Test frame work improvements (#1452)

* Add -hide-ignored
Made API filter when enbled filter out non API tests.

* Add ability to set categories at file level.
Added wave, wave-mask and wave-active categories.

* Added -api-only flag.

* Don't synthesize tests from only CPU tests.

Co-authored-by: Tim Foley <tfoleyNV@users.noreply.github.com>

35 files changed, 198 insertions, 164 deletions

diff --git a/tests/hlsl-intrinsic/wave-active-product.slang b/tests/hlsl-intrinsic/wave-active-product.slang
index 67b9ca587..7fd42e038 100644
--- a/tests/hlsl-intrinsic/wave-active-product.slang
+++ b/tests/hlsl-intrinsic/wave-active-product.slang
@@ -1,8 +1,9 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
+//TEST_CATEGORY(wave, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-broadcast-lane-at-vk.slang b/tests/hlsl-intrinsic/wave-broadcast-lane-at-vk.slang
index 3c746476a..b67bb601b 100644
--- a/tests/hlsl-intrinsic/wave-broadcast-lane-at-vk.slang
+++ b/tests/hlsl-intrinsic/wave-broadcast-lane-at-vk.slang
@@ -1,5 +1,6 @@
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST_CATEGORY(wave, compute)
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-broadcast-lane-at.slang b/tests/hlsl-intrinsic/wave-broadcast-lane-at.slang
index bb1523219..ee5ce45ab 100644
--- a/tests/hlsl-intrinsic/wave-broadcast-lane-at.slang
+++ b/tests/hlsl-intrinsic/wave-broadcast-lane-at.slang
@@ -1,9 +1,10 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST_CATEGORY(wave, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
 // Disabled on VK because glsl can't do WaveReadLaneAt on matrix. 
-//DISABLE_TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
+//DISABLE_TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-diverge.slang b/tests/hlsl-intrinsic/wave-diverge.slang
index 2d500d0c8..1f0e6aac5 100644
--- a/tests/hlsl-intrinsic/wave-diverge.slang
+++ b/tests/hlsl-intrinsic/wave-diverge.slang
@@ -1,8 +1,9 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
+//TEST_CATEGORY(wave, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-equality.slang b/tests/hlsl-intrinsic/wave-equality.slang
index 7ed67b632..fc8d28f10 100644
--- a/tests/hlsl-intrinsic/wave-equality.slang
+++ b/tests/hlsl-intrinsic/wave-equality.slang
@@ -1,8 +1,9 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
+//TEST_CATEGORY(wave, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-get-lane-index.slang b/tests/hlsl-intrinsic/wave-get-lane-index.slang
index 50cf9822c..84b1c3fea 100644
--- a/tests/hlsl-intrinsic/wave-get-lane-index.slang
+++ b/tests/hlsl-intrinsic/wave-get-lane-index.slang
@@ -1,8 +1,9 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute
+//TEST_CATEGORY(wave, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-is-first-lane.slang b/tests/hlsl-intrinsic/wave-is-first-lane.slang
index e1f0faef3..4b78647ee 100644
--- a/tests/hlsl-intrinsic/wave-is-first-lane.slang
+++ b/tests/hlsl-intrinsic/wave-is-first-lane.slang
@@ -1,8 +1,9 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
+//TEST_CATEGORY(wave, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-mask/wave-active-product.slang b/tests/hlsl-intrinsic/wave-mask/wave-active-product.slang
index 3135586a5..682622580 100644
--- a/tests/hlsl-intrinsic/wave-mask/wave-active-product.slang
+++ b/tests/hlsl-intrinsic/wave-mask/wave-active-product.slang
@@ -1,8 +1,9 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute
+//TEST_CATEGORY(wave-mask, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-mask/wave-broadcast-lane-at-vk.slang b/tests/hlsl-intrinsic/wave-mask/wave-broadcast-lane-at-vk.slang
index 94a44df97..552ef7144 100644
--- a/tests/hlsl-intrinsic/wave-mask/wave-broadcast-lane-at-vk.slang
+++ b/tests/hlsl-intrinsic/wave-mask/wave-broadcast-lane-at-vk.slang
@@ -1,5 +1,6 @@
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST_CATEGORY(wave-mask, compute)
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-mask/wave-broadcast-lane-at.slang b/tests/hlsl-intrinsic/wave-mask/wave-broadcast-lane-at.slang
index 62d29085f..1fe8a54b2 100644
--- a/tests/hlsl-intrinsic/wave-mask/wave-broadcast-lane-at.slang
+++ b/tests/hlsl-intrinsic/wave-mask/wave-broadcast-lane-at.slang
@@ -1,9 +1,10 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST_CATEGORY(wave-mask, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
 // Disabled on VK because glsl can't do WaveReadLaneAt on matrix. 
-//DISABLE_TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute
+//DISABLE_TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-mask/wave-diverge.slang b/tests/hlsl-intrinsic/wave-mask/wave-diverge.slang
