.cc b/src/tint/lang/hlsl/writer/access_test.cc
@@ -718,9 +718,8 @@ void foo() {
   uint v_2 = 0u;
   sb.GetDimensions(v_2);
   uint v_3 = (((v_2 - 16u) / 128u) - 1u);
-  uint v_4 = (uint(min(uint(i), v_3)) * 128u);
-  uint v_5 = (uint(min(v, 2u)) * 32u);
-  float x = asfloat(sb.Load((((48u + v_4) + v_5) + (uint(min(uint(v_1), 2u)) * 4u))));
+  uint v_4 = (min(uint(i), v_3) * 128u);
+  float x = asfloat(sb.Load((((48u + v_4) + (min(v, 2u) * 32u)) + (min(uint(v_1), 2u) * 4u))));
 }
 
 )");
@@ -763,9 +762,7 @@ void foo() {
   uint j = 1u;
   uint v = 0u;
   sb.GetDimensions(v);
-  uint v_1 = min(j, 2u);
-  uint v_2 = (uint(min(4u, (((v - 16u) / 128u) - 1u))) * 128u);
-  float x = asfloat(sb.Load(((56u + v_2) + (uint(v_1) * 32u))));
+  float x = asfloat(sb.Load(((56u + (min(4u, (((v - 16u) / 128u) - 1u)) * 128u)) + (min(j, 2u) * 32u))));
 }
 
 )");

diff --git a/src/tint/lang/hlsl/writer/arraylength_test.cc b/src/tint/lang/hlsl/writer/arraylength_test.cc
@@ -247,7 +247,7 @@ RWByteAddressBuffer dst : register(u1);
 void foo() {
   uint v = 0u;
   dst.GetDimensions(v);
-  dst.Store((0u + (uint(min(0u, ((v / 4u) - 1u))) * 4u)), 123u);
+  dst.Store((0u + (min(0u, ((v / 4u) - 1u)) * 4u)), 123u);
 }
 
 )");
@@ -275,7 +275,7 @@ cbuffer cbuffer_tint_storage_buffer_sizes : register(b0, space30) {
   uint4 tint_storage_buffer_sizes[1];
 };
 void foo() {
-  dst.Store((0u + (uint(min(0u, ((tint_storage_buffer_sizes[0u].x / 4u) - 1u))) * 4u)), 123u);
+  dst.Store((0u + (min(0u, ((tint_storage_buffer_sizes[0u].x / 4u) - 1u)) * 4u)), 123u);
 }
 
 )");

diff --git a/src/tint/lang/hlsl/writer/raise/decompose_storage_access.cc b/src/tint/lang/hlsl/writer/raise/decompose_storage_access.cc
@@ -359,8 +359,11 @@ struct State {
                 offset->byte_offset += idx_value->Value()->ValueAs<uint32_t>() * elm_size;
             },
             [&](core::ir::Value* val) {
-                offset->expr.Push(
-                    b.Multiply(ty.u32(), b.Convert(ty.u32(), val), u32(elm_size))->Result(0));
+                auto* idx = val;
+                if (val->Type() != ty.u32()) {
+                    idx = b.Convert(ty.u32(), val)->Result(0);
+                }
+                offset->expr.Push(b.Multiply(ty.u32(), idx, u32(elm_size))->Result(0));
             },
             TINT_ICE_ON_NO_MATCH);
     }

