Enable triton kernels matmul tests

wdziurdz · wdziurdz · commit 0c5358e16327 · 2025-10-31T12:43:16.000Z
Signed-off-by: Witold Dziurdz &lt;witold.dziurdz@intel.com&gt;
diff --git a/python/triton_kernels/triton_kernels/matmul_ogs.py b/python/triton_kernels/triton_kernels/matmul_ogs.py
@@ -596,7 +596,7 @@ def matmul_ogs(x, w, bias,
         out_matmul_scale = out_matmul_scale.data.view(torch.uint8)
         if has_scratchpad and "mx_out_scale" in memory["scratchpad"]:
             out_matmul_scale = memory["scratchpad"]["mx_out_scale"]
-    out_matmul_has_mx = out_matmul_scale is not None and out_matmul.element_size() == 1
+    out_matmul_has_mx = out_matmul_scale is not None and bitwidth(out_dtype) == 8
     # matrix multiplication
     flex = precision_config.flex_ctx
     bias_stride = None if bias is None else bias.stride(0)
diff --git a/scripts/skiplist/a770/triton_kernels.txt b/scripts/skiplist/a770/triton_kernels.txt
@@ -1,3 +1,2 @@
-tests/test_matmul.py::test_op
 tests/test_matmul.py::test_fused_act
 tests/test_matmul.py::test_zero_reduction_dim
diff --git a/scripts/skiplist/arl-h/triton_kernels.txt b/scripts/skiplist/arl-h/triton_kernels.txt
@@ -1,3 +1,2 @@
-tests/test_matmul.py::test_op
 tests/test_matmul.py::test_fused_act
 tests/test_matmul.py::test_zero_reduction_dim
diff --git a/scripts/skiplist/arl-s/triton_kernels.txt b/scripts/skiplist/arl-s/triton_kernels.txt
@@ -1,3 +1,2 @@
-tests/test_matmul.py::test_op
 tests/test_matmul.py::test_fused_act
 tests/test_matmul.py::test_zero_reduction_dim
diff --git a/scripts/skiplist/default/triton_kernels.txt b/scripts/skiplist/default/triton_kernels.txt
@@ -1,28 +1,16 @@
 # https://github.com/intel/intel-xpu-backend-for-triton/issues/5074
-tests/test_matmul.py::test_op[False-False-False-False-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-False-False-False-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-False-False-False-None-128-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-False-False-False-None-128-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-False-False-False-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-False-False-False-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-False-False-False-None-16-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-False-False-False-None-16-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-False-True-False-None-128-1000-400-400-ragged-float8_e4m3fn-float8_e4m3fn-3-1-1-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-False-False-True-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-False-False-True-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-False-True-False-None-128-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-False-True-False-None-128-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-False-True-False-None-16-1000-400-400-ragged-float8_e4m3fn-float8_e4m3fn-3-1-1-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-False-False-True-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-False-False-True-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-False-True-False-None-16-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-False-True-False-None-16-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-False-True-False-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-False-True-False-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-True-False-False-None-128-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-True-False-False-None-128-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-False-True-False-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-False-True-False-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-True-False-False-None-16-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-True-False-False-None-16-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-True-True-False-None-128-1000-400-400-ragged-float8_e4m3fn-float8_e4m3fn-3-1-1-False-None-False-False-False]
@@ -35,30 +23,18 @@ tests/test_matmul.py::test_op[False-False-True-True-False-None-16-1000-704-800-r
 tests/test_matmul.py::test_op[False-False-True-True-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-True-True-False-None-16-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-True-True-False-None-16-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-False-False-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-False-False-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-False-False-False-None-128-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-False-False-False-None-128-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-False-False-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-False-False-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-False-False-False-None-16-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-False-False-False-None-16-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-False-True-False-None-128-1000-400-400-ragged-float8_e4m3fn-float8_e4m3fn-3-1-1-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-False-True-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-False-True-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-False-True-False-None-128-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-False-True-False-None-128-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-False-True-False-None-16-1000-400-400-ragged-float8_e4m3fn-float8_e4m3fn-3-1-1-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-False-True-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-False-True-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-False-True-False-None-16-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-False-True-False-None-16-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-True-False-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-True-False-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-True-False-False-None-128-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-True-False-False-None-128-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-True-False-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-True-False-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-True-False-False-None-16-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-True-False-False-None-16-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-True-True-False-None-128-1000-400-400-ragged-float8_e4m3fn-float8_e4m3fn-3-1-1-False-None-False-False-False]
