1be168c0dSopenharmony_ciFrom 48ef088b1f2d594b4d756c19574ead05492d3f84 Mon Sep 17 00:00:00 2001 2be168c0dSopenharmony_ciFrom: zhangyanhui <zhangyanhui17@huawei.com> 3be168c0dSopenharmony_ciDate: Mon, 8 Jul 2024 18:27:42 +0800 4be168c0dSopenharmony_ciSubject: [PATCH] bugfix for depthwise when kernel size is longer than input_w 5be168c0dSopenharmony_ci 6be168c0dSopenharmony_ci--- 7be168c0dSopenharmony_ci .../device/cpu/kernel/nnacl/assembly/arm32/ConvDwFp32Row.S | 2 +- 8be168c0dSopenharmony_ci .../device/cpu/kernel/nnacl/assembly/arm64/ConvDwFp32Row.S | 2 +- 9be168c0dSopenharmony_ci 2 files changed, 2 insertions(+), 2 deletions(-) 10be168c0dSopenharmony_ci 11be168c0dSopenharmony_cidiff --git a/mindspore/ccsrc/plugin/device/cpu/kernel/nnacl/assembly/arm32/ConvDwFp32Row.S b/mindspore/ccsrc/plugin/device/cpu/kernel/nnacl/assembly/arm32/ConvDwFp32Row.S 12be168c0dSopenharmony_ciindex 8d1d819d..19750920 100644 13be168c0dSopenharmony_ci--- a/mindspore/ccsrc/plugin/device/cpu/kernel/nnacl/assembly/arm32/ConvDwFp32Row.S 14be168c0dSopenharmony_ci+++ b/mindspore/ccsrc/plugin/device/cpu/kernel/nnacl/assembly/arm32/ConvDwFp32Row.S 15be168c0dSopenharmony_ci@@ -35,7 +35,7 @@ asm_function ConvDwFp32Row 16be168c0dSopenharmony_ci mov r6, #4 17be168c0dSopenharmony_ci mul r5, r5, r6 18be168c0dSopenharmony_ci cmp r3, #0 19be168c0dSopenharmony_ci- beq End 20be168c0dSopenharmony_ci+ ble End 21be168c0dSopenharmony_ci 22be168c0dSopenharmony_ci LoopNumPixel: 23be168c0dSopenharmony_ci mov r6, r1 // input_ptr 24be168c0dSopenharmony_cidiff --git a/mindspore/ccsrc/plugin/device/cpu/kernel/nnacl/assembly/arm64/ConvDwFp32Row.S b/mindspore/ccsrc/plugin/device/cpu/kernel/nnacl/assembly/arm64/ConvDwFp32Row.S 25be168c0dSopenharmony_ciindex 417aa4ce..e50e2ee5 100644 26be168c0dSopenharmony_ci--- a/mindspore/ccsrc/plugin/device/cpu/kernel/nnacl/assembly/arm64/ConvDwFp32Row.S 27be168c0dSopenharmony_ci+++ b/mindspore/ccsrc/plugin/device/cpu/kernel/nnacl/assembly/arm64/ConvDwFp32Row.S 28be168c0dSopenharmony_ci@@ -30,7 +30,7 @@ asm_function ConvDwFp32Row 29be168c0dSopenharmony_ci // x19 ~ x29 should be also preserved 30be168c0dSopenharmony_ci // whereas our coding style do not permit such amount of parameters 31be168c0dSopenharmony_ci cmp x3, #0 32be168c0dSopenharmony_ci-beq End 33be168c0dSopenharmony_ci+ble End 34be168c0dSopenharmony_ci 35be168c0dSopenharmony_ci mov x9, x0 36be168c0dSopenharmony_ci mov x12, #4 37be168c0dSopenharmony_ci-- 38be168c0dSopenharmony_ci2.25.1 39be168c0dSopenharmony_ci 40