[BACKEND] fix test, rm benchmark.select_cpu_backend code

tsingmicro-public-e · tsingmicro-public-e · commit 0c3ef6b49f06 · 2026-03-12T19:10:07.000+08:00
diff --git a/third_party/tsingmicro/examples/bare_matmul.py b/third_party/tsingmicro/examples/bare_matmul.py
@@ -45,7 +45,6 @@ def bench_matmul(N, provider):
 
 
 if __name__ == "__main__":
-    # benchmark.select_cpu_backend()
     for provider in ['test']:
         bench_matmul(16, provider)
     # for X in [2**i for i in range(7, 10, 1)]:
diff --git a/third_party/tsingmicro/examples/bare_matmul_acc.py b/third_party/tsingmicro/examples/bare_matmul_acc.py
@@ -41,7 +41,6 @@ def bench_matmul(N, provider):
 
 
 if __name__ == "__main__":
-    benchmark.select_cpu_backend()
     for X in [2**i for i in range(7, 10, 1)]:
         for provider in ['test', 'torch', 'triton']:
             bench_matmul(X, provider)
diff --git a/third_party/tsingmicro/examples/bare_matmul_autotune.py b/third_party/tsingmicro/examples/bare_matmul_autotune.py
@@ -66,5 +66,4 @@ def bench_matmul(M, N, K):
 
 
 if __name__ == "__main__":
-    # benchmark.select_cpu_backend()
     bench_matmul(16, 16, 16)
diff --git a/third_party/tsingmicro/examples/mult_ir.py b/third_party/tsingmicro/examples/mult_ir.py
@@ -182,7 +182,6 @@ def bench_matmul(a, b):
 
 
 if __name__ == "__main__":
-    # benchmark.select_cpu_backend()
     M = 1024
     K = 1024
     N = 1024
diff --git a/third_party/tsingmicro/examples/profile_matmul.py b/third_party/tsingmicro/examples/profile_matmul.py
@@ -187,7 +187,6 @@ def bench_matmul(M, N, K, provider):
 
 
 if __name__ == "__main__":
-    # benchmark.select_cpu_backend()
     bench_matmul(4096, 4096, 4096, 'triton')
     # for X in [128 * i for i in range(2, 7)]:
     #     for provider in ['torch', 'triton']:
diff --git a/third_party/tsingmicro/examples/test_argmax2d.py b/third_party/tsingmicro/examples/test_argmax2d.py
@@ -4,8 +4,6 @@
 import pytest
 import benchmark
 
-benchmark.select_cpu_backend()
-
 
 @triton.jit
 def argmax_kernel_2d(
diff --git a/third_party/tsingmicro/examples/test_bare_matmul.py b/third_party/tsingmicro/examples/test_bare_matmul.py
@@ -58,7 +58,6 @@ def test_bare_matmul(N, dtype, device='cpu'):
 
 
 if __name__ == "__main__":
-    benchmark.select_cpu_backend()
     test_bare_matmul(128, torch.float32)
     # for X in [2**i for i in range(7, 10, 1)]:
     #     for provider in ['test', 'torch', 'triton']:
diff --git a/third_party/tsingmicro/examples/test_bare_matmul_acc.py b/third_party/tsingmicro/examples/test_bare_matmul_acc.py
@@ -59,7 +59,6 @@ def test_bare_matmul_acc(N, dtype, device='cpu'):
 
 
 if __name__ == "__main__":
-    benchmark.select_cpu_backend()
     for X in [2**i for i in range(7, 10, 1)]:
         for provider in ['test', 'torch', 'triton']:
             bench_matmul(X, provider)
diff --git a/third_party/tsingmicro/examples/test_cdiv.py b/third_party/tsingmicro/examples/test_cdiv.py
@@ -101,6 +101,5 @@ def benchmark_cdiv_triton(size, dtype, provider):
 
