From b361b3025ff8d0feecdcab06e75e7857f63316eb Mon Sep 17 00:00:00 2001 From: Liu Liu Date: Fri, 18 Oct 2024 19:52:16 -0400 Subject: [PATCH] Revert previous change, also disable flashattn in the unit tests for now. --- .github/workflows/analyze.yaml | 2 +- .github/workflows/coverage.yaml | 2 +- .github/workflows/cuda-int-tests.yaml | 2 +- .github/workflows/ubsan-unit-tests.yaml | 2 +- .../ccv_cnnp_model_gradient_checkpointing.c | 25 ------------------- 5 files changed, 4 insertions(+), 29 deletions(-) diff --git a/.github/workflows/analyze.yaml b/.github/workflows/analyze.yaml index 48799df3c..ccb2fc2f8 100644 --- a/.github/workflows/analyze.yaml +++ b/.github/workflows/analyze.yaml @@ -12,7 +12,7 @@ jobs: fetch-depth: 0 - name: configure run: | - cd lib && ./configure --enable-sm80 --disable-openmp && cd .. + cd lib && ./configure --disable-openmp && cd .. - name: analyze run: | cd lib && scan-build -o ../_analyze --use-cc clang make -j 64 diff --git a/.github/workflows/coverage.yaml b/.github/workflows/coverage.yaml index f6c8d8014..e4987c81f 100644 --- a/.github/workflows/coverage.yaml +++ b/.github/workflows/coverage.yaml @@ -12,7 +12,7 @@ jobs: fetch-depth: 0 - name: configure & build run: | - cd lib && ./configure --enable-sm80 && cd .. + cd lib && ./configure && cd .. make -j 64 -C test COVER=1 all.tests - name: tests run: | diff --git a/.github/workflows/cuda-int-tests.yaml b/.github/workflows/cuda-int-tests.yaml index 83c9d34f1..fa3d81c62 100644 --- a/.github/workflows/cuda-int-tests.yaml +++ b/.github/workflows/cuda-int-tests.yaml @@ -12,7 +12,7 @@ jobs: fetch-depth: 0 - name: configure & build run: | - cd lib && ./configure --enable-sm80 && cd .. + cd lib && ./configure && cd .. make -j 64 -C test/int/nnc - name: integration tests run: | diff --git a/.github/workflows/ubsan-unit-tests.yaml b/.github/workflows/ubsan-unit-tests.yaml index 0f4bccb5f..58d3898af 100644 --- a/.github/workflows/ubsan-unit-tests.yaml +++ b/.github/workflows/ubsan-unit-tests.yaml @@ -12,7 +12,7 @@ jobs: fetch-depth: 0 - name: configure & build run: | - cd lib && ./configure --enable-sm80 && cd .. + cd lib && ./configure && cd .. make -j 64 -C lib ubsan make -j 64 -C bin ubsan make -j 64 -C bin/nnc ubsan diff --git a/lib/nnc/ccv_cnnp_model_gradient_checkpointing.c b/lib/nnc/ccv_cnnp_model_gradient_checkpointing.c index 89977886b..20351e998 100644 --- a/lib/nnc/ccv_cnnp_model_gradient_checkpointing.c +++ b/lib/nnc/ccv_cnnp_model_gradient_checkpointing.c @@ -201,31 +201,6 @@ void ccv_cnnp_model_apply_gradient_checkpoints(ccv_cnnp_compiled_data_t* const c } if (input_execs->rnum <= 0 || output_execs->rnum <= 0) continue; - ccv_nnc_graph_visit_t* const reverse_visit = ccv_nnc_graph_visit_new(graph, reversed_nodes, exec_rnum, (ccv_nnc_graph_exec_symbol_t*)ccv_array_get(output_execs, 0), output_execs->rnum, (ccv_nnc_graph_exec_symbol_t*)ccv_array_get(input_execs, 0), input_execs->rnum, 1); - ccv_nnc_graph_visit_for(reverse_visit, exec_info, node, idx) { - if (idx < exec_rnum && !CCV_NNC_GRAPH_EXEC_IS_DEAD(node->flags)) - maskbit[idx >> 5] |= (1u << (idx & 0x1f)); - } ccv_nnc_graph_visit_endfor - // Check if any of the items in input_execs is not marked, if it is not, there is no connection from it to output_execs, no need to visit. - for (j = 0; j < input_execs->rnum;) - { - const int idx = ((ccv_nnc_tensor_symbol_t*)ccv_array_get(input_execs, j))->d; - if (maskbit[idx >> 5] & (1u << (idx & 0x1f))) - { - ++j; - continue; - } - // Not marked, remove this one. - if (j < input_execs->rnum - 1) - *(ccv_nnc_tensor_symbol_t*)ccv_array_get(input_execs, j) = *(ccv_nnc_tensor_symbol_t*)ccv_array_get(input_execs, input_execs->rnum - 1); - --input_execs->rnum; - } - // Reset maskbit back. - ccv_nnc_graph_visit_for(reverse_visit, exec_info, node, idx) { - if (idx < exec_rnum && !CCV_NNC_GRAPH_EXEC_IS_DEAD(node->flags)) - maskbit[idx >> 5] &= ~(1u << (idx & 0x1f)); - } ccv_nnc_graph_visit_endfor - ccv_nnc_graph_visit_free(reverse_visit); // Fill in blanks (i.e. the backward ops that are not showing in above, but should be included to avoid excluding necessary ones). This is done by flowing gradients from outputs back all the way to inputs. ccv_array_clear(input_gradient_execs); ccv_array_clear(output_gradient_execs);