HLSL: force FXC to never unroll loops

Emit the "[loop]" attribute on "for" and "while" so that FXC does not
attempt to unroll them. This is to work around an FXC bug where it fails
to unroll loops with gradient operations.

FXC ostensibly unrolls such loops because gradient operations require
uniform control flow, and loops that have varying iterations may
possibly not be uniform. Tint will eventually validate that control flow
is indeed uniform, so forcing FXC to avoid unrolling in these cases
should be fine.

Bug: tint:1112
Change-Id: I10077f8b62fbbb230a0003f3864c75a8fe0e1d18
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/69880
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Commit-Queue: Antonio Maiorano <amaiorano@google.com>

src/writer/hlsl/generator_impl.cc

@@ -115,6 +115,13 @@ std::ostream& operator<<(std::ostream& s, const RegisterAndSpace& rs) {
   return s;
 }
 
+const char* LoopAttribute() {
+  // Force loops not to be unrolled to work around FXC compilation issues when
+  // it attempts and fails to unroll loops when it contains gradient operations.
+  // https://docs.microsoft.com/en-us/windows/win32/direct3dhlsl/dx-graphics-hlsl-while
+  return "[loop] ";
+}
+
 }  // namespace
 
 SanitizedResult Sanitize(const Program* in,
@@ -2804,7 +2811,7 @@ bool GeneratorImpl::EmitLoop(const ast::LoopStatement* stmt) {
   };
 
   TINT_SCOPED_ASSIGNMENT(emit_continuing_, emit_continuing);
-  line() << "while (true) {";
+  line() << LoopAttribute() << "while (true) {";
   {
     ScopedIndent si(this);
     if (!EmitStatements(stmt->body->statements)) {
@@ -2874,7 +2881,7 @@ bool GeneratorImpl::EmitForLoop(const ast::ForLoopStatement* stmt) {
     };
 
     TINT_SCOPED_ASSIGNMENT(emit_continuing_, emit_continuing);
-    line() << "while (true) {";
+    line() << LoopAttribute() << "while (true) {";
     increment_indent();
     TINT_DEFER({
       decrement_indent();
@@ -2897,7 +2904,7 @@ bool GeneratorImpl::EmitForLoop(const ast::ForLoopStatement* stmt) {
     // For-loop can be generated.
     {
       auto out = line();
-      out << "for";
+      out << LoopAttribute() << "for";
       {
         ScopedParen sp(out);
 

src/writer/hlsl/generator_impl_continue_test.cc

@@ -30,7 +30,7 @@ TEST_F(HlslGeneratorImplTest_Continue, Emit_Continue) {
   gen.increment_indent();
 
   ASSERT_TRUE(gen.EmitStatement(loop)) << gen.error();
-  EXPECT_EQ(gen.result(), R"(  while (true) {
+  EXPECT_EQ(gen.result(), R"(  [loop] while (true) {
     continue;
   }
 )");

src/writer/hlsl/generator_impl_loop_test.cc

@@ -34,7 +34,7 @@ TEST_F(HlslGeneratorImplTest_Loop, Emit_Loop) {
   gen.increment_indent();
 
   ASSERT_TRUE(gen.EmitStatement(l)) << gen.error();
-  EXPECT_EQ(gen.result(), R"(  while (true) {
+  EXPECT_EQ(gen.result(), R"(  [loop] while (true) {
     discard;
   }
 )");
@@ -54,7 +54,7 @@ TEST_F(HlslGeneratorImplTest_Loop, Emit_LoopWithContinuing) {
   gen.increment_indent();
 
   ASSERT_TRUE(gen.EmitStatement(l)) << gen.error();
-  EXPECT_EQ(gen.result(), R"(  while (true) {
+  EXPECT_EQ(gen.result(), R"(  [loop] while (true) {
     discard;
     {
       a_statement();
@@ -88,8 +88,8 @@ TEST_F(HlslGeneratorImplTest_Loop, Emit_LoopNestedWithContinuing) {
   gen.increment_indent();
 
   ASSERT_TRUE(gen.EmitStatement(outer)) << gen.error();
-  EXPECT_EQ(gen.result(), R"(  while (true) {
+  EXPECT_EQ(gen.result(), R"(  [loop] while (true) {
-    while (true) {
+    [loop] while (true) {
       discard;
       {
         a_statement();
@@ -142,7 +142,7 @@ TEST_F(HlslGeneratorImplTest_Loop, Emit_LoopWithVarUsedInContinuing) {
   gen.increment_indent();
 
   ASSERT_TRUE(gen.EmitStatement(outer)) << gen.error();
-  EXPECT_EQ(gen.result(), R"(  while (true) {
+  EXPECT_EQ(gen.result(), R"(  [loop] while (true) {
     float lhs = 2.400000095f;
     float other = 0.0f;
     {
@@ -169,7 +169,7 @@ TEST_F(HlslGeneratorImplTest_Loop, Emit_ForLoop) {
 
   ASSERT_TRUE(gen.EmitStatement(f)) << gen.error();
   EXPECT_EQ(gen.result(), R"(  {
-    for(; ; ) {
+    [loop] for(; ; ) {
       a_statement();
     }
   }
@@ -193,7 +193,7 @@ TEST_F(HlslGeneratorImplTest_Loop, Emit_ForLoopWithSimpleInit) {
 
   ASSERT_TRUE(gen.EmitStatement(f)) << gen.error();
   EXPECT_EQ(gen.result(), R"(  {
-    for(int i = 0; ; ) {
+    [loop] for(int i = 0; ; ) {
       a_statement();
     }
   }
@@ -223,7 +223,7 @@ TEST_F(HlslGeneratorImplTest_Loop, Emit_ForLoopWithMultiStmtInit) {
       tint_tmp = false;
     }
     bool b = (tint_tmp);
-    for(; ; ) {
+    [loop] for(; ; ) {
       a_statement();
     }
   }
@@ -246,7 +246,7 @@ TEST_F(HlslGeneratorImplTest_Loop, Emit_ForLoopWithSimpleCond) {
 