index 51b7d3aeb..63176bac1 100644
--- a/tests/hlsl-intrinsic/wave-mask/wave-diverge.slang
+++ b/tests/hlsl-intrinsic/wave-mask/wave-diverge.slang
@@ -1,8 +1,9 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute
+//TEST_CATEGORY(wave-mask, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-mask/wave-equality.slang b/tests/hlsl-intrinsic/wave-mask/wave-equality.slang
index 13887b2c8..020e6415a 100644
--- a/tests/hlsl-intrinsic/wave-mask/wave-equality.slang
+++ b/tests/hlsl-intrinsic/wave-mask/wave-equality.slang
@@ -1,8 +1,9 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
+//TEST_CATEGORY(wave-mask, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-mask/wave-get-active.slang b/tests/hlsl-intrinsic/wave-mask/wave-get-active.slang
index 393cf7199..7f2c63b86 100644
--- a/tests/hlsl-intrinsic/wave-mask/wave-get-active.slang
+++ b/tests/hlsl-intrinsic/wave-mask/wave-get-active.slang
@@ -1,8 +1,9 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute
+//TEST_CATEGORY(wave-mask, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cuda -compute
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-mask/wave-get-converged.slang b/tests/hlsl-intrinsic/wave-mask/wave-get-converged.slang
index 9b2e814f2..d6c6cd380 100644
--- a/tests/hlsl-intrinsic/wave-mask/wave-get-converged.slang
+++ b/tests/hlsl-intrinsic/wave-mask/wave-get-converged.slang
@@ -1,8 +1,9 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute
+//TEST_CATEGORY(wave-mask, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-mask/wave-is-first-lane.slang b/tests/hlsl-intrinsic/wave-mask/wave-is-first-lane.slang
index 2b332bfb5..8d126f2ec 100644
--- a/tests/hlsl-intrinsic/wave-mask/wave-is-first-lane.slang
+++ b/tests/hlsl-intrinsic/wave-mask/wave-is-first-lane.slang
@@ -1,8 +1,9 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-cuda -compute
+//TEST_CATEGORY(wave-mask, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-mask/wave-mask-prefix.slang b/tests/hlsl-intrinsic/wave-mask/wave-mask-prefix.slang
index a2f25d71b..e80714d02 100644
--- a/tests/hlsl-intrinsic/wave-mask/wave-mask-prefix.slang
+++ b/tests/hlsl-intrinsic/wave-mask/wave-mask-prefix.slang
@@ -1,11 +1,12 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
+//TEST_CATEGORY(wave-mask, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
 // We need SM6.5 for these tests
 // Disable because version of dxc we are currently using doesn't support SM6.5
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile sm_6_5
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile sm_6_5
 // Disabled because we don't have GLSL intrinsics for these it seems 
-//DISABLE_TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
+//DISABLE_TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-mask/wave-matrix.slang b/tests/hlsl-intrinsic/wave-mask/wave-matrix.slang
index dadce4051..3fd551be9 100644
--- a/tests/hlsl-intrinsic/wave-mask/wave-matrix.slang
+++ b/tests/hlsl-intrinsic/wave-mask/wave-matrix.slang
@@ -1,8 +1,9 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//DISABLE_TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute
+//TEST_CATEGORY(wave-mask, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//DISABLE_TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-mask/wave-prefix-product.slang b/tests/hlsl-intrinsic/wave-mask/wave-prefix-product.slang
index 14ac8d652..56dbf7e4e 100644
--- a/tests/hlsl-intrinsic/wave-mask/wave-prefix-product.slang
+++ b/tests/hlsl-intrinsic/wave-mask/wave-prefix-product.slang
@@ -1,8 +1,9 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute
+//TEST_CATEGORY(wave-mask, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-mask/wave-prefix-sum.slang b/tests/hlsl-intrinsic/wave-mask/wave-prefix-sum.slang
index f756398b8..6149d1cca 100644
--- a/tests/hlsl-intrinsic/wave-mask/wave-prefix-sum.slang
+++ b/tests/hlsl-intrinsic/wave-mask/wave-prefix-sum.slang
@@ -1,8 +1,9 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute
+//TEST_CATEGORY(wave-mask, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-mask/wave-read-lane-at-vk.slang b/tests/hlsl-intrinsic/wave-mask/wave-read-lane-at-vk.slang
index e0464ae68..39b01bec3 100644
--- a/tests/hlsl-intrinsic/wave-mask/wave-read-lane-at-vk.slang
+++ b/tests/hlsl-intrinsic/wave-mask/wave-read-lane-at-vk.slang
@@ -1,8 +1,9 @@
 // This is similar to wave-lane-at.slang but tests more limited supported types for vk.
 // We have this 'simple' test, because we can't do matrix (or imat) operations on GLSL/Vk target
 
