diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/kernels/sve_interleaved_u8u32_dot_8x3VL/generic.cpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_gemm/kernels/sve_interleaved_u8u32_dot_8x3VL/generic.cpp | 20 |
1 files changed, 10 insertions, 10 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/sve_interleaved_u8u32_dot_8x3VL/generic.cpp b/src/core/NEON/kernels/arm_gemm/kernels/sve_interleaved_u8u32_dot_8x3VL/generic.cpp index dc31a73c13..eaa3ad2428 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/sve_interleaved_u8u32_dot_8x3VL/generic.cpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/sve_interleaved_u8u32_dot_8x3VL/generic.cpp @@ -55,25 +55,25 @@ void sve_interleaved_u8u32_dot_8x3VL( "2:" // Width loop "ldr x20, [%x[args_ptr], %[offsetof_K]]\n" "mov %x[Apanel], x21\n" + "cmp x20, #0x2\n" "mov z8.s, #0x0\n" "mov z9.s, #0x0\n" + "ld1rqb { z0.b }, p0/Z, [%x[Apanel]]\n" "mov z10.s, #0x0\n" - "ld1b { z4.b }, p0/Z, [x22]\n" "mov z11.s, #0x0\n" + "ld1rqb { z1.b }, p0/Z, [%x[Apanel], #16]\n" "mov z12.s, #0x0\n" - "ld1b { z5.b }, p0/Z, [x22, #1, MUL VL]\n" - "cmp x20, #0x2\n" "mov z13.s, #0x0\n" + "ld1b { z4.b }, p0/Z, [x22]\n" "mov z14.s, #0x0\n" "mov z15.s, #0x0\n" + "ld1b { z5.b }, p0/Z, [x22, #1, MUL VL]\n" "mov z16.s, #0x0\n" - "ld1rqb { z0.b }, p0/Z, [%x[Apanel]]\n" "mov z17.s, #0x0\n" + "ld1b { z6.b }, p0/Z, [x22, #2, MUL VL]\n" "mov z18.s, #0x0\n" - "ld1rqb { z1.b }, p0/Z, [%x[Apanel], #16]\n" "mov z19.s, #0x0\n" "mov z20.s, #0x0\n" - "ld1b { z6.b }, p0/Z, [x22, #2, MUL VL]\n" "mov z21.s, #0x0\n" "mov z22.s, #0x0\n" "mov z23.s, #0x0\n" @@ -151,12 +151,12 @@ void sve_interleaved_u8u32_dot_8x3VL( "ld1b { z6.b }, p0/Z, [x22, #2, MUL VL]\n" "bge 3b\n" "4:" // main loop skip - "add %x[Apanel], %x[Apanel], #0x20\n" - "addvl x22, x22, #3\n" "udot z8.s, z4.b, z0.b[0]\n" "udot z11.s, z4.b, z0.b[1]\n" + "add %x[Apanel], %x[Apanel], #0x20\n" "udot z14.s, z4.b, z0.b[2]\n" "udot z17.s, z4.b, z0.b[3]\n" + "addvl x22, x22, #3\n" "udot z20.s, z4.b, z1.b[0]\n" "udot z23.s, z4.b, z1.b[1]\n" "udot z26.s, z4.b, z1.b[2]\n" @@ -183,13 +183,13 @@ void sve_interleaved_u8u32_dot_8x3VL( "add %x[Apanel], %x[Apanel], #0x20\n" "ld1b { z2.b }, p0/Z, [x22]\n" "ld1b { z1.b }, p0/Z, [x22, #1, MUL VL]\n" - "ld1b { z0.b }, p0/Z, [x22, #2, MUL VL]\n" - "addvl x22, x22, #3\n" "udot z8.s, z2.b, z4.b[0]\n" + "ld1b { z0.b }, p0/Z, [x22, #2, MUL VL]\n" "udot z11.s, z2.b, z4.b[1]\n" "udot z14.s, z2.b, z4.b[2]\n" "udot z17.s, z2.b, z4.b[3]\n" "udot z20.s, z2.b, z3.b[0]\n" + "addvl x22, x22, #3\n" "udot z23.s, z2.b, z3.b[1]\n" "udot z26.s, z2.b, z3.b[2]\n" "udot z29.s, z2.b, z3.b[3]\n" |