   ASSERT_TRUE(gen.EmitStatement(f)) << gen.error();
   EXPECT_EQ(gen.result(), R"(  {
-    for(; true; ) {
+    [loop] for(; true; ) {
       a_statement();
     }
   }
@@ -272,7 +272,7 @@ TEST_F(HlslGeneratorImplTest_Loop, Emit_ForLoopWithMultiStmtCond) {
 
   ASSERT_TRUE(gen.EmitStatement(f)) << gen.error();
   EXPECT_EQ(gen.result(), R"(  {
-    while (true) {
+    [loop] while (true) {
       bool tint_tmp = true;
       if (tint_tmp) {
         tint_tmp = false;
@@ -302,7 +302,7 @@ TEST_F(HlslGeneratorImplTest_Loop, Emit_ForLoopWithSimpleCont) {
 
   ASSERT_TRUE(gen.EmitStatement(f)) << gen.error();
   EXPECT_EQ(gen.result(), R"(  {
-    for(; ; i = (i + 1)) {
+    [loop] for(; ; i = (i + 1)) {
       a_statement();
     }
   }
@@ -329,7 +329,7 @@ TEST_F(HlslGeneratorImplTest_Loop, Emit_ForLoopWithMultiStmtCont) {
 
   ASSERT_TRUE(gen.EmitStatement(f)) << gen.error();
   EXPECT_EQ(gen.result(), R"(  {
-    while (true) {
+    [loop] while (true) {
       a_statement();
       bool tint_tmp = true;
       if (tint_tmp) {
@@ -358,7 +358,7 @@ TEST_F(HlslGeneratorImplTest_Loop, Emit_ForLoopWithSimpleInitCondCont) {
 
   ASSERT_TRUE(gen.EmitStatement(f)) << gen.error();
   EXPECT_EQ(gen.result(), R"(  {
-    for(int i = 0; true; i = (i + 1)) {
+    [loop] for(int i = 0; true; i = (i + 1)) {
       a_statement();
     }
   }
@@ -394,7 +394,7 @@ TEST_F(HlslGeneratorImplTest_Loop, Emit_ForLoopWithMultiStmtInitCondCont) {
       tint_tmp = false;
     }
     bool i = (tint_tmp);
-    while (true) {
+    [loop] while (true) {
       bool tint_tmp_1 = true;
       if (tint_tmp_1) {
         tint_tmp_1 = false;

test/array/assign_to_function_var.wgsl.expected.hlsl

@@ -32,7 +32,7 @@ typedef tint_padded_array_element tint_symbol_1_ret[4];
 tint_symbol_1_ret tint_symbol_1(uint4 buffer[4], uint offset) {
   tint_padded_array_element arr_1[4] = (tint_padded_array_element[4])0;
   {
-    for(uint i = 0u; (i < 4u); i = (i + 1u)) {
+    [loop] for(uint i = 0u; (i < 4u); i = (i + 1u)) {
       const uint scalar_offset = ((offset + (i * 16u))) / 4;
       arr_1[i].el = asint(buffer[scalar_offset / 4][scalar_offset % 4]);
     }
@@ -44,7 +44,7 @@ typedef tint_padded_array_element tint_symbol_3_ret[4];
 tint_symbol_3_ret tint_symbol_3(RWByteAddressBuffer buffer, uint offset) {
   tint_padded_array_element arr_2[4] = (tint_padded_array_element[4])0;
   {
-    for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
+    [loop] for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
       arr_2[i_1].el = asint(buffer.Load((offset + (i_1 * 16u))));
     }
   }

test/array/assign_to_private_var.wgsl.expected.hlsl

@@ -34,7 +34,7 @@ typedef tint_padded_array_element tint_symbol_1_ret[4];
 tint_symbol_1_ret tint_symbol_1(uint4 buffer[4], uint offset) {
   tint_padded_array_element arr_1[4] = (tint_padded_array_element[4])0;
   {
-    for(uint i = 0u; (i < 4u); i = (i + 1u)) {
+    [loop] for(uint i = 0u; (i < 4u); i = (i + 1u)) {
       const uint scalar_offset = ((offset + (i * 16u))) / 4;
       arr_1[i].el = asint(buffer[scalar_offset / 4][scalar_offset % 4]);
     }
@@ -46,7 +46,7 @@ typedef tint_padded_array_element tint_symbol_3_ret[4];
 tint_symbol_3_ret tint_symbol_3(RWByteAddressBuffer buffer, uint offset) {
   tint_padded_array_element arr_2[4] = (tint_padded_array_element[4])0;
   {
-    for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
+    [loop] for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
       arr_2[i_1].el = asint(buffer.Load((offset + (i_1 * 16u))));
     }
   }

