From 9262dcbb0a442cb7818dd77205eb6a72a4ebb794 Mon Sep 17 00:00:00 2001 From: zhentaoyu Date: Tue, 20 Aug 2024 17:14:13 +0800 Subject: [PATCH] test: comment new test_cases for only local testing Signed-off-by: zhentaoyu --- tests/test-backend-ops.cpp | 20 +++++++++++--------- 1 file changed, 11 insertions(+), 9 deletions(-) diff --git a/tests/test-backend-ops.cpp b/tests/test-backend-ops.cpp index 040a1f20d..e2b720844 100644 --- a/tests/test-backend-ops.cpp +++ b/tests/test-backend-ops.cpp @@ -2145,11 +2145,12 @@ static bool test_backend(ggml_backend_t backend, test_mode mode, const char * op test_cases.emplace_back(new test_im2col(GGML_TYPE_F32, GGML_TYPE_F16, GGML_TYPE_F16, {3000, 128, 1, 1}, {3, 128, 1280, 1}, 1, 0, 1, 0, 1, 0, false)); test_cases.emplace_back(new test_im2col(GGML_TYPE_F32, GGML_TYPE_F16, GGML_TYPE_F32, {3000, 128, 1, 1}, {3, 128, 1280, 1}, 1, 0, 1, 0, 1, 0, false)); -#ifdef GGML_USE_SYCL - // test cases for 2D im2col with large input H and W - test_cases.emplace_back(new test_im2col(GGML_TYPE_F32, GGML_TYPE_F16, GGML_TYPE_F16, {1024, 1024, 256, 1}, {3, 3, 256, 1}, 1, 1, 1, 1, 1, 1, true)); - test_cases.emplace_back(new test_im2col(GGML_TYPE_F32, GGML_TYPE_F16, GGML_TYPE_F32, {1024, 1024, 256, 1}, {3, 3, 256, 1}, 1, 1, 1, 1, 1, 1, true)); -#endif + // sycl backend will limit task global_range < MAX_INT + // test cases for 2D im2col with large input W and H (occurs in stable-diffusion) + // however these cases need to alloc more memory which may fail in some devices (Intel Arc770, etc.) + // these cases are verified (pass) in Intel(R) Data Center GPU Max 1100 (sycl backend) and NV A30 (cuda backend) + // test_cases.emplace_back(new test_im2col(GGML_TYPE_F32, GGML_TYPE_F16, GGML_TYPE_F16, {1024, 1024, 256, 1}, {3, 3, 256, 1}, 1, 1, 1, 1, 1, 1, true)); + // test_cases.emplace_back(new test_im2col(GGML_TYPE_F32, GGML_TYPE_F16, GGML_TYPE_F32, {1024, 1024, 256, 1}, {3, 3, 256, 1}, 1, 1, 1, 1, 1, 1, true)); test_cases.emplace_back(new test_conv_transpose_1d()); test_cases.emplace_back(new test_conv_transpose_1d({3,2,1,1}, {2,3,2,1}, 3, 0, 1)); @@ -2293,10 +2294,11 @@ static bool test_backend(ggml_backend_t backend, test_mode mode, const char * op test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F16, GGML_TYPE_F32, 64, 45, 128, { 8, 1}, {4, 1})); test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F16, GGML_TYPE_F32, 128, 45, 64, { 8, 1}, {4, 1})); -#ifdef GGML_USE_SYCL - // test case for fp16-convert-to-fp32 kernel with large k under fp32 compute dtype - test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F16, GGML_TYPE_F16, 512, 262144, 9216, {1, 1}, {1, 1})); -#endif + // sycl backend will limit task global_range < MAX_INT + // test case for f16-type-convert-to-fp32 kernel with large k under fp32 compute dtype (occurs in stable-diffusion) + // however this case needs to alloc more memory which may fail in some devices (Intel Arc770, etc.) + // this case is verified (pass) in Intel(R) Data Center GPU Max 1100 (sycl backend) and NV A30 (cuda backend) + // test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F16, GGML_TYPE_F16, 512, 262144, 9216, {1, 1}, {1, 1})); for (ggml_type type_a : base_types) { for (ggml_type type_b : {GGML_TYPE_F32 /*, GGML_TYPE_F16 */}) {