Update MSL generator to handle casts of packed types.

Currently in the MSL backend we cast int values to uint in order to get the correct WGSL behaviour for over/under flow. This fails in the case of host shareable buffers as they use `packed` types which need to get cast to the non-packed version first. Bug: tint:1677 Change-Id: I57b70abaa8ca614472a26d63f19c1aef2bd64668 Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/103986 Reviewed-by: Ben Clayton <bclayton@google.com> Kokoro: Kokoro <noreply+kokoro@google.com> Commit-Queue: Dan Sinclair <dsinclair@chromium.org>
2022-09-29 19:44:58 +00:00 · 2022-09-29 19:44:58 +00:00 · 2bcade246a
parent d1a5f93630
commit 2bcade246a
27 changed files with 197 additions and 30 deletions
--- a/src/tint/writer/msl/generator_impl.cc
+++ b/src/tint/writer/msl/generator_impl.cc
@ -517,8 +517,18 @@ bool GeneratorImpl::EmitBinary(std::ostream& out, const ast::BinaryExpression* e
        ScopedParen sp(out);
        {
            ScopedBitCast lhs_uint_cast(this, out, lhs_type, unsigned_type_of(target_type));
-            if (!EmitExpression(out, expr->lhs)) {
-                return false;
+
+            // In case the type is packed, cast to our own type in order to remove the packing.
+            // Otherwise, this just casts to itself.
+            if (lhs_type->is_signed_integer_vector()) {
+                ScopedBitCast lhs_self_cast(this, out, lhs_type, lhs_type);
+                if (!EmitExpression(out, expr->lhs)) {
+                    return false;
+                }
+            } else {
+                if (!EmitExpression(out, expr->lhs)) {
+                    return false;
+                }
            }
        }
        if (!emit_op()) {
@ -526,8 +536,18 @@ bool GeneratorImpl::EmitBinary(std::ostream& out, const ast::BinaryExpression* e
        }
        {
            ScopedBitCast rhs_uint_cast(this, out, rhs_type, unsigned_type_of(target_type));
-            if (!EmitExpression(out, expr->rhs)) {
-                return false;
+
+            // In case the type is packed, cast to our own type in order to remove the packing.
+            // Otherwise, this just casts to itself.
+            if (rhs_type->is_signed_integer_vector()) {
+                ScopedBitCast rhs_self_cast(this, out, rhs_type, rhs_type);
+                if (!EmitExpression(out, expr->rhs)) {
+                    return false;
+                }
+            } else {
+                if (!EmitExpression(out, expr->rhs)) {
+                    return false;
+                }
            }
        }
        return true;
--- a/test/tint/bug/tint/1121.wgsl.expected.msl
+++ b/test/tint/bug/tint/1121.wgsl.expected.msl
@ -88,7 +88,7 @@ void tint_symbol_inner(uint3 GlobalInvocationID, const constant Config* const ti
    for(int x_1 = 0; (x_1 < TILE_COUNT_X); x_1 = as_type<int>((as_type<uint>(x_1) + as_type<uint>(1)))) {
      int2 tilePixel0Idx = int2(as_type<int>((as_type<uint>(x_1) * as_type<uint>(TILE_SIZE))), as_type<int>((as_type<uint>(y_1) * as_type<uint>(TILE_SIZE))));
      float2 floorCoord = (((2.0f * float2(tilePixel0Idx)) / float4((*(tint_symbol_3)).fullScreenSize).xy) - float2(1.0f));
-      float2 ceilCoord = (((2.0f * float2(as_type<int2>((as_type<uint2>(tilePixel0Idx) + as_type<uint2>(int2(TILE_SIZE)))))) / float4((*(tint_symbol_3)).fullScreenSize).xy) - float2(1.0f));