test/array/assign_to_storage_var.wgsl.expected.hlsl

@@ -33,7 +33,7 @@ S ret_struct_arr() {
 void tint_symbol_1(RWByteAddressBuffer buffer, uint offset, tint_padded_array_element value[4]) {
   tint_padded_array_element array[4] = value;
   {
-    for(uint i = 0u; (i < 4u); i = (i + 1u)) {
+    [loop] for(uint i = 0u; (i < 4u); i = (i + 1u)) {
       buffer.Store((offset + (i * 16u)), asuint(array[i].el));
     }
   }
@@ -43,7 +43,7 @@ typedef tint_padded_array_element tint_symbol_3_ret[4];
 tint_symbol_3_ret tint_symbol_3(uint4 buffer[4], uint offset) {
   tint_padded_array_element arr_1[4] = (tint_padded_array_element[4])0;
   {
-    for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
+    [loop] for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
       const uint scalar_offset = ((offset + (i_1 * 16u))) / 4;
       arr_1[i_1].el = asint(buffer[scalar_offset / 4][scalar_offset % 4]);
     }
@@ -55,7 +55,7 @@ typedef tint_padded_array_element tint_symbol_5_ret[4];
 tint_symbol_5_ret tint_symbol_5(RWByteAddressBuffer buffer, uint offset) {
   tint_padded_array_element arr_2[4] = (tint_padded_array_element[4])0;
   {
-    for(uint i_2 = 0u; (i_2 < 4u); i_2 = (i_2 + 1u)) {
+    [loop] for(uint i_2 = 0u; (i_2 < 4u); i_2 = (i_2 + 1u)) {
       arr_2[i_2].el = asint(buffer.Load((offset + (i_2 * 16u))));
     }
   }
@@ -65,7 +65,7 @@ tint_symbol_5_ret tint_symbol_5(RWByteAddressBuffer buffer, uint offset) {
 void tint_symbol_9(RWByteAddressBuffer buffer, uint offset, int value[2]) {
   int array_3[2] = value;
   {
-    for(uint i_3 = 0u; (i_3 < 2u); i_3 = (i_3 + 1u)) {
+    [loop] for(uint i_3 = 0u; (i_3 < 2u); i_3 = (i_3 + 1u)) {
       buffer.Store((offset + (i_3 * 4u)), asuint(array_3[i_3]));
     }
   }
@@ -74,7 +74,7 @@ void tint_symbol_9(RWByteAddressBuffer buffer, uint offset, int value[2]) {
 void tint_symbol_8(RWByteAddressBuffer buffer, uint offset, int value[3][2]) {
   int array_2[3][2] = value;
   {
-    for(uint i_4 = 0u; (i_4 < 3u); i_4 = (i_4 + 1u)) {
+    [loop] for(uint i_4 = 0u; (i_4 < 3u); i_4 = (i_4 + 1u)) {
       tint_symbol_9(buffer, (offset + (i_4 * 8u)), array_2[i_4]);
     }
   }
@@ -83,7 +83,7 @@ void tint_symbol_8(RWByteAddressBuffer buffer, uint offset, int value[3][2]) {
 void tint_symbol_7(RWByteAddressBuffer buffer, uint offset, int value[4][3][2]) {
   int array_1[4][3][2] = value;
   {
-    for(uint i_5 = 0u; (i_5 < 4u); i_5 = (i_5 + 1u)) {
+    [loop] for(uint i_5 = 0u; (i_5 < 4u); i_5 = (i_5 + 1u)) {
       tint_symbol_8(buffer, (offset + (i_5 * 24u)), array_1[i_5]);
     }
   }

test/array/assign_to_workgroup_var.wgsl.expected.hlsl

@@ -34,7 +34,7 @@ typedef tint_padded_array_element tint_symbol_1_ret[4];
 tint_symbol_1_ret tint_symbol_1(uint4 buffer[4], uint offset) {
   tint_padded_array_element arr_1[4] = (tint_padded_array_element[4])0;
   {
-    for(uint i = 0u; (i < 4u); i = (i + 1u)) {
+    [loop] for(uint i = 0u; (i < 4u); i = (i + 1u)) {
       const uint scalar_offset = ((offset + (i * 16u))) / 4;
       arr_1[i].el = asint(buffer[scalar_offset / 4][scalar_offset % 4]);
     }
@@ -46,7 +46,7 @@ typedef tint_padded_array_element tint_symbol_3_ret[4];
 tint_symbol_3_ret tint_symbol_3(RWByteAddressBuffer buffer, uint offset) {
   tint_padded_array_element arr_2[4] = (tint_padded_array_element[4])0;
   {
-    for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
+    [loop] for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
       arr_2[i_1].el = asint(buffer.Load((offset + (i_1 * 16u))));
     }
   }

test/buffer/storage/dynamic_index/read.wgsl.expected.hlsl

@@ -16,7 +16,7 @@ typedef int4 tint_symbol_11_ret[4];
 tint_symbol_11_ret tint_symbol_11(ByteAddressBuffer buffer, uint offset) {
   int4 arr_1[4] = (int4[4])0;
   {
-    for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
+    [loop] for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
       arr_1[i_1] = asint(buffer.Load4((offset + (i_1 * 16u))));
     }
   }

test/buffer/storage/dynamic_index/write.wgsl.expected.hlsl

@@ -18,7 +18,7 @@ void tint_symbol_9(RWByteAddressBuffer buffer, uint offset, float3x2 value) {
 void tint_symbol_11(RWByteAddressBuffer buffer, uint offset, int4 value[4]) {
   int4 array[4] = value;
   {
-    for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
+    [loop] for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
       buffer.Store4((offset + (i_1 * 16u)), asuint(array[i_1]));
     }
   }

test/buffer/storage/static_index/read.wgsl.expected.hlsl

@@ -24,7 +24,7 @@ typedef tint_padded_array_element tint_symbol_10_ret[4];
 tint_symbol_10_ret tint_symbol_10(ByteAddressBuffer buffer, uint offset) {
   tint_padded_array_element arr[4] = (tint_padded_array_element[4])0;
   {
-    for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
+    [loop] for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
       arr[i_1].el = tint_symbol_9(buffer, (offset + (i_1 * 16u)));
     }
   }

test/buffer/storage/static_index/write.wgsl.expected.hlsl

@@ -25,7 +25,7 @@ void tint_symbol_9(RWByteAddressBuffer buffer, uint offset, Inner value) {
 void tint_symbol_10(RWByteAddressBuffer buffer, uint offset, tint_padded_array_element value[4]) {
   tint_padded_array_element array[4] = value;
   {
-    for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
+    [loop] for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
       tint_symbol_9(buffer, (offset + (i_1 * 16u)), array[i_1].el);
     }
   }

test/buffer/uniform/dynamic_index/read.wgsl.expected.hlsl

@@ -26,7 +26,7 @@ typedef int4 tint_symbol_12_ret[4];
 tint_symbol_12_ret tint_symbol_12(uint4 buffer[96], uint offset) {
   int4 arr_1[4] = (int4[4])0;
   {
-    for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
+    [loop] for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
       const uint scalar_offset_5 = ((offset + (i_1 * 16u))) / 4;
       arr_1[i_1] = asint(buffer[scalar_offset_5 / 4]);
     }

test/buffer/uniform/static_index/read.wgsl.expected.hlsl

@@ -35,7 +35,7 @@ typedef tint_padded_array_element tint_symbol_11_ret[4];
 tint_symbol_11_ret tint_symbol_11(uint4 buffer[13], uint offset) {
   tint_padded_array_element arr[4] = (tint_padded_array_element[4])0;
   {
-    for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
+    [loop] for(uint i_1 = 0u; (i_1 < 4u); i_1 = (i_1 + 1u)) {
       arr[i_1].el = tint_symbol_10(buffer, (offset + (i_1 * 16u)));
     }
   }