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST_CATEGORY(wave-mask, compute)
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-mask/wave-read-lane-at.slang b/tests/hlsl-intrinsic/wave-mask/wave-read-lane-at.slang
index 1b231fbc1..e2dccffea 100644
--- a/tests/hlsl-intrinsic/wave-mask/wave-read-lane-at.slang
+++ b/tests/hlsl-intrinsic/wave-mask/wave-read-lane-at.slang
@@ -1,9 +1,10 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST_CATEGORY(wave-mask, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
 // Disabled on VK because glsl can't do WaveReadLaneAt on matrix. 
-//DISABLE_TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute
+//DISABLE_TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-mask/wave-shuffle-vk.slang b/tests/hlsl-intrinsic/wave-mask/wave-shuffle-vk.slang
index fa9f4b3e5..fe4bedfe7 100644
--- a/tests/hlsl-intrinsic/wave-mask/wave-shuffle-vk.slang
+++ b/tests/hlsl-intrinsic/wave-mask/wave-shuffle-vk.slang
@@ -1,9 +1,10 @@
+//TEST_CATEGORY(wave-mask, compute)
 // Disabled because main tests is wave-shuffle.slang, this just tests VK 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-mask/wave-shuffle.slang b/tests/hlsl-intrinsic/wave-mask/wave-shuffle.slang
index a559aae8f..d108fc39e 100644
--- a/tests/hlsl-intrinsic/wave-mask/wave-shuffle.slang
+++ b/tests/hlsl-intrinsic/wave-mask/wave-shuffle.slang
@@ -1,10 +1,11 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
+//TEST_CATEGORY(wave-mask, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
 //Disabled on D3D, because in general WaveShuffle requires hardware that doesn't have the 'uniform laneId across Wave' restriction. 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
 // Disabled because vk doesn't currently support matrix types. See wave-shuffle-vk.slang
-//DISABLE_TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute
+//DISABLE_TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-mask/wave-vector.slang b/tests/hlsl-intrinsic/wave-mask/wave-vector.slang
index 083e1b5b9..dc6947bfd 100644
--- a/tests/hlsl-intrinsic/wave-mask/wave-vector.slang
+++ b/tests/hlsl-intrinsic/wave-mask/wave-vector.slang
@@ -1,8 +1,9 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute
+//TEST_CATEGORY(wave-mask, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-mask/wave.slang b/tests/hlsl-intrinsic/wave-mask/wave.slang
index 346940cb2..9d6fbac46 100644
--- a/tests/hlsl-intrinsic/wave-mask/wave.slang
+++ b/tests/hlsl-intrinsic/wave-mask/wave.slang
@@ -1,8 +1,9 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute
+//TEST_CATEGORY(wave-mask, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0  0 0 0 0  0 0 0 0  0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-matrix.slang b/tests/hlsl-intrinsic/wave-matrix.slang
index 50acc4a71..dd239a155 100644
--- a/tests/hlsl-intrinsic/wave-matrix.slang
+++ b/tests/hlsl-intrinsic/wave-matrix.slang
@@ -1,8 +1,9 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//DISABLE_TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
+//TEST_CATEGORY(wave, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//DISABLE_TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-multi-prefix.slang b/tests/hlsl-intrinsic/wave-multi-prefix.slang
index a1eb0e7a9..a3adc1a80 100644
--- a/tests/hlsl-intrinsic/wave-multi-prefix.slang
+++ b/tests/hlsl-intrinsic/wave-multi-prefix.slang
@@ -1,11 +1,12 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
+//TEST_CATEGORY(wave, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
 // We need SM6.5 for these tests
 // Disable because version of dxc we are currently using doesn't support SM6.5
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile sm_6_5
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile sm_6_5