+      float2 ceilCoord = (((2.0f * float2(as_type<int2>((as_type<uint2>(as_type<int2>(tilePixel0Idx)) + as_type<uint2>(as_type<int2>(int2(TILE_SIZE))))))) / float4((*(tint_symbol_3)).fullScreenSize).xy) - float2(1.0f));
      float2 viewFloorCoord = float2((((-(viewNear) * floorCoord[0]) - (M[2][0] * viewNear)) / M[0][0]), (((-(viewNear) * floorCoord[1]) - (M[2][1] * viewNear)) / M[1][1]));
      float2 viewCeilCoord = float2((((-(viewNear) * ceilCoord[0]) - (M[2][0] * viewNear)) / M[0][0]), (((-(viewNear) * ceilCoord[1]) - (M[2][1] * viewNear)) / M[1][1]));
      frustumPlanes[0] = float4(1.0f, 0.0f, (-(viewFloorCoord[0]) / viewNear), 0.0f);
--- a/test/tint/bug/tint/1520.spvasm.expected.msl
+++ b/test/tint/bug/tint/1520.spvasm.expected.msl
@ -45,13 +45,13 @@ bool test_int_S1_c0_b(const constant UniformBuffer* const tint_symbol_5) {
  ok = x_41;
  int4 const x_44 = int4(x_27, x_27, x_27, x_27);
  val = x_44;
-  int4 const x_47 = as_type<int4>((as_type<uint4>(x_44) + as_type<uint4>(int4(1))));
+  int4 const x_47 = as_type<int4>((as_type<uint4>(as_type<int4>(x_44)) + as_type<uint4>(as_type<int4>(int4(1)))));
  val = x_47;
-  int4 const x_48 = as_type<int4>((as_type<uint4>(x_47) - as_type<uint4>(int4(1))));
+  int4 const x_48 = as_type<int4>((as_type<uint4>(as_type<int4>(x_47)) - as_type<uint4>(as_type<int4>(int4(1)))));
  val = x_48;
-  int4 const x_49 = as_type<int4>((as_type<uint4>(x_48) + as_type<uint4>(int4(1))));
+  int4 const x_49 = as_type<int4>((as_type<uint4>(as_type<int4>(x_48)) + as_type<uint4>(as_type<int4>(int4(1)))));
  val = x_49;
-  int4 const x_50 = as_type<int4>((as_type<uint4>(x_49) - as_type<uint4>(int4(1))));
+  int4 const x_50 = as_type<int4>((as_type<uint4>(as_type<int4>(x_49)) - as_type<uint4>(as_type<int4>(int4(1)))));
  val = x_50;
  x_55 = false;
  if (x_41) {
@ -59,11 +59,11 @@ bool test_int_S1_c0_b(const constant UniformBuffer* const tint_symbol_5) {
    x_55 = x_54;
  }
  ok = x_55;
-  int4 const x_58 = as_type<int4>((as_type<uint4>(x_50) * as_type<uint4>(int4(2))));
+  int4 const x_58 = as_type<int4>((as_type<uint4>(as_type<int4>(x_50)) * as_type<uint4>(as_type<int4>(int4(2)))));
  val = x_58;
  int4 const x_59 = (x_58 / int4(2));
  val = x_59;
-  int4 const x_60 = as_type<int4>((as_type<uint4>(x_59) * as_type<uint4>(int4(2))));
+  int4 const x_60 = as_type<int4>((as_type<uint4>(as_type<int4>(x_59)) * as_type<uint4>(as_type<int4>(int4(2)))));
  val = x_60;
  int4 const x_61 = (x_60 / int4(2));
  val = x_61;
--- a/test/tint/bug/tint/1677.wgsl
+++ b/test/tint/bug/tint/1677.wgsl
@ -0,0 +1,10 @@
+struct Input {
+  position : vec3<i32>,
+}
+
+@group(0) @binding(0) var<storage, read> input : Input;
+
+@compute @workgroup_size(1, 1, 1)
+fn main(@builtin(global_invocation_id) id : vec3<u32>) {
+  let pos = input.position - vec3<i32>(0);
+}
--- a/test/tint/bug/tint/1677.wgsl.expected.dxc.hlsl
+++ b/test/tint/bug/tint/1677.wgsl.expected.dxc.hlsl
@ -0,0 +1,15 @@
+ByteAddressBuffer input : register(t0, space0);
+
+struct tint_symbol_1 {
+  uint3 id : SV_DispatchThreadID;
+};
+
+void main_inner(uint3 id) {
+  const int3 pos = (asint(input.Load3(0u)) - (0).xxx);
+}
+
+[numthreads(1, 1, 1)]
+void main(tint_symbol_1 tint_symbol) {
+  main_inner(tint_symbol.id);
+  return;
+}
--- a/test/tint/bug/tint/1677.wgsl.expected.fxc.hlsl
+++ b/test/tint/bug/tint/1677.wgsl.expected.fxc.hlsl
@ -0,0 +1,15 @@
+ByteAddressBuffer input : register(t0, space0);
+
+struct tint_symbol_1 {