test/bug/fxc/dyn_array_idx/read/workgroup.wgsl.expected.hlsl

@@ -15,7 +15,7 @@ struct tint_symbol_2 {
 
 void f_inner(uint local_invocation_index) {
   {
-    for(uint idx = local_invocation_index; (idx < 64u); idx = (idx + 1u)) {
+    [loop] for(uint idx = local_invocation_index; (idx < 64u); idx = (idx + 1u)) {
       const uint i = idx;
       s.data[i] = 0;
     }

test/bug/fxc/dyn_array_idx/write/workgroup.wgsl.expected.hlsl

@@ -15,7 +15,7 @@ struct tint_symbol_2 {
 
 void f_inner(uint local_invocation_index) {
   {
-    for(uint idx = local_invocation_index; (idx < 64u); idx = (idx + 1u)) {
+    [loop] for(uint idx = local_invocation_index; (idx < 64u); idx = (idx + 1u)) {
       const uint i = idx;
       s.data[i] = 0;
     }

test/bug/fxc/vector_assignment_in_loop/loop_call_with_loop.wgsl.expected.hlsl

@@ -21,7 +21,7 @@ static bool2 v2b = bool2(false, false);
 
 void foo() {
   {
-    for(int i = 0; (i < 2); i = (i + 1)) {
+    [loop] for(int i = 0; (i < 2); i = (i + 1)) {
       set_float2(v2f, i, 1.0f);
       set_int3(v3i, i, 1);
       set_uint4(v4u, i, 1u);
@@ -33,7 +33,7 @@ void foo() {
 [numthreads(1, 1, 1)]
 void main() {
   {
-    for(int i = 0; (i < 2); i = (i + 1)) {
+    [loop] for(int i = 0; (i < 2); i = (i + 1)) {
       foo();
     }
   }

test/bug/fxc/vector_assignment_in_loop/loop_call_with_no_loop.wgsl.expected.hlsl

@@ -30,7 +30,7 @@ void foo() {
 [numthreads(1, 1, 1)]
 void main() {
   {
-    for(int i = 0; (i < 2); i = (i + 1)) {
+    [loop] for(int i = 0; (i < 2); i = (i + 1)) {
       foo();
     }
   }

test/bug/fxc/vector_assignment_in_loop/loop_types_all.wgsl.expected.hlsl

@@ -61,7 +61,7 @@ void main() {
   bool3 v3b = bool3(false, false, false);
   bool4 v4b = bool4(false, false, false, false);
   {
-    for(int i = 0; (i < 2); i = (i + 1)) {
+    [loop] for(int i = 0; (i < 2); i = (i + 1)) {
       set_float2(v2f, i, 1.0f);
       set_float3(v3f, i, 1.0f);
       set_float4(v4f, i, 1.0f);

test/bug/fxc/vector_assignment_in_loop/loop_types_repeated.wgsl.expected.hlsl

@@ -25,7 +25,7 @@ void main() {
   bool2 v2b = bool2(false, false);
   bool2 v2b_2 = bool2(false, false);
   {
-    for(int i = 0; (i < 2); i = (i + 1)) {
+    [loop] for(int i = 0; (i < 2); i = (i + 1)) {
       set_float2(v2f, i, 1.0f);
       set_int3(v3i, i, 1);
       set_uint4(v4u, i, 1u);

test/bug/fxc/vector_assignment_in_loop/loop_types_some.wgsl.expected.hlsl

@@ -61,7 +61,7 @@ void main() {
   bool3 v3b = bool3(false, false, false);
   bool4 v4b = bool4(false, false, false, false);
   {
-    for(int i = 0; (i < 2); i = (i + 1)) {
+    [loop] for(int i = 0; (i < 2); i = (i + 1)) {
       set_float2(v2f, i, 1.0f);
       set_int2(v2i, i, 1);
       set_uint2(v2u, i, 1u);

test/bug/tint/1064.wgsl.expected.hlsl

@@ -1,5 +1,5 @@
 void main() {
-  while (true) {
+  [loop] while (true) {
     if (false) {
     } else {
       break;

test/bug/tint/1081.wgsl.expected.hlsl

@@ -22,7 +22,7 @@ struct tint_symbol_2 {
 
 int main_inner(int3 x) {
   int y = x.x;
-  while (true) {
+  [loop] while (true) {
     const int r = f(y);
     if ((r == 0)) {
       break;

test/bug/tint/1121.wgsl.expected.hlsl

@@ -52,9 +52,9 @@ void main_inner(uint3 GlobalInvocationID) {
   const int TILE_SIZE = 16;
   const int TILE_COUNT_X = 2;
   {
-    for(int y_1 = 0; (y_1 < 2); y_1 = (y_1 + 1)) {
+    [loop] for(int y_1 = 0; (y_1 < 2); y_1 = (y_1 + 1)) {
       {
-        for(int x_1 = 0; (x_1 < TILE_COUNT_X); x_1 = (x_1 + 1)) {
+        [loop] for(int x_1 = 0; (x_1 < TILE_COUNT_X); x_1 = (x_1 + 1)) {
           int2 tilePixel0Idx = int2((x_1 * TILE_SIZE), (y_1 * TILE_SIZE));
           float2 floorCoord = (((2.0f * float2(tilePixel0Idx)) / asfloat(uniforms[10]).xy) - float2((1.0f).xx));
           float2 ceilCoord = (((2.0f * float2((tilePixel0Idx + int2((TILE_SIZE).xx)))) / asfloat(uniforms[10]).xy) - float2((1.0f).xx));
@@ -66,7 +66,7 @@ void main_inner(uint3 GlobalInvocationID) {
           frustumPlanes[3] = float4(0.0f, -1.0f, (viewCeilCoord.y / viewNear), 0.0f);
           float dp = 0.0f;
           {
-            for(uint i = 0u; (i < 6u); i = (i + 1u)) {
+            [loop] for(uint i = 0u; (i < 6u); i = (i + 1u)) {
               float4 p = float4(0.0f, 0.0f, 0.0f, 0.0f);
               if ((frustumPlanes[i].x > 0.0f)) {
                 p.x = boxMax.x;