 // Disabled because we don't have GLSL intrinsics for these it seems 
-//DISABLE_TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
+//DISABLE_TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-prefix-product.slang b/tests/hlsl-intrinsic/wave-prefix-product.slang
index 436ce06f7..888d3ba22 100644
--- a/tests/hlsl-intrinsic/wave-prefix-product.slang
+++ b/tests/hlsl-intrinsic/wave-prefix-product.slang
@@ -1,8 +1,9 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
+//TEST_CATEGORY(wave, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-prefix-sum.slang b/tests/hlsl-intrinsic/wave-prefix-sum.slang
index 358debedc..4b86da763 100644
--- a/tests/hlsl-intrinsic/wave-prefix-sum.slang
+++ b/tests/hlsl-intrinsic/wave-prefix-sum.slang
@@ -1,8 +1,9 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
+//TEST_CATEGORY(wave, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-read-lane-at-vk.slang b/tests/hlsl-intrinsic/wave-read-lane-at-vk.slang
index 3bd6b36b8..f085081b0 100644
--- a/tests/hlsl-intrinsic/wave-read-lane-at-vk.slang
+++ b/tests/hlsl-intrinsic/wave-read-lane-at-vk.slang
@@ -1,8 +1,8 @@
 // This is similar to wave-lane-at.slang but tests more limited supported types for vk.
 // We have this 'simple' test, because we can't do matrix (or imat) operations on GLSL/Vk target
-
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST_CATEGORY(wave, compute)
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-read-lane-at.slang b/tests/hlsl-intrinsic/wave-read-lane-at.slang
index 4d840ad0f..bc32cf1f8 100644
--- a/tests/hlsl-intrinsic/wave-read-lane-at.slang
+++ b/tests/hlsl-intrinsic/wave-read-lane-at.slang
@@ -1,9 +1,10 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST_CATEGORY(wave, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
 // Disabled on VK because glsl can't do WaveReadLaneAt on matrix. 
-//DISABLE_TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
+//DISABLE_TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-shuffle-vk.slang b/tests/hlsl-intrinsic/wave-shuffle-vk.slang
index 820a69e89..ecd325c95 100644
--- a/tests/hlsl-intrinsic/wave-shuffle-vk.slang
+++ b/tests/hlsl-intrinsic/wave-shuffle-vk.slang
@@ -1,9 +1,10 @@
+//TEST_CATEGORY(wave, compute)
 // Disabled because main tests is wave-shuffle.slang, this just tests VK 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-shuffle.slang b/tests/hlsl-intrinsic/wave-shuffle.slang
index 608e57ffd..30f3db3c8 100644
--- a/tests/hlsl-intrinsic/wave-shuffle.slang
+++ b/tests/hlsl-intrinsic/wave-shuffle.slang
@@ -1,10 +1,11 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
+//TEST_CATEGORY(wave, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
 //Disabled on D3D, because in general WaveShuffle requires hardware that doesn't have the 'uniform laneId across Wave' restriction. 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
 // Disabled because vk doesn't currently support matrix types. See wave-shuffle-vk.slang
-//DISABLE_TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
+//DISABLE_TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave-vector.slang b/tests/hlsl-intrinsic/wave-vector.slang
index 784f16c16..3a3f4b4be 100644
--- a/tests/hlsl-intrinsic/wave-vector.slang
+++ b/tests/hlsl-intrinsic/wave-vector.slang
@@ -1,8 +1,9 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
+//TEST_CATEGORY(wave, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/wave.slang b/tests/hlsl-intrinsic/wave.slang
index eb647d0fe..04a6c2048 100644
--- a/tests/hlsl-intrinsic/wave.slang
+++ b/tests/hlsl-intrinsic/wave.slang
@@ -1,8 +1,9 @@
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
+//TEST_CATEGORY(wave, compute)
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute
+//TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;