+  uint3 id : SV_DispatchThreadID;
+};
+
+void main_inner(uint3 id) {
+  const int3 pos = (asint(input.Load3(0u)) - (0).xxx);
+}
+
+[numthreads(1, 1, 1)]
+void main(tint_symbol_1 tint_symbol) {
+  main_inner(tint_symbol.id);
+  return;
+}
--- a/test/tint/bug/tint/1677.wgsl.expected.glsl
+++ b/test/tint/bug/tint/1677.wgsl.expected.glsl
@ -0,0 +1,16 @@
+#version 310 es
+
+layout(binding = 0, std430) buffer Input_ssbo {
+  ivec3 position;
+  uint pad;
+} tint_symbol;
+
+void tint_symbol_1(uvec3 id) {
+  ivec3 pos = (tint_symbol.position - ivec3(0));
+}
+
+layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
+void main() {
+  tint_symbol_1(gl_GlobalInvocationID);
+  return;
+}
--- a/test/tint/bug/tint/1677.wgsl.expected.msl
+++ b/test/tint/bug/tint/1677.wgsl.expected.msl
@ -0,0 +1,30 @@
+#include <metal_stdlib>
+
+using namespace metal;
+
+template<typename T, size_t N>
+struct tint_array {
+    const constant T& operator[](size_t i) const constant { return elements[i]; }
+    device T& operator[](size_t i) device { return elements[i]; }
+    const device T& operator[](size_t i) const device { return elements[i]; }
+    thread T& operator[](size_t i) thread { return elements[i]; }
+    const thread T& operator[](size_t i) const thread { return elements[i]; }
+    threadgroup T& operator[](size_t i) threadgroup { return elements[i]; }
+    const threadgroup T& operator[](size_t i) const threadgroup { return elements[i]; }
+    T elements[N];
+};
+
+struct Input {
+  /* 0x0000 */ packed_int3 position;
+  /* 0x000c */ tint_array<int8_t, 4> tint_pad;
+};
+
+void tint_symbol_inner(uint3 id, const device Input* const tint_symbol_1) {
+  int3 const pos = as_type<int3>((as_type<uint3>(as_type<int3>((*(tint_symbol_1)).position)) - as_type<uint3>(as_type<int3>(int3(0)))));
+}
+
+kernel void tint_symbol(const device Input* tint_symbol_2 [[buffer(0)]], uint3 id [[thread_position_in_grid]]) {
+  tint_symbol_inner(id, tint_symbol_2);
+  return;
+}
+
--- a/test/tint/bug/tint/1677.wgsl.expected.spvasm
+++ b/test/tint/bug/tint/1677.wgsl.expected.spvasm
@ -0,0 +1,51 @@
+; SPIR-V
+; Version: 1.3
+; Generator: Google Tint Compiler; 0
+; Bound: 26
+; Schema: 0
+               OpCapability Shader
+               OpMemoryModel Logical GLSL450
+               OpEntryPoint GLCompute %main "main" %id_1
+               OpExecutionMode %main LocalSize 1 1 1
+               OpName %id_1 "id_1"
+               OpName %Input "Input"
+               OpMemberName %Input 0 "position"
+               OpName %input "input"
+               OpName %main_inner "main_inner"
+               OpName %id "id"
+               OpName %main "main"
+               OpDecorate %id_1 BuiltIn GlobalInvocationId
+               OpDecorate %Input Block
+               OpMemberDecorate %Input 0 Offset 0
+               OpDecorate %input NonWritable
+               OpDecorate %input DescriptorSet 0
+               OpDecorate %input Binding 0
+       %uint = OpTypeInt 32 0
+     %v3uint = OpTypeVector %uint 3
+%_ptr_Input_v3uint = OpTypePointer Input %v3uint
+       %id_1 = OpVariable %_ptr_Input_v3uint Input
+        %int = OpTypeInt 32 1
+      %v3int = OpTypeVector %int 3
+      %Input = OpTypeStruct %v3int
+%_ptr_StorageBuffer_Input = OpTypePointer StorageBuffer %Input
+      %input = OpVariable %_ptr_StorageBuffer_Input StorageBuffer
+       %void = OpTypeVoid