test/bug/tint/221.wgsl.expected.hlsl

@@ -3,7 +3,7 @@ RWByteAddressBuffer b : register(u0, space0);
 [numthreads(1, 1, 1)]
 void main() {
   uint i = 0u;
-  while (true) {
+  [loop] while (true) {
     if ((i >= b.Load(0u))) {
       break;
     }

test/bug/tint/534.wgsl.expected.hlsl

@@ -32,7 +32,7 @@ void main_inner(uint3 GlobalInvocationID) {
   uint4 srcColorBits = uint4(0u, 0u, 0u, 0u);
   uint4 dstColorBits = uint4(dstColor);
   {
-    for(uint i = 0u; (i < uniforms[0].w); i = (i + 1u)) {
+    [loop] for(uint i = 0u; (i < uniforms[0].w); i = (i + 1u)) {
       set_uint4(srcColorBits, i, ConvertToFp16FloatValue(srcColor[i]));
       bool tint_tmp_1 = success;
       if (tint_tmp_1) {

test/bug/tint/744.wgsl.expected.hlsl

@@ -15,7 +15,7 @@ void main_inner(uint3 global_id) {
   const uint dimOutter = uniforms[1].y;
   uint result = 0u;
   {
-    for(uint i = 0u; (i < dimInner); i = (i + 1u)) {
+    [loop] for(uint i = 0u; (i < dimInner); i = (i + 1u)) {
       const uint a = (i + (resultCell.x * dimInner));
       const uint b = (resultCell.y + (i * dimOutter));
       result = (result + (firstMatrix.Load((4u * a)) * secondMatrix.Load((4u * b))));

test/bug/tint/749.spvasm.expected.hlsl

@@ -153,7 +153,7 @@ int performPartition_i1_i1_(inout int l, inout int h) {
   const QuicksortObject tint_symbol_11 = {tint_symbol_10};
   obj = tint_symbol_11;
   obj = x_960;
-  while (true) {
+  [loop] while (true) {
     const int x_961 = pivot;
     pivot = 0;
     pivot = x_961;
@@ -453,7 +453,7 @@ void quicksort_() {
   p = 0;
   p = x_1027;
   stack[x_100_save] = x_99;
-  while (true) {
+  [loop] while (true) {
     const float3 x_566 = float3(x_563.x, x_563.x, x_563.x);
     const int x_1028 = h_1;
     h_1 = 0;

test/bug/tint/757.wgsl.expected.hlsl

@@ -14,7 +14,7 @@ void main_inner(uint3 GlobalInvocationID) {
   flatIndex = (flatIndex * 1u);
   float4 texel = myTexture.Load(int4(int3(int2(GlobalInvocationID.xy), 0), 0));
   {
-    for(uint i = 0u; (i < 1u); i = (i + 1u)) {
+    [loop] for(uint i = 0u; (i < 1u); i = (i + 1u)) {
       result.Store((4u * (flatIndex + i)), asuint(texel.r));
     }
   }

test/bug/tint/870.spvasm.expected.hlsl

@@ -4,7 +4,7 @@ typedef int tint_symbol_ret[6];
 tint_symbol_ret tint_symbol(ByteAddressBuffer buffer, uint offset) {
   int arr[6] = (int[6])0;
   {
-    for(uint i = 0u; (i < 6u); i = (i + 1u)) {
+    [loop] for(uint i = 0u; (i < 6u); i = (i + 1u)) {
       arr[i] = asint(buffer.Load((offset + (i * 4u))));
     }
   }