diff --git a/src/tint/lang/hlsl/writer/raise/decompose_storage_access_test.cc b/src/tint/lang/hlsl/writer/raise/decompose_storage_access_test.cc
@@ -1183,16 +1183,15 @@ SB = struct @align(16) {
     %8:i32 = load %k
     %9:u32 = convert %4
     %10:u32 = mul %9, 128u
-    %11:u32 = convert %6
-    %12:u32 = mul %11, 32u
-    %13:u32 = convert %8
-    %14:u32 = mul %13, 4u
-    %15:u32 = add 48u, %10
-    %16:u32 = add %15, %12
-    %17:u32 = add %16, %14
-    %18:u32 = %sb.Load %17
-    %19:f32 = bitcast %18
-    %x:f32 = let %19
+    %11:u32 = mul %6, 32u
+    %12:u32 = convert %8
+    %13:u32 = mul %12, 4u
+    %14:u32 = add 48u, %10
+    %15:u32 = add %14, %11
+    %16:u32 = add %15, %13
+    %17:u32 = %sb.Load %16
+    %18:f32 = bitcast %17
+    %x:f32 = let %18
     ret
   }
 }
@@ -1294,12 +1293,11 @@ SB = struct @align(16) {
   $B2: {
     %j:ptr<function, u32, read_write> = var, 1u
     %4:u32 = load %j
-    %5:u32 = convert %4
-    %6:u32 = mul %5, 32u
-    %7:u32 = add 568u, %6
-    %8:u32 = %sb.Load %7
-    %9:f32 = bitcast %8
-    %x:f32 = let %9
+    %5:u32 = mul %4, 32u
+    %6:u32 = add 568u, %5
+    %7:u32 = %sb.Load %6
+    %8:f32 = bitcast %7
+    %x:f32 = let %8
     ret
   }
 }
@@ -1436,15 +1434,13 @@ S2 = struct @align(16) {
 
 %foo = @fragment func(%3:u32):void {
   $B2: {
-    %4:u32 = convert %3
-    %5:u32 = mul %4, 32u
-    %6:u32 = convert %3
-    %7:u32 = mul %6, 4u
-    %8:ptr<function, i32, read_write> = var, 0i
-    %9:u32 = add 16u, %5
-    %10:u32 = add %9, %7
-    %11:i32 = convert %10
-    %12:void = %v.InterlockedExchange %11, 123i, %8
+    %4:u32 = mul %3, 32u
+    %5:u32 = mul %3, 4u
+    %6:ptr<function, i32, read_write> = var, 0i
+    %7:u32 = add 16u, %4
+    %8:u32 = add %7, %5
+    %9:i32 = convert %8
+    %10:void = %v.InterlockedExchange %9, 123i, %6
     ret
   }
 }
@@ -1629,17 +1625,15 @@ S2 = struct @align(16) {
 
 %foo = @fragment func(%3:u32):void {
   $B2: {
-    %4:u32 = convert %3
-    %5:u32 = mul %4, 32u
-    %6:u32 = convert %3
-    %7:u32 = mul %6, 4u
-    %8:ptr<function, i32, read_write> = var, 0i
-    %9:u32 = add 16u, %5
-    %10:u32 = add %9, %7
-    %11:i32 = convert %10
-    %12:void = %v.InterlockedOr %11, 0i, %8
-    %13:i32 = load %8
-    %x:i32 = let %13
+    %4:u32 = mul %3, 32u
+    %5:u32 = mul %3, 4u
+    %6:ptr<function, i32, read_write> = var, 0i
+    %7:u32 = add 16u, %4
+    %8:u32 = add %7, %5
+    %9:i32 = convert %8
+    %10:void = %v.InterlockedOr %9, 0i, %6
+    %11:i32 = load %6
+    %x:i32 = let %11
     ret
   }
 }
@@ -1828,18 +1822,16 @@ S2 = struct @align(16) {
 
 %foo = @fragment func(%3:u32):void {
   $B2: {
-    %4:u32 = convert %3
-    %5:u32 = mul %4, 32u
-    %6:u32 = convert %3
-    %7:u32 = mul %6, 4u
-    %8:ptr<function, i32, read_write> = var, 0i
-    %9:i32 = sub 0i, 123i
-    %10:u32 = add 16u, %5
-    %11:u32 = add %10, %7
-    %12:i32 = convert %11
-    %13:void = %v.InterlockedAdd %12, %9, %8
-    %14:i32 = load %8
-    %x:i32 = let %14
+    %4:u32 = mul %3, 32u
+    %5:u32 = mul %3, 4u
+    %6:ptr<function, i32, read_write> = var, 0i
+    %7:i32 = sub 0i, 123i
+    %8:u32 = add 16u, %4
+    %9:u32 = add %8, %5
+    %10:i32 = convert %9
+    %11:void = %v.InterlockedAdd %10, %7, %6
+    %12:i32 = load %6
+    %x:i32 = let %12
     ret
   }
 }
@@ -2053,19 +2045,17 @@ __atomic_compare_exchange_result_i32 = struct @align(4) {
 