 
 if __name__ == "__main__":
-    # benchmark.select_cpu_backend()
     for X in [2**i for i in range(22, 25, 1)]:
         benchmark_cdiv_triton(X, torch.int32, "triton")
diff --git a/third_party/tsingmicro/examples/test_ceil.py b/third_party/tsingmicro/examples/test_ceil.py
@@ -91,6 +91,5 @@ def benchmark_ceil_triton(size, dtype, provider):
 
 
 if __name__ == "__main__":
-    benchmark.select_cpu_backend()
     for size in [2**i for i in range(22, 25, 1)]:
         benchmark_ceil_triton(size, torch.float32, provider="triton")
diff --git a/third_party/tsingmicro/examples/test_clamp.py b/third_party/tsingmicro/examples/test_clamp.py
@@ -97,6 +97,5 @@ def benchmark_clamp_triton(size, dtype, provider):
 
 
 if __name__ == "__main__":
-    benchmark.select_cpu_backend()
     for size in [2**i for i in range(22, 25, 1)]:
         benchmark_clamp_triton(size, torch.float32, "triton")
diff --git a/third_party/tsingmicro/examples/test_cos.py b/third_party/tsingmicro/examples/test_cos.py
@@ -94,6 +94,5 @@ def benchmark_cos_triton(size, dtype, provider):
 
 
 if __name__ == "__main__":
-    # benchmark.select_cpu_backend()
     for size in [2**i for i in range(22, 25, 1)]:
         benchmark_cos_triton(size, torch.float32, "triton")
diff --git a/third_party/tsingmicro/examples/test_div_rn.py b/third_party/tsingmicro/examples/test_div_rn.py
@@ -97,6 +97,5 @@ def benchmark_div_rn_triton(size, dtype, provider):
 
 
 if __name__ == "__main__":
-    benchmark.select_cpu_backend()
     for i in [2**i for i in range(22, 25, 1)]:
         benchmark_div_rn_triton(i, torch.float32, provider="triton")
diff --git a/third_party/tsingmicro/examples/test_dot_scaled.py b/third_party/tsingmicro/examples/test_dot_scaled.py
@@ -5,8 +5,6 @@
 import itertools
 import benchmark
 
-benchmark.select_cpu_backend()
-
 from triton._internal_testing import (
     integral_dtypes,
     int_dtypes,
diff --git a/third_party/tsingmicro/examples/test_embedding.py b/third_party/tsingmicro/examples/test_embedding.py
@@ -90,5 +90,4 @@ def test_embedding(M, N, dtype, device='cpu'):
 
 
 if __name__ == "__main__":
-    # benchmark.select_cpu_backend()
     test_embedding(1151, 8192, torch.float32)
diff --git a/third_party/tsingmicro/examples/test_exp.py b/third_party/tsingmicro/examples/test_exp.py
@@ -87,6 +87,5 @@ def benchmark_exp_triton(size, dtype, provider):
 
 
 if __name__ == "__main__":
-    benchmark.select_cpu_backend()
     for size in [i**2 for i in range(22, 25, 1)]:
         benchmark_exp_triton(size, torch.float32, "triton")
diff --git a/third_party/tsingmicro/examples/test_exp2.py b/third_party/tsingmicro/examples/test_exp2.py
@@ -89,6 +89,5 @@ def benchmark_exp2_triton(size, dtype, provider):
 
 
 if __name__ == "__main__":
-    benchmark.select_cpu_backend()
     for size in [i**2 for i in range(22, 25, 1)]:
         benchmark_exp2_triton(size, torch.float32, "triton")
diff --git a/third_party/tsingmicro/examples/test_fdiv.py b/third_party/tsingmicro/examples/test_fdiv.py
@@ -94,6 +94,5 @@ def benchmark_fdiv_triton(size, dtype, provider):
 
 
 if __name__ == "__main__":
-    benchmark.select_cpu_backend()
     for size in [i**2 for i in range(22, 25, 1)]:
         benchmark_fdiv_triton(size, torch.float32, "triton")
diff --git a/third_party/tsingmicro/examples/test_flip.py b/third_party/tsingmicro/examples/test_flip.py
@@ -4,7 +4,6 @@
 import triton.language as tl
 import benchmark
 