test/bug/tint/914.wgsl.expected.hlsl

@@ -56,7 +56,7 @@ struct tint_symbol_1 {
 
 void main_inner(uint3 local_id, uint3 global_id, uint local_invocation_index) {
   {
-    for(uint idx = local_invocation_index; (idx < 4096u); idx = (idx + 256u)) {
+    [loop] for(uint idx = local_invocation_index; (idx < 4096u); idx = (idx + 256u)) {
       const uint i = (idx / 64u);
       const uint i_1 = (idx % 64u);
       mm_Asub[i][i_1] = 0.0f;
@@ -73,7 +73,7 @@ void main_inner(uint3 local_id, uint3 global_id, uint local_invocation_index) {
   float ACached = 0.0f;
   float BCached[4] = (float[4])0;
   {
-    for(uint index = 0u; (index < (RowPerThread * ColPerThread)); index = (index + 1u)) {
+    [loop] for(uint index = 0u; (index < (RowPerThread * ColPerThread)); index = (index + 1u)) {
       acc[index] = 0.0f;
     }
   }
@@ -82,11 +82,11 @@ void main_inner(uint3 local_id, uint3 global_id, uint local_invocation_index) {
   const uint RowPerThreadB = (TileInner / 16u);
   const uint tileRowB = (local_id.y * RowPerThreadB);
   {
-    for(uint t = 0u; (t < numTiles); t = (t + 1u)) {
+    [loop] for(uint t = 0u; (t < numTiles); t = (t + 1u)) {
       {
-        for(uint innerRow = 0u; (innerRow < RowPerThread); innerRow = (innerRow + 1u)) {
+        [loop] for(uint innerRow = 0u; (innerRow < RowPerThread); innerRow = (innerRow + 1u)) {
           {
-            for(uint innerCol = 0u; (innerCol < ColPerThreadA); innerCol = (innerCol + 1u)) {
+            [loop] for(uint innerCol = 0u; (innerCol < ColPerThreadA); innerCol = (innerCol + 1u)) {
               const uint inputRow = (tileRow + innerRow);
               const uint inputCol = (tileColA + innerCol);
               mm_Asub[inputRow][inputCol] = mm_readA((globalRow + innerRow), ((t * TileInner) + inputCol));
@@ -95,9 +95,9 @@ void main_inner(uint3 local_id, uint3 global_id, uint local_invocation_index) {
         }
       }
       {
-        for(uint innerRow = 0u; (innerRow < RowPerThreadB); innerRow = (innerRow + 1u)) {
+        [loop] for(uint innerRow = 0u; (innerRow < RowPerThreadB); innerRow = (innerRow + 1u)) {
           {
-            for(uint innerCol = 0u; (innerCol < ColPerThread); innerCol = (innerCol + 1u)) {
+            [loop] for(uint innerCol = 0u; (innerCol < ColPerThread); innerCol = (innerCol + 1u)) {
               const uint inputRow = (tileRowB + innerRow);
               const uint inputCol = (tileCol + innerCol);
               mm_Bsub[innerCol][inputCol] = mm_readB(((t * TileInner) + inputRow), (globalCol + innerCol));
@@ -107,17 +107,17 @@ void main_inner(uint3 local_id, uint3 global_id, uint local_invocation_index) {
       }
       GroupMemoryBarrierWithGroupSync();
       {
-        for(uint k = 0u; (k < TileInner); k = (k + 1u)) {
+        [loop] for(uint k = 0u; (k < TileInner); k = (k + 1u)) {
           {
-            for(uint inner = 0u; (inner < ColPerThread); inner = (inner + 1u)) {
+            [loop] for(uint inner = 0u; (inner < ColPerThread); inner = (inner + 1u)) {
               BCached[inner] = mm_Bsub[k][(tileCol + inner)];
             }
           }
           {
-            for(uint innerRow = 0u; (innerRow < RowPerThread); innerRow = (innerRow + 1u)) {
+            [loop] for(uint innerRow = 0u; (innerRow < RowPerThread); innerRow = (innerRow + 1u)) {
               ACached = mm_Asub[(tileRow + innerRow)][k];
               {
-                for(uint innerCol = 0u; (innerCol < ColPerThread); innerCol = (innerCol + 1u)) {
+                [loop] for(uint innerCol = 0u; (innerCol < ColPerThread); innerCol = (innerCol + 1u)) {
                   const uint index = ((innerRow * ColPerThread) + innerCol);
                   acc[index] = (acc[index] + (ACached * BCached[innerCol]));
                 }
@@ -130,9 +130,9 @@ void main_inner(uint3 local_id, uint3 global_id, uint local_invocation_index) {
     }
   }
   {
-    for(uint innerRow = 0u; (innerRow < RowPerThread); innerRow = (innerRow + 1u)) {
+    [loop] for(uint innerRow = 0u; (innerRow < RowPerThread); innerRow = (innerRow + 1u)) {
       {
-        for(uint innerCol = 0u; (innerCol < ColPerThread); innerCol = (innerCol + 1u)) {
+        [loop] for(uint innerCol = 0u; (innerCol < ColPerThread); innerCol = (innerCol + 1u)) {
           const uint index = ((innerRow * ColPerThread) + innerCol);
           mm_write((globalRow + innerRow), (globalCol + innerCol), acc[index]);
         }

test/bug/tint/942.wgsl.expected.hlsl

@@ -18,7 +18,7 @@ struct tint_symbol_1 {
 
 void main_inner(uint3 WorkGroupID, uint3 LocalInvocationID, uint local_invocation_index) {
   {
-    for(uint idx = local_invocation_index; (idx < 1024u); idx = (idx + 64u)) {
+    [loop] for(uint idx = local_invocation_index; (idx < 1024u); idx = (idx + 64u)) {
       const uint i_1 = (idx / 256u);
       const uint i_2 = (idx % 256u);
       tile[i_1][i_2] = float3(0.0f, 0.0f, 0.0f);
@@ -31,9 +31,9 @@ void main_inner(uint3 WorkGroupID, uint3 LocalInvocationID, uint local_invocatio
   const int2 dims = tint_tmp.xy;
   const int2 baseIndex = (int2(((WorkGroupID.xy * uint2(params[0].y, 4u)) + (LocalInvocationID.xy * uint2(4u, 1u)))) - int2(int(filterOffset), 0));
   {
-    for(uint r = 0u; (r < 4u); r = (r + 1u)) {
+    [loop] for(uint r = 0u; (r < 4u); r = (r + 1u)) {
       {
-        for(uint c = 0u; (c < 4u); c = (c + 1u)) {
+        [loop] for(uint c = 0u; (c < 4u); c = (c + 1u)) {
           int2 loadIndex = (baseIndex + int2(int(c), int(r)));
           if ((flip[0].x != 0u)) {
             loadIndex = loadIndex.yx;
@@ -45,9 +45,9 @@ void main_inner(uint3 WorkGroupID, uint3 LocalInvocationID, uint local_invocatio
   }
   GroupMemoryBarrierWithGroupSync();
   {
-    for(uint r = 0u; (r < 4u); r = (r + 1u)) {
+    [loop] for(uint r = 0u; (r < 4u); r = (r + 1u)) {
       {
-        for(uint c = 0u; (c < 4u); c = (c + 1u)) {
+        [loop] for(uint c = 0u; (c < 4u); c = (c + 1u)) {
           int2 writeIndex = (baseIndex + int2(int(c), int(r)));
           if ((flip[0].x != 0u)) {
             writeIndex = writeIndex.yx;
@@ -64,7 +64,7 @@ void main_inner(uint3 WorkGroupID, uint3 LocalInvocationID, uint local_invocatio
           if ((tint_tmp_1)) {
             float3 acc = float3(0.0f, 0.0f, 0.0f);
             {
-              for(uint f = 0u; (f < params[0].x); f = (f + 1u)) {
+              [loop] for(uint f = 0u; (f < params[0].x); f = (f + 1u)) {
                 uint i = ((center + f) - filterOffset);
                 acc = (acc + ((1.0f / float(params[0].x)) * tile[r][i]));
               }