+         %10 = OpTypeFunction %void %v3uint
+     %uint_0 = OpConstant %uint 0
+%_ptr_StorageBuffer_v3int = OpTypePointer StorageBuffer %v3int
+         %19 = OpConstantNull %v3int
+         %21 = OpTypeFunction %void
+ %main_inner = OpFunction %void None %10
+         %id = OpFunctionParameter %v3uint
+         %14 = OpLabel
+         %17 = OpAccessChain %_ptr_StorageBuffer_v3int %input %uint_0
+         %18 = OpLoad %v3int %17
+         %20 = OpISub %v3int %18 %19
+               OpReturn
+               OpFunctionEnd
+       %main = OpFunction %void None %21
+         %23 = OpLabel
+         %25 = OpLoad %v3uint %id_1
+         %24 = OpFunctionCall %void %main_inner %25
+               OpReturn
+               OpFunctionEnd
--- a/test/tint/bug/tint/1677.wgsl.expected.wgsl
+++ b/test/tint/bug/tint/1677.wgsl.expected.wgsl
@ -0,0 +1,10 @@
+struct Input {
+  position : vec3<i32>,
+}
+
+@group(0) @binding(0) var<storage, read> input : Input;
+
+@compute @workgroup_size(1, 1, 1)
+fn main(@builtin(global_invocation_id) id : vec3<u32>) {
+  let pos = (input.position - vec3<i32>(0));
+}
--- a/test/tint/bug/tint/942.wgsl.expected.msl
+++ b/test/tint/bug/tint/942.wgsl.expected.msl
@ -32,10 +32,10 @@ void tint_symbol_inner(uint3 WorkGroupID, uint3 LocalInvocationID, uint local_in
  threadgroup_barrier(mem_flags::mem_threadgroup);
  uint const filterOffset = (((*(tint_symbol_2)).filterDim - 1u) / 2u);
  int2 const dims = int2(tint_symbol_3.get_width(0), tint_symbol_3.get_height(0));
-  int2 const baseIndex = as_type<int2>((as_type<uint2>(int2(((uint3(WorkGroupID).xy * uint2((*(tint_symbol_2)).blockDim, 4u)) + (uint3(LocalInvocationID).xy * uint2(4u, 1u))))) - as_type<uint2>(int2(int(filterOffset), 0))));
+  int2 const baseIndex = as_type<int2>((as_type<uint2>(as_type<int2>(int2(((uint3(WorkGroupID).xy * uint2((*(tint_symbol_2)).blockDim, 4u)) + (uint3(LocalInvocationID).xy * uint2(4u, 1u)))))) - as_type<uint2>(as_type<int2>(int2(int(filterOffset), 0)))));
  for(uint r = 0u; (r < 4u); r = (r + 1u)) {
    for(uint c = 0u; (c < 4u); c = (c + 1u)) {
-      int2 loadIndex = as_type<int2>((as_type<uint2>(baseIndex) + as_type<uint2>(int2(int(c), int(r)))));
+      int2 loadIndex = as_type<int2>((as_type<uint2>(as_type<int2>(baseIndex)) + as_type<uint2>(as_type<int2>(int2(int(c), int(r))))));
      if (((*(tint_symbol_4)).value != 0u)) {
        loadIndex = int2(loadIndex).yx;
      }
@ -45,7 +45,7 @@ void tint_symbol_inner(uint3 WorkGroupID, uint3 LocalInvocationID, uint local_in
  threadgroup_barrier(mem_flags::mem_threadgroup);
  for(uint r = 0u; (r < 4u); r = (r + 1u)) {
    for(uint c = 0u; (c < 4u); c = (c + 1u)) {
-      int2 writeIndex = as_type<int2>((as_type<uint2>(baseIndex) + as_type<uint2>(int2(int(c), int(r)))));
+      int2 writeIndex = as_type<int2>((as_type<uint2>(as_type<int2>(baseIndex)) + as_type<uint2>(as_type<int2>(int2(int(c), int(r))))));
      if (((*(tint_symbol_4)).value != 0u)) {
        writeIndex = int2(writeIndex).yx;
      }
--- a/test/tint/expressions/binary/add/scalar-vec3/i32.wgsl.expected.msl
+++ b/test/tint/expressions/binary/add/scalar-vec3/i32.wgsl.expected.msl
@ -4,7 +4,7 @@ using namespace metal;
 kernel void f() {
  int const a = 4;
  int3 const b = int3(1, 2, 3);
-  int3 const r = as_type<int3>((as_type<uint>(a) + as_type<uint3>(b)));
+  int3 const r = as_type<int3>((as_type<uint>(a) + as_type<uint3>(as_type<int3>(b))));
  return;
 }