-# benchmark.select_cpu_backend()
 from triton._internal_testing import numpy_random
 
 
diff --git a/third_party/tsingmicro/examples/test_floor.py b/third_party/tsingmicro/examples/test_floor.py
@@ -91,6 +91,5 @@ def benchmark_floor_triton(size, dtype, provider):
 
 
 if __name__ == "__main__":
-    benchmark.select_cpu_backend()
     for size in [i**2 for i in range(22, 25, 1)]:
         benchmark_floor_triton(size, torch.float32, provider="triton")
diff --git a/third_party/tsingmicro/examples/test_fma.py b/third_party/tsingmicro/examples/test_fma.py
@@ -101,6 +101,5 @@ def benchmark_fma_triton(size, dtype, provider):
 
 
 if __name__ == "__main__":
-    benchmark.select_cpu_backend()
     for size in [i**2 for i in range(22, 25, 1)]:
         benchmark_fma_triton(size, torch.float32, "triton")
diff --git a/third_party/tsingmicro/examples/test_histogram.py b/third_party/tsingmicro/examples/test_histogram.py
@@ -32,8 +32,6 @@ def histogram_kernel(x_ptr, z_ptr, M: tl.constexpr, N: tl.constexpr):
 
 
 if __name__ == "__main__":
-    import benchmark
-    benchmark.select_cpu_backend()
     test_histogram(1024, 8, 'cpu')
     # test_histogram_2d(32, 16, 8, 'cpu')
     # test_histogram(2048, 2, 'cpu')
diff --git a/third_party/tsingmicro/examples/test_layernorm.py b/third_party/tsingmicro/examples/test_layernorm.py
@@ -175,7 +175,6 @@ def bench_layernorm(size, provider):
 
 
 if __name__ == "__main__":
-    # benchmark.select_cpu_backend()
     for X in [2**i for i in range(10, 13, 1)]:
         for provider in ['triton']:
             bench_layernorm(X, provider)
diff --git a/third_party/tsingmicro/examples/test_load6d.py b/third_party/tsingmicro/examples/test_load6d.py
@@ -4,8 +4,6 @@
 import pytest
 import benchmark
 
-benchmark.select_cpu_backend()
-
 
 @triton.jit
 def six_dim_load(T_ptr, output_ptr, B, C, D1, D2, D3, D4, stride_b, stride_c, stride_d1, stride_d2, stride_d3,
diff --git a/third_party/tsingmicro/examples/test_log.py b/third_party/tsingmicro/examples/test_log.py
@@ -87,6 +87,5 @@ def benchmark_log_triton(size, dtype, provider):
 
 
 if __name__ == "__main__":
-    benchmark.select_cpu_backend()
     for size in [i**2 for i in range(22, 25, 1)]:
         benchmark_log_triton(size, torch.float32, "triton")
diff --git a/third_party/tsingmicro/examples/test_log2.py b/third_party/tsingmicro/examples/test_log2.py
@@ -87,6 +87,5 @@ def benchmark_log2_triton(size, dtype, provider):
 
 
 if __name__ == "__main__":
-    benchmark.select_cpu_backend()
     for size in [i**2 for i in range(22, 25, 1)]:
         benchmark_log2_triton(size, torch.float32, "triton")
diff --git a/third_party/tsingmicro/examples/test_math_erf_op.py b/third_party/tsingmicro/examples/test_math_erf_op.py
@@ -10,7 +10,6 @@
 import inspect
 import benchmark
 