 %foo = @fragment func(%3:u32):void {
   $B2: {
-    %4:u32 = convert %3
-    %5:u32 = mul %4, 32u
-    %6:u32 = convert %3
-    %7:u32 = mul %6, 4u
-    %8:ptr<function, i32, read_write> = var, 0i
-    %9:u32 = add 16u, %5
-    %10:u32 = add %9, %7
-    %11:i32 = convert %10
-    %12:void = %v.InterlockedCompareExchange %11, 123i, 345i, %8
-    %13:i32 = load %8
-    %14:bool = eq %13, 123i
-    %15:__atomic_compare_exchange_result_i32 = construct %13, %14
-    %x:__atomic_compare_exchange_result_i32 = let %15
+    %4:u32 = mul %3, 32u
+    %5:u32 = mul %3, 4u
+    %6:ptr<function, i32, read_write> = var, 0i
+    %7:u32 = add 16u, %4
+    %8:u32 = add %7, %5
+    %9:i32 = convert %8
+    %10:void = %v.InterlockedCompareExchange %9, 123i, 345i, %6
+    %11:i32 = load %6
+    %12:bool = eq %11, 123i
+    %13:__atomic_compare_exchange_result_i32 = construct %11, %12
+    %x:__atomic_compare_exchange_result_i32 = let %13
     ret
   }
 }
@@ -2331,18 +2321,16 @@ S2 = struct @align(4) {
 
 %foo = @fragment func(%3:u32):void {
   $B2: {
-    %4:u32 = convert %3
-    %5:u32 = mul %4, 12u
-    %6:u32 = convert %3
-    %7:u32 = mul %6, 4u
-    %8:ptr<function, u32, read_write> = var, 0u
-    %9:u32 = add 0u, %5
-    %10:u32 = add %9, %7
-    %11:u32 = convert %10
-    %12:void = %v.)" +
-                  std::string(param.interlock) + R"( %11, 123u, %8
-    %13:u32 = load %8
-    %x:u32 = let %13
+    %4:u32 = mul %3, 12u
+    %5:u32 = mul %3, 4u
+    %6:ptr<function, u32, read_write> = var, 0u
+    %7:u32 = add 0u, %4
+    %8:u32 = add %7, %5
+    %9:u32 = convert %8
+    %10:void = %v.)" +
+                  std::string(param.interlock) + R"( %9, 123u, %6
+    %11:u32 = load %6
+    %x:u32 = let %11
     ret
   }
 }

diff --git a/test/tint/buffer/storage/dynamic_index/read.wgsl.expected.ir.dxc.hlsl b/test/tint/buffer/storage/dynamic_index/read.wgsl.expected.ir.dxc.hlsl
@@ -68,28 +68,28 @@ float2x2 v_12(uint offset) {
 }
 