test/bug/tint/943.spvasm.expected.hlsl

@@ -186,10 +186,10 @@ void mm_matMul_i1_i1_i1_(inout int dimAOuter, inout int dimInner, inout int dimB
   numTiles = (((x_152 - 1) / 64) + 1);
   innerRow = 0;
   {
-    for(; (innerRow < 1); innerRow = (innerRow + 1)) {
+    [loop] for(; (innerRow < 1); innerRow = (innerRow + 1)) {
       innerCol = 0;
       {
-        for(; (innerCol < 1); innerCol = (innerCol + 1)) {
+        [loop] for(; (innerCol < 1); innerCol = (innerCol + 1)) {
           acc[innerRow][innerCol] = 0.0f;
         }
       }
@@ -201,13 +201,13 @@ void mm_matMul_i1_i1_i1_(inout int dimAOuter, inout int dimInner, inout int dimB
   tileRowB = (asint(x_192) * 1);
   t = 0;
   {
-    for(; (t < numTiles); t = (t + 1)) {
+    [loop] for(; (t < numTiles); t = (t + 1)) {
       innerRow_1 = 0;
       {
-        for(; (innerRow_1 < 1); innerRow_1 = (innerRow_1 + 1)) {
+        [loop] for(; (innerRow_1 < 1); innerRow_1 = (innerRow_1 + 1)) {
           innerCol_1 = 0;
           {
-            for(; (innerCol_1 < 64); innerCol_1 = (innerCol_1 + 1)) {
+            [loop] for(; (innerCol_1 < 64); innerCol_1 = (innerCol_1 + 1)) {
               inputRow = (tileRow + innerRow_1);
               inputCol = (tileColA + innerCol_1);
               const int x_233 = inputRow;
@@ -224,10 +224,10 @@ void mm_matMul_i1_i1_i1_(inout int dimAOuter, inout int dimInner, inout int dimB
       }
       innerRow_2 = 0;
       {
-        for(; (innerRow_2 < 1); innerRow_2 = (innerRow_2 + 1)) {
+        [loop] for(; (innerRow_2 < 1); innerRow_2 = (innerRow_2 + 1)) {
           innerCol_2 = 0;
           {
-            for(; (innerCol_2 < 1); innerCol_2 = (innerCol_2 + 1)) {
+            [loop] for(; (innerCol_2 < 1); innerCol_2 = (innerCol_2 + 1)) {
               inputRow_1 = (tileRowB + innerRow_2);
               inputCol_1 = (tileCol + innerCol_2);
               const int x_278 = inputRow_1;
@@ -245,10 +245,10 @@ void mm_matMul_i1_i1_i1_(inout int dimAOuter, inout int dimInner, inout int dimB
       GroupMemoryBarrierWithGroupSync();
       k = 0;
       {
-        for(; (k < 64); k = (k + 1)) {
+        [loop] for(; (k < 64); k = (k + 1)) {
           inner = 0;
           {
-            for(; (inner < 1); inner = (inner + 1)) {
+            [loop] for(; (inner < 1); inner = (inner + 1)) {
               const int x_314 = inner;
               const float x_320 = mm_Bsub[k][(tileCol + inner)];
               BCached[x_314] = x_320;
@@ -256,12 +256,12 @@ void mm_matMul_i1_i1_i1_(inout int dimAOuter, inout int dimInner, inout int dimB
           }
           innerRow_3 = 0;
           {
-            for(; (innerRow_3 < 1); innerRow_3 = (innerRow_3 + 1)) {
+            [loop] for(; (innerRow_3 < 1); innerRow_3 = (innerRow_3 + 1)) {
               const float x_338 = mm_Asub[(tileRow + innerRow_3)][k];
               ACached = x_338;
               innerCol_3 = 0;
               {
-                for(; (innerCol_3 < 1); innerCol_3 = (innerCol_3 + 1)) {
+                [loop] for(; (innerCol_3 < 1); innerCol_3 = (innerCol_3 + 1)) {
                   const int x_347 = innerRow_3;
                   const int x_348 = innerCol_3;
                   const float x_349 = ACached;
@@ -279,9 +279,9 @@ void mm_matMul_i1_i1_i1_(inout int dimAOuter, inout int dimInner, inout int dimB
   }
   innerRow_4 = 0;
   {
-    for(; (innerRow_4 < 1); innerRow_4 = (innerRow_4 + 1)) {
+    [loop] for(; (innerRow_4 < 1); innerRow_4 = (innerRow_4 + 1)) {
       innerCol_4 = 0;
-      while (true) {
+      [loop] while (true) {
         bool x_393 = false;
         bool x_394_phi = false;
         if ((innerCol_4 < 1)) {
@@ -352,7 +352,7 @@ void main_inner(uint3 gl_LocalInvocationID_param, uint3 gl_GlobalInvocationID_pa
     mm_Bsub[i_1][i_2] = 0.0f;
   }
   {
-    for(uint idx = local_invocation_index; (idx < 4096u); idx = (idx + 64u)) {
+    [loop] for(uint idx = local_invocation_index; (idx < 4096u); idx = (idx + 64u)) {
       const uint i = (idx / 64u);
       const uint i_1 = (idx % 64u);
       mm_Asub[i][i_1] = 0.0f;

test/bug/tint/948.wgsl.expected.hlsl

@@ -62,7 +62,7 @@ void main_1() {
   stageUnits = (float2(1.0f, 1.0f) / x_111);
   i = 0;
   {
-    for(; (i < 2); i = (i + 1)) {
+    [loop] for(; (i < 2); i = (i + 1)) {
       switch(i) {
         case 1: {
           const float2 x_150 = tileID;
@@ -93,7 +93,7 @@ void main_1() {
         mt = ((x_181 * x_184) % 1.0f);
         f = 0.0f;
         {
-          for(; (f < 8.0f); f = (f + 1.0f)) {
+          [loop] for(; (f < 8.0f); f = (f + 1.0f)) {
             const float x_197 = animationData.y;
             if ((x_197 > mt)) {
               const float x_203 = animationData.x;

test/bug/tint/949.wgsl.expected.hlsl

@@ -244,7 +244,7 @@ void main_1() {
   currSampledHeight = 1.0f;
   i = 0;
   {
-    for(; (i < 15); i = (i + 1)) {
+    [loop] for(; (i < 15); i = (i + 1)) {
       const float4 x_397 = TextureSamplerTexture.Sample(TextureSamplerSampler, (v_uv + vCurrOffset));
       currSampledHeight = x_397.w;
       if ((currSampledHeight > currRayHeight)) {