-benchmark.select_cpu_backend()
 from numpy.random import RandomState
 
 from triton._internal_testing import (
diff --git a/third_party/tsingmicro/examples/test_matmul.py b/third_party/tsingmicro/examples/test_matmul.py
@@ -179,9 +179,8 @@ def bench_matmul(M, N, K, provider):
 
 
 if __name__ == "__main__":
-    # benchmark.select_cpu_backend()
     # test_matmul(179,167,321,torch.float32)
-    test_matmul("txda")
-    # for X in [128 * i for i in range(2, 7)]:
-    #     for provider in ['torch', 'triton']:
-    #         bench_matmul(X, X, X, provider)
+    # test_matmul("txda")
+    for X in [128 * i for i in range(2, 7)]:
+        for provider in ['torch', 'triton']:
+            bench_matmul(X, X, X, provider)
diff --git a/third_party/tsingmicro/examples/test_maximum.py b/third_party/tsingmicro/examples/test_maximum.py
@@ -92,6 +92,5 @@ def benchmark_maximum_triton(size, dtype, provider):
 
 
 if __name__ == "__main__":
-    benchmark.select_cpu_backend()
     for size in [i**2 for i in range(22, 25, 1)]:
         benchmark_maximum_triton(size, torch.float32, provider="triton")
diff --git a/third_party/tsingmicro/examples/test_minimum.py b/third_party/tsingmicro/examples/test_minimum.py
@@ -92,6 +92,5 @@ def benchmark_minimum_triton(size, dtype, provider):
 
 
 if __name__ == "__main__":
-    benchmark.select_cpu_backend()
     for size in [i**2 for i in range(22, 25, 1)]:
         benchmark_minimum_triton(size, torch.float32, provider="triton")
diff --git a/third_party/tsingmicro/examples/test_reduce1d.py b/third_party/tsingmicro/examples/test_reduce1d.py
@@ -5,8 +5,6 @@
 import triton.language as tl
 import benchmark
 
-benchmark.select_cpu_backend()
-
 
 @triton.jit
 def reduce_kernel_1d(
diff --git a/third_party/tsingmicro/examples/test_rsqrt.py b/third_party/tsingmicro/examples/test_rsqrt.py
@@ -87,6 +87,5 @@ def benchmark_rsqrt_triton(size, dtype, provider):
 
 
 if __name__ == "__main__":
-    benchmark.select_cpu_backend()
     for size in [i**2 for i in range(22, 25, 1)]:
         benchmark_rsqrt_triton(size, torch.float32, "triton")
diff --git a/third_party/tsingmicro/examples/test_scan.py b/third_party/tsingmicro/examples/test_scan.py
@@ -63,6 +63,5 @@ def scan_kernel_reverse(out_ptr, in_ptr, n_elements, M: tl.constexpr, N: tl.cons
 
 
 if __name__ == "__main__":
-    benchmark.select_cpu_backend()
     test_scan_1d(32, 2, False, "cpu")
     test_scan_1d(32, 2, True, "cpu")
diff --git a/third_party/tsingmicro/examples/test_scan2d.py b/third_party/tsingmicro/examples/test_scan2d.py
@@ -206,7 +206,6 @@ def kernel(X, Y, Z, BLOCK_M: tl.constexpr, BLOCK_N: tl.constexpr, AXIS: tl.const
 
 
 if __name__ == "__main__":
-    benchmark.select_cpu_backend()
     test_scan2d('cumsum', 'float32', (8, 32), 1, False, 'cpu')
     test_scan2d('cumsum', 'float32', (8, 32), 0, True, 'cpu')
     test_scan2d('cumsum', 'float32', (8, 32), 0, False, 'cpu')
diff --git a/third_party/tsingmicro/examples/test_scan3d.py b/third_party/tsingmicro/examples/test_scan3d.py
@@ -152,7 +152,6 @@ def kernel(X, Y, Z, BLOCK_M: tl.constexpr, BLOCK_N: tl.constexpr, BLOCK_K: tl.co
 