 void main_inner(uint idx) {
-  float scalar_f32 = asfloat(sb.Load((0u + (uint(idx) * 544u))));
-  int scalar_i32 = asint(sb.Load((4u + (uint(idx) * 544u))));
-  uint scalar_u32 = sb.Load((8u + (uint(idx) * 544u)));
-  float2 vec2_f32 = asfloat(sb.Load2((16u + (uint(idx) * 544u))));
-  int2 vec2_i32 = asint(sb.Load2((24u + (uint(idx) * 544u))));
-  uint2 vec2_u32 = sb.Load2((32u + (uint(idx) * 544u)));
-  float3 vec3_f32 = asfloat(sb.Load3((48u + (uint(idx) * 544u))));
-  int3 vec3_i32 = asint(sb.Load3((64u + (uint(idx) * 544u))));
-  uint3 vec3_u32 = sb.Load3((80u + (uint(idx) * 544u)));
-  float4 vec4_f32 = asfloat(sb.Load4((96u + (uint(idx) * 544u))));
-  int4 vec4_i32 = asint(sb.Load4((112u + (uint(idx) * 544u))));
-  uint4 vec4_u32 = sb.Load4((128u + (uint(idx) * 544u)));
-  float2x2 mat2x2_f32 = v_12((144u + (uint(idx) * 544u)));
-  float2x3 mat2x3_f32 = v_11((160u + (uint(idx) * 544u)));
-  float2x4 mat2x4_f32 = v_10((192u + (uint(idx) * 544u)));
-  float3x2 mat3x2_f32 = v_9((224u + (uint(idx) * 544u)));
-  float3x3 mat3x3_f32 = v_8((256u + (uint(idx) * 544u)));
-  float3x4 mat3x4_f32 = v_7((304u + (uint(idx) * 544u)));
-  float4x2 mat4x2_f32 = v_6((352u + (uint(idx) * 544u)));
-  float4x3 mat4x3_f32 = v_5((384u + (uint(idx) * 544u)));
-  float4x4 mat4x4_f32 = v_4((448u + (uint(idx) * 544u)));
-  float3 arr2_vec3_f32[2] = v((512u + (uint(idx) * 544u)));
+  float scalar_f32 = asfloat(sb.Load((0u + (idx * 544u))));
+  int scalar_i32 = asint(sb.Load((4u + (idx * 544u))));
+  uint scalar_u32 = sb.Load((8u + (idx * 544u)));
+  float2 vec2_f32 = asfloat(sb.Load2((16u + (idx * 544u))));
+  int2 vec2_i32 = asint(sb.Load2((24u + (idx * 544u))));
+  uint2 vec2_u32 = sb.Load2((32u + (idx * 544u)));
+  float3 vec3_f32 = asfloat(sb.Load3((48u + (idx * 544u))));
+  int3 vec3_i32 = asint(sb.Load3((64u + (idx * 544u))));
+  uint3 vec3_u32 = sb.Load3((80u + (idx * 544u)));
+  float4 vec4_f32 = asfloat(sb.Load4((96u + (idx * 544u))));
+  int4 vec4_i32 = asint(sb.Load4((112u + (idx * 544u))));
+  uint4 vec4_u32 = sb.Load4((128u + (idx * 544u)));
+  float2x2 mat2x2_f32 = v_12((144u + (idx * 544u)));
+  float2x3 mat2x3_f32 = v_11((160u + (idx * 544u)));
+  float2x4 mat2x4_f32 = v_10((192u + (idx * 544u)));
+  float3x2 mat3x2_f32 = v_9((224u + (idx * 544u)));
+  float3x3 mat3x3_f32 = v_8((256u + (idx * 544u)));
+  float3x4 mat3x4_f32 = v_7((304u + (idx * 544u)));
+  float4x2 mat4x2_f32 = v_6((352u + (idx * 544u)));
+  float4x3 mat4x3_f32 = v_5((384u + (idx * 544u)));
+  float4x4 mat4x4_f32 = v_4((448u + (idx * 544u)));
+  float3 arr2_vec3_f32[2] = v((512u + (idx * 544u)));
   int v_13 = (tint_f32_to_i32(scalar_f32) + scalar_i32);
   int v_14 = (v_13 + int(scalar_u32));
   int v_15 = ((v_14 + tint_f32_to_i32(vec2_f32.x)) + vec2_i32.x);

diff --git a/test/tint/buffer/storage/dynamic_index/read.wgsl.expected.ir.fxc.hlsl b/test/tint/buffer/storage/dynamic_index/read.wgsl.expected.ir.fxc.hlsl
@@ -68,28 +68,28 @@ float2x2 v_12(uint offset) {
 }
 