diff --git a/scripts/skiplist/lts/triton_kernels.txt b/scripts/skiplist/lts/triton_kernels.txt
@@ -1,28 +1,16 @@
 # https://github.com/intel/intel-xpu-backend-for-triton/issues/5074
-tests/test_matmul.py::test_op[False-False-False-False-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-False-False-False-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-False-False-False-None-128-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-False-False-False-None-128-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-False-False-False-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-False-False-False-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-False-False-False-None-16-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-False-False-False-None-16-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-False-True-False-None-128-1000-400-400-ragged-float8_e4m3fn-float8_e4m3fn-3-1-1-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-False-False-True-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-False-False-True-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-False-True-False-None-128-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-False-True-False-None-128-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-False-True-False-None-16-1000-400-400-ragged-float8_e4m3fn-float8_e4m3fn-3-1-1-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-False-False-True-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-False-False-True-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-False-True-False-None-16-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-False-True-False-None-16-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-False-True-False-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-False-True-False-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-True-False-False-None-128-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-True-False-False-None-128-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-False-True-False-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-False-True-False-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-True-False-False-None-16-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-True-False-False-None-16-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-True-True-False-None-128-1000-400-400-ragged-float8_e4m3fn-float8_e4m3fn-3-1-1-False-None-False-False-False]
@@ -35,30 +23,18 @@ tests/test_matmul.py::test_op[False-False-True-True-False-None-16-1000-704-800-r
 tests/test_matmul.py::test_op[False-False-True-True-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-True-True-False-None-16-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-False-True-True-False-None-16-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-False-False-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-False-False-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-False-False-False-None-128-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-False-False-False-None-128-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-False-False-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-False-False-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-False-False-False-None-16-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-False-False-False-None-16-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-False-True-False-None-128-1000-400-400-ragged-float8_e4m3fn-float8_e4m3fn-3-1-1-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-False-True-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-False-True-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-False-True-False-None-128-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-False-True-False-None-128-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-False-True-False-None-16-1000-400-400-ragged-float8_e4m3fn-float8_e4m3fn-3-1-1-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-False-True-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-False-True-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-False-True-False-None-16-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-False-True-False-None-16-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-True-False-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-True-False-False-None-128-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-True-False-False-None-128-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-True-False-False-None-128-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-True-False-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-False-None-False-False-False]
-tests/test_matmul.py::test_op[False-True-True-False-False-None-16-1000-704-800-ragged-mxfloat8_e4m3fn-mxfloat4_e2m1-8-2-9-True-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-True-False-False-None-16-16-16-1000-batched-float16-float16-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-True-False-False-None-16-16-16-1000-batched-float8_e5m2-float8_e5m2-5-1-None-False-None-False-False-False]
 tests/test_matmul.py::test_op[False-True-True-True-False-None-128-1000-400-400-ragged-float8_e4m3fn-float8_e4m3fn-3-1-1-False-None-False-False-False]
diff --git a/scripts/skiplist/xe2/triton_kernels.txt b/scripts/skiplist/xe2/triton_kernels.txt

Original file line number	Diff line number	Diff line change
`@@ -1,3 +1,2 @@`
`1`		`-tests/test_matmul.py::test_op`
`2`	`1`	`tests/test_matmul.py::test_fused_act`
`3`	`2`	`tests/test_matmul.py::test_zero_reduction_dim`