--- a/test/tint/expressions/binary/add/vec3-scalar/i32.wgsl.expected.msl
+++ b/test/tint/expressions/binary/add/vec3-scalar/i32.wgsl.expected.msl
@ -4,7 +4,7 @@ using namespace metal;
 kernel void f() {
  int3 const a = int3(1, 2, 3);
  int const b = 4;
-  int3 const r = as_type<int3>((as_type<uint3>(a) + as_type<uint>(b)));
+  int3 const r = as_type<int3>((as_type<uint3>(as_type<int3>(a)) + as_type<uint>(b)));
  return;
 }

--- a/test/tint/expressions/binary/add/vec3-vec3/i32.wgsl.expected.msl
+++ b/test/tint/expressions/binary/add/vec3-vec3/i32.wgsl.expected.msl
@ -4,7 +4,7 @@ using namespace metal;
 kernel void f() {
  int3 const a = int3(1, 2, 3);
  int3 const b = int3(4, 5, 6);
-  int3 const r = as_type<int3>((as_type<uint3>(a) + as_type<uint3>(b)));
+  int3 const r = as_type<int3>((as_type<uint3>(as_type<int3>(a)) + as_type<uint3>(as_type<int3>(b))));
  return;
 }

--- a/test/tint/expressions/binary/div_by_zero/by_expression/scalar-vec3/i32.wgsl.expected.msl
+++ b/test/tint/expressions/binary/div_by_zero/by_expression/scalar-vec3/i32.wgsl.expected.msl
@ -4,7 +4,7 @@ using namespace metal;
 kernel void f() {
  int a = 4;
  int3 b = int3(0, 2, 0);
-  int3 const r = (a / as_type<int3>((as_type<uint3>(b) + as_type<uint3>(b))));
+  int3 const r = (a / as_type<int3>((as_type<uint3>(as_type<int3>(b)) + as_type<uint3>(as_type<int3>(b)))));
  return;
 }