 void main_inner(uint idx) {
-  float scalar_f32 = asfloat(sb.Load((0u + (uint(idx) * 544u))));
-  int scalar_i32 = asint(sb.Load((4u + (uint(idx) * 544u))));
-  uint scalar_u32 = sb.Load((8u + (uint(idx) * 544u)));
-  float2 vec2_f32 = asfloat(sb.Load2((16u + (uint(idx) * 544u))));
-  int2 vec2_i32 = asint(sb.Load2((24u + (uint(idx) * 544u))));
-  uint2 vec2_u32 = sb.Load2((32u + (uint(idx) * 544u)));
-  float3 vec3_f32 = asfloat(sb.Load3((48u + (uint(idx) * 544u))));
-  int3 vec3_i32 = asint(sb.Load3((64u + (uint(idx) * 544u))));
-  uint3 vec3_u32 = sb.Load3((80u + (uint(idx) * 544u)));
-  float4 vec4_f32 = asfloat(sb.Load4((96u + (uint(idx) * 544u))));
-  int4 vec4_i32 = asint(sb.Load4((112u + (uint(idx) * 544u))));
-  uint4 vec4_u32 = sb.Load4((128u + (uint(idx) * 544u)));
-  float2x2 mat2x2_f32 = v_12((144u + (uint(idx) * 544u)));
-  float2x3 mat2x3_f32 = v_11((160u + (uint(idx) * 544u)));
-  float2x4 mat2x4_f32 = v_10((192u + (uint(idx) * 544u)));
-  float3x2 mat3x2_f32 = v_9((224u + (uint(idx) * 544u)));
-  float3x3 mat3x3_f32 = v_8((256u + (uint(idx) * 544u)));
-  float3x4 mat3x4_f32 = v_7((304u + (uint(idx) * 544u)));
-  float4x2 mat4x2_f32 = v_6((352u + (uint(idx) * 544u)));
-  float4x3 mat4x3_f32 = v_5((384u + (uint(idx) * 544u)));
-  float4x4 mat4x4_f32 = v_4((448u + (uint(idx) * 544u)));
-  float3 arr2_vec3_f32[2] = v((512u + (uint(idx) * 544u)));
+  float scalar_f32 = asfloat(sb.Load((0u + (idx * 544u))));
+  int scalar_i32 = asint(sb.Load((4u + (idx * 544u))));
+  uint scalar_u32 = sb.Load((8u + (idx * 544u)));
+  float2 vec2_f32 = asfloat(sb.Load2((16u + (idx * 544u))));
+  int2 vec2_i32 = asint(sb.Load2((24u + (idx * 544u))));
+  uint2 vec2_u32 = sb.Load2((32u + (idx * 544u)));
+  float3 vec3_f32 = asfloat(sb.Load3((48u + (idx * 544u))));
+  int3 vec3_i32 = asint(sb.Load3((64u + (idx * 544u))));
+  uint3 vec3_u32 = sb.Load3((80u + (idx * 544u)));
+  float4 vec4_f32 = asfloat(sb.Load4((96u + (idx * 544u))));
+  int4 vec4_i32 = asint(sb.Load4((112u + (idx * 544u))));
+  uint4 vec4_u32 = sb.Load4((128u + (idx * 544u)));
+  float2x2 mat2x2_f32 = v_12((144u + (idx * 544u)));
+  float2x3 mat2x3_f32 = v_11((160u + (idx * 544u)));
+  float2x4 mat2x4_f32 = v_10((192u + (idx * 544u)));
+  float3x2 mat3x2_f32 = v_9((224u + (idx * 544u)));
+  float3x3 mat3x3_f32 = v_8((256u + (idx * 544u)));
+  float3x4 mat3x4_f32 = v_7((304u + (idx * 544u)));
+  float4x2 mat4x2_f32 = v_6((352u + (idx * 544u)));
+  float4x3 mat4x3_f32 = v_5((384u + (idx * 544u)));
+  float4x4 mat4x4_f32 = v_4((448u + (idx * 544u)));
+  float3 arr2_vec3_f32[2] = v((512u + (idx * 544u)));
   int v_13 = (tint_f32_to_i32(scalar_f32) + scalar_i32);
   int v_14 = (v_13 + int(scalar_u32));
   int v_15 = ((v_14 + tint_f32_to_i32(vec2_f32.x)) + vec2_i32.x);