 
 if __name__ == "__main__":
-    benchmark.select_cpu_backend()
     test_scan3d('cumsum', 'float32', (2, 2, 32), 2, False, 'cpu')
     test_scan3d('cumsum', 'float32', (2, 2, 32), 0, False, 'cpu')
     test_scan3d('cumsum', 'float32', (2, 2, 32), 1, False, 'cpu')
diff --git a/third_party/tsingmicro/examples/test_sigmoid.py b/third_party/tsingmicro/examples/test_sigmoid.py
@@ -87,6 +87,5 @@ def benchmark_sigmoid_triton(size, dtype, provider):
 
 
 if __name__ == "__main__":
-    benchmark.select_cpu_backend()
     for size in [i**2 for i in range(22, 25, 1)]:
         benchmark_sigmoid_triton(size, torch.float32, "triton")
diff --git a/third_party/tsingmicro/examples/test_sin.py b/third_party/tsingmicro/examples/test_sin.py
@@ -95,6 +95,5 @@ def benchmark_sin_triton(size, dtype, provider):
 
 
 if __name__ == "__main__":
-    # benchmark.select_cpu_backend()
     for size in [i**2 for i in range(22, 25, 1)]:
         benchmark_sin_triton(size, torch.float32, "triton")
diff --git a/third_party/tsingmicro/examples/test_softmax.py b/third_party/tsingmicro/examples/test_softmax.py
@@ -83,7 +83,6 @@ def bench_softmax(size, provider):
 
 
 if __name__ == "__main__":
-    # benchmark.select_cpu_backend()
     for X in [2**i for i in range(10, 14, 1)]:
         for provider in ['torch', 'triton']:
             bench_softmax(X, provider)
diff --git a/third_party/tsingmicro/examples/test_sqrt.py b/third_party/tsingmicro/examples/test_sqrt.py
@@ -89,6 +89,5 @@ def benchmark_sqrt_triton(size, dtype, provider):
 
 
 if __name__ == "__main__":
-    benchmark.select_cpu_backend()
     for size in [i**2 for i in range(22, 25, 1)]:
         benchmark_sqrt_triton(size, torch.float32, "triton")
diff --git a/third_party/tsingmicro/examples/test_sqrt_rn.py b/third_party/tsingmicro/examples/test_sqrt_rn.py
@@ -88,6 +88,5 @@ def benchmark_sqrt_rn_triton(size, dtype, provider):
 
 
 if __name__ == "__main__":
-    benchmark.select_cpu_backend()
     for size in [i**2 for i in range(22, 25, 1)]:
         benchmark_sqrt_rn_triton(size, torch.float32, "triton")
diff --git a/third_party/tsingmicro/examples/test_trans2d.py b/third_party/tsingmicro/examples/test_trans2d.py
@@ -30,5 +30,4 @@ def kernel(In, Out, in_shape1: tl.constexpr, in_shape2: tl.constexpr, ou_shape1:
 
 
 if __name__ == "__main__":
-    benchmark.select_cpu_backend()
     test_trans_2d('float32', (32, 16), (1, 0), 'cpu')
diff --git a/third_party/tsingmicro/examples/time1.py b/third_party/tsingmicro/examples/time1.py
@@ -182,7 +182,6 @@ def bench_matmul(a, b):
 
 
 if __name__ == "__main__":
-    # benchmark.select_cpu_backend()
     M = 4096
     K = 4096
     N = 4096
diff --git a/third_party/tsingmicro/examples/time_zs_opt2.py b/third_party/tsingmicro/examples/time_zs_opt2.py
@@ -134,7 +134,6 @@ def bench_matmul(a, b):
 
 
 if __name__ == "__main__":
-    # benchmark.select_cpu_backend()
     M = 4096
     N = 4096
     K = 4096

Original file line number	Diff line number	Diff line change
`@@ -66,5 +66,4 @@ def bench_matmul(M, N, K):`
`66`	`66`
`67`	`67`
`68`	`68`	`if __name__ == "__main__":`
`69`		`- # benchmark.select_cpu_backend()`
`70`	`69`	`bench_matmul(16, 16, 16)`