--- a/test/tint/expressions/binary/div_by_zero/by_expression/vec3-vec3/i32.wgsl.expected.msl
+++ b/test/tint/expressions/binary/div_by_zero/by_expression/vec3-vec3/i32.wgsl.expected.msl
@ -4,7 +4,7 @@ using namespace metal;
 kernel void f() {
  int3 a = int3(1, 2, 3);
  int3 b = int3(0, 5, 0);
-  int3 const r = (a / as_type<int3>((as_type<uint3>(b) + as_type<uint3>(b))));
+  int3 const r = (a / as_type<int3>((as_type<uint3>(as_type<int3>(b)) + as_type<uint3>(as_type<int3>(b)))));
  return;
 }

--- a/test/tint/expressions/binary/mod_by_zero/by_expression/scalar-vec3/i32.wgsl.expected.msl
+++ b/test/tint/expressions/binary/mod_by_zero/by_expression/scalar-vec3/i32.wgsl.expected.msl
@ -4,7 +4,7 @@ using namespace metal;
 kernel void f() {
  int a = 4;
  int3 b = int3(0, 2, 0);
-  int3 const r = (a % as_type<int3>((as_type<uint3>(b) + as_type<uint3>(b))));
+  int3 const r = (a % as_type<int3>((as_type<uint3>(as_type<int3>(b)) + as_type<uint3>(as_type<int3>(b)))));
  return;
 }

--- a/test/tint/expressions/binary/mod_by_zero/by_expression/vec3-vec3/i32.wgsl.expected.msl
+++ b/test/tint/expressions/binary/mod_by_zero/by_expression/vec3-vec3/i32.wgsl.expected.msl
@ -4,7 +4,7 @@ using namespace metal;
 kernel void f() {
  int3 a = int3(1, 2, 3);
  int3 b = int3(0, 5, 0);
-  int3 const r = (a % as_type<int3>((as_type<uint3>(b) + as_type<uint3>(b))));
+  int3 const r = (a % as_type<int3>((as_type<uint3>(as_type<int3>(b)) + as_type<uint3>(as_type<int3>(b)))));
  return;
 }

--- a/test/tint/expressions/binary/mul/scalar-vec3/i32.wgsl.expected.msl
+++ b/test/tint/expressions/binary/mul/scalar-vec3/i32.wgsl.expected.msl
@ -4,7 +4,7 @@ using namespace metal;
 kernel void f() {
  int const a = 4;
  int3 const b = int3(1, 2, 3);
-  int3 const r = as_type<int3>((as_type<uint>(a) * as_type<uint3>(b)));
+  int3 const r = as_type<int3>((as_type<uint>(a) * as_type<uint3>(as_type<int3>(b))));
  return;
 }

--- a/test/tint/expressions/binary/mul/vec3-scalar/i32.wgsl.expected.msl
+++ b/test/tint/expressions/binary/mul/vec3-scalar/i32.wgsl.expected.msl
@ -4,7 +4,7 @@ using namespace metal;
 kernel void f() {
  int3 const a = int3(1, 2, 3);
  int const b = 4;
-  int3 const r = as_type<int3>((as_type<uint3>(a) * as_type<uint>(b)));
+  int3 const r = as_type<int3>((as_type<uint3>(as_type<int3>(a)) * as_type<uint>(b)));
  return;
 }

--- a/test/tint/expressions/binary/mul/vec3-vec3/i32.wgsl.expected.msl
+++ b/test/tint/expressions/binary/mul/vec3-vec3/i32.wgsl.expected.msl
@ -4,7 +4,7 @@ using namespace metal;
 kernel void f() {
  int3 const a = int3(1, 2, 3);
  int3 const b = int3(4, 5, 6);
-  int3 const r = as_type<int3>((as_type<uint3>(a) * as_type<uint3>(b)));
+  int3 const r = as_type<int3>((as_type<uint3>(as_type<int3>(a)) * as_type<uint3>(as_type<int3>(b))));
  return;
 }