test/bug/tint/990.wgsl.expected.hlsl

@@ -6,7 +6,7 @@ void unused_entry_point() {
 void f() {
   int i = 0;
   {
-    for(; ; ) {
+    [loop] for(; ; ) {
     }
   }
 }

test/layout/storage/mat2x2/stride/16.spvasm.expected.hlsl

@@ -18,7 +18,7 @@ typedef tint_padded_array_element tint_symbol_ret[2];
 tint_symbol_ret tint_symbol(RWByteAddressBuffer buffer, uint offset) {
   tint_padded_array_element arr_1[2] = (tint_padded_array_element[2])0;
   {
-    for(uint i = 0u; (i < 2u); i = (i + 1u)) {
+    [loop] for(uint i = 0u; (i < 2u); i = (i + 1u)) {
       arr_1[i].el = asfloat(buffer.Load2((offset + (i * 16u))));
     }
   }
@@ -28,7 +28,7 @@ tint_symbol_ret tint_symbol(RWByteAddressBuffer buffer, uint offset) {
 void tint_symbol_2(RWByteAddressBuffer buffer, uint offset, tint_padded_array_element value[2]) {
   tint_padded_array_element array[2] = value;
   {
-    for(uint i_1 = 0u; (i_1 < 2u); i_1 = (i_1 + 1u)) {
+    [loop] for(uint i_1 = 0u; (i_1 < 2u); i_1 = (i_1 + 1u)) {
       buffer.Store2((offset + (i_1 * 16u)), asuint(array[i_1].el));
     }
   }

test/loops/loop.wgsl.expected.hlsl

@@ -5,7 +5,7 @@ void unused_entry_point() {
 
 int f() {
   int i = 0;
-  while (true) {
+  [loop] while (true) {
     i = (i + 1);
     if ((i > 4)) {
       return i;

test/loops/loop_with_continuing.wgsl.expected.hlsl

@@ -5,7 +5,7 @@ void unused_entry_point() {
 
 int f() {
   int i = 0;
-  while (true) {
+  [loop] while (true) {
     if ((i > 4)) {
       return i;
     }

test/loops/nested_loops.wgsl.expected.hlsl

@@ -6,12 +6,12 @@ void unused_entry_point() {
 int f() {
   int i = 0;
   int j = 0;
-  while (true) {
+  [loop] while (true) {
     i = (i + 1);
     if ((i > 4)) {
       return 1;
     }
-    while (true) {
+    [loop] while (true) {
       j = (j + 1);
       if ((j > 4)) {
         return 2;

test/loops/nested_loops_with_continuing.wgsl.expected.hlsl

@@ -6,11 +6,11 @@ void unused_entry_point() {
 int f() {
   int i = 0;
   int j = 0;
-  while (true) {
+  [loop] while (true) {
     if ((i > 4)) {
       return 1;
     }
-    while (true) {
+    [loop] while (true) {
       if ((j > 4)) {
         return 2;
       }

test/samples/compute_boids.wgsl.expected.hlsl

@@ -60,7 +60,7 @@ void comp_main_inner(uint3 gl_GlobalInvocationID) {
   float2 pos = float2(0.0f, 0.0f);
   float2 vel = float2(0.0f, 0.0f);
   {
-    for(uint i = 0u; (i < 5u); i = (i + 1u)) {
+    [loop] for(uint i = 0u; (i < 5u); i = (i + 1u)) {
       if ((i == index)) {
         continue;
       }

test/statements/for/basic.wgsl.expected.hlsl

@@ -8,7 +8,7 @@ void some_loop_body() {
 
 void f() {
   {
-    for(int i = 0; (i < 5); i = (i + 1)) {
+    [loop] for(int i = 0; (i < 5); i = (i + 1)) {
       some_loop_body();
     }
   }

test/statements/for/complex.wgsl.expected.hlsl

@@ -10,7 +10,7 @@ void f() {
   int j = 0;
   {
     int i = 0;
-    while (true) {
+    [loop] while (true) {
       bool tint_tmp = (i < 5);
       if (tint_tmp) {
         tint_tmp = (j < 10);

test/statements/for/condition.wgsl.expected.hlsl

@@ -6,7 +6,7 @@ void unused_entry_point() {
 void f() {
   int i = 0;
   {
-    for(; (i < 4); ) {
+    [loop] for(; (i < 4); ) {
     }
   }
 }

test/statements/for/continuing.wgsl.expected.hlsl

@@ -6,7 +6,7 @@ void unused_entry_point() {
 void f() {
   int i = 0;
   {
-    for(; ; i = (i + 1)) {
+    [loop] for(; ; i = (i + 1)) {
     }
   }
 }

test/statements/for/empty.wgsl.expected.hlsl

@@ -5,7 +5,7 @@ void unused_entry_point() {
 
 void f() {
   {
-    for(; ; ) {
+    [loop] for(; ; ) {
     }
   }
 }

test/statements/for/initializer.wgsl.expected.hlsl

@@ -5,7 +5,7 @@ void unused_entry_point() {
 
 void f() {
   {
-    for(int i = 0; ; ) {
+    [loop] for(int i = 0; ; ) {
     }
   }
 }

test/statements/for/scoping.wgsl.expected.hlsl

@@ -5,7 +5,7 @@ void unused_entry_point() {
 
 void f() {
   {
-    for(int must_not_collide = 0; ; ) {
+    [loop] for(int must_not_collide = 0; ; ) {
     }
   }
   int must_not_collide = 0;

test/var/initialization/workgroup/array.wgsl.expected.hlsl

@@ -6,7 +6,7 @@ struct tint_symbol_1 {
 
 void main_inner(uint local_invocation_index) {
   {
-    for(uint idx = local_invocation_index; (idx < 3u); idx = (idx + 1u)) {
+    [loop] for(uint idx = local_invocation_index; (idx < 3u); idx = (idx + 1u)) {
       const uint i = idx;
       v[i] = 0;
     }