diff --git a/mindspore/lite/nnacl/assembly/arm32/MatmulFp32Opt12x4.S b/mindspore/lite/nnacl/assembly/arm32/MatmulFp32Opt12x4.S index 5ed821be34..03f0832923 100644 --- a/mindspore/lite/nnacl/assembly/arm32/MatmulFp32Opt12x4.S +++ b/mindspore/lite/nnacl/assembly/arm32/MatmulFp32Opt12x4.S @@ -22,7 +22,8 @@ MatmulFloatNeon32Opt12x4: // r4-r8 and q4-q7 must be saved according to https://static.docs.arm.com/ihi0042/i/aapcs32.pdf push {r0-r8, r10, r11, lr} - add sp, sp, #48 + vpush {q4-q7} + add sp, sp, #112 ldr r5, [sp, #4] ldr r6, [sp, #8] @@ -400,6 +401,7 @@ LoopRow: b LoopRow LoopRowEnd: - sub sp, sp, #48 + sub sp, sp, #112 + vpop {q4-q7} pop {r0-r8, r10, r11, pc} #endif