--- a/test/tint/expressions/binary/sub/scalar-vec3/i32.wgsl.expected.msl
+++ b/test/tint/expressions/binary/sub/scalar-vec3/i32.wgsl.expected.msl
@ -4,7 +4,7 @@ using namespace metal;
 kernel void f() {
  int const a = 4;
  int3 const b = int3(1, 2, 3);
-  int3 const r = as_type<int3>((as_type<uint>(a) - as_type<uint3>(b)));
+  int3 const r = as_type<int3>((as_type<uint>(a) - as_type<uint3>(as_type<int3>(b))));
  return;
 }

--- a/test/tint/expressions/binary/sub/vec3-scalar/i32.wgsl.expected.msl
+++ b/test/tint/expressions/binary/sub/vec3-scalar/i32.wgsl.expected.msl
@ -4,7 +4,7 @@ using namespace metal;
 kernel void f() {
  int3 const a = int3(1, 2, 3);
  int const b = 4;
-  int3 const r = as_type<int3>((as_type<uint3>(a) - as_type<uint>(b)));
+  int3 const r = as_type<int3>((as_type<uint3>(as_type<int3>(a)) - as_type<uint>(b)));
  return;
 }

--- a/test/tint/expressions/binary/sub/vec3-vec3/i32.wgsl.expected.msl
+++ b/test/tint/expressions/binary/sub/vec3-vec3/i32.wgsl.expected.msl
@ -4,7 +4,7 @@ using namespace metal;
 kernel void f() {
  int3 const a = int3(1, 2, 3);
  int3 const b = int3(4, 5, 6);
-  int3 const r = as_type<int3>((as_type<uint3>(a) - as_type<uint3>(b)));
+  int3 const r = as_type<int3>((as_type<uint3>(as_type<int3>(a)) - as_type<uint3>(as_type<int3>(b))));
  return;
 }

--- a/test/tint/statements/compound_assign/vector/minus.wgsl.expected.msl
+++ b/test/tint/statements/compound_assign/vector/minus.wgsl.expected.msl
@ -6,6 +6,6 @@ struct S {
 };

 void foo(device S* const tint_symbol) {
-  (*(tint_symbol)).a = as_type<int4>((as_type<uint4>((*(tint_symbol)).a) - as_type<uint4>(int4(2))));
+  (*(tint_symbol)).a = as_type<int4>((as_type<uint4>(as_type<int4>((*(tint_symbol)).a)) - as_type<uint4>(as_type<int4>(int4(2)))));
 }

--- a/test/tint/statements/compound_assign/vector/plus.wgsl.expected.msl
+++ b/test/tint/statements/compound_assign/vector/plus.wgsl.expected.msl
@ -6,6 +6,6 @@ struct S {
 };

 void foo(device S* const tint_symbol) {
-  (*(tint_symbol)).a = as_type<int4>((as_type<uint4>((*(tint_symbol)).a) + as_type<uint4>(int4(2))));
+  (*(tint_symbol)).a = as_type<int4>((as_type<uint4>(as_type<int4>((*(tint_symbol)).a)) + as_type<uint4>(as_type<int4>(int4(2)))));
 }

--- a/test/tint/statements/compound_assign/vector/times.wgsl.expected.msl
+++ b/test/tint/statements/compound_assign/vector/times.wgsl.expected.msl
@ -6,6 +6,6 @@ struct S {
 };

 void foo(device S* const tint_symbol) {
-  (*(tint_symbol)).a = as_type<int4>((as_type<uint4>((*(tint_symbol)).a) * as_type<uint4>(int4(2))));
+  (*(tint_symbol)).a = as_type<int4>((as_type<uint4>(as_type<int4>((*(tint_symbol)).a)) * as_type<uint4>(as_type<int4>(int4(2)))));
 }