diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/transforms/sme_transpose_interleave_4VL_1x4.hpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_gemm/transforms/sme_transpose_interleave_4VL_1x4.hpp | 58 |
1 files changed, 29 insertions, 29 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/transforms/sme_transpose_interleave_4VL_1x4.hpp b/src/core/NEON/kernels/arm_gemm/transforms/sme_transpose_interleave_4VL_1x4.hpp index c7d4882b42..94d1c0840a 100644 --- a/src/core/NEON/kernels/arm_gemm/transforms/sme_transpose_interleave_4VL_1x4.hpp +++ b/src/core/NEON/kernels/arm_gemm/transforms/sme_transpose_interleave_4VL_1x4.hpp @@ -43,43 +43,43 @@ void sme_transpose_interleave_4VL_1x4(uint8_t *out, const uint8_t *in, size_t wi "ptrue p1.b\n" "1:" // Main row loop: Head "mov x25, %x[in]\n" - "cmp %x[height], #0x3\n" "add x24, x25, %x[in_stride]\n" - "mov x23, %x[out]\n" - "add x22, x24, %x[in_stride]\n" - "mov x21, %x[width]\n" - "add x20, x22, %x[in_stride]\n" - "csel x22, x22, %x[pad_row], GE\n" - "add %x[in], x20, %x[in_stride]\n" - "csel x20, x20, %x[pad_row], GT\n" + "add x23, x24, %x[in_stride]\n" + "add x22, x23, %x[in_stride]\n" + "cmp %x[height], #0x3\n" + "add %x[in], x22, %x[in_stride]\n" + "csel x22, x22, %x[pad_row], GT\n" + "csel x23, x23, %x[pad_row], GE\n" "cmp %x[height], #0x1\n" - "sub %x[height], %x[height], #0x4\n" + "mov x21, %x[out]\n" "csel x24, x24, %x[pad_row], GT\n" + "sub %x[height], %x[height], #0x4\n" + "mov x20, %x[width]\n" "2:" // Main row loop: Column loop - "whilelt p0.b, XZR, x21\n" - "decw x21, ALL, MUL #4\n" - "ld1b { z20.b }, p0/Z, [x25]\n" - "cmp x21, #0x0\n" - "addvl x25, x25, #1\n" + "whilelt p0.b, XZR, x20\n" + "ld1b { z17.b }, p0/Z, [x25]\n" + "decw x20, ALL, MUL #4\n" "ld1b { z19.b }, p0/Z, [x24]\n" + "cmp x20, #0x0\n" + "addvl x25, x25, #1\n" + "ld1b { z16.b }, p0/Z, [x23]\n" + "zip1 z18.b, z17.b, z16.b\n" + "zip2 z20.b, z17.b, z16.b\n" "addvl x24, x24, #1\n" - "ld1b { z17.b }, p0/Z, [x22]\n" - "addvl x22, x22, #1\n" - "ld1b { z16.b }, p0/Z, [x20]\n" - "addvl x20, x20, #1\n" - "zip1 z18.b, z20.b, z17.b\n" - "zip2 z20.b, z20.b, z17.b\n" + "ld1b { z16.b }, p0/Z, [x22]\n" "zip1 z17.b, z19.b, z16.b\n" - "zip2 z16.b, z19.b, z16.b\n" - "zip1 z19.b, z18.b, z17.b\n" + "zip2 z19.b, z19.b, z16.b\n" + "addvl x23, x23, #1\n" + "addvl x22, x22, #1\n" + "zip1 z16.b, z18.b, z17.b\n" "zip2 z18.b, z18.b, z17.b\n" - "zip1 z17.b, z20.b, z16.b\n" - "zip2 z16.b, z20.b, z16.b\n" - "st1b { z19.b }, p1, [x23]\n" - "st1b { z18.b }, p1, [x23, #1, MUL VL]\n" - "st1b { z17.b }, p1, [x23, #2, MUL VL]\n" - "st1b { z16.b }, p1, [x23, #3, MUL VL]\n" - "add x23, x23, %x[out_stride]\n" + "st1b { z16.b }, p1, [x21]\n" + "zip1 z17.b, z20.b, z19.b\n" + "zip2 z16.b, z20.b, z19.b\n" + "st1b { z18.b }, p1, [x21, #1, MUL VL]\n" + "st1b { z17.b }, p1, [x21, #2, MUL VL]\n" + "st1b { z16.b }, p1, [x21, #3, MUL VL]\n" + "add x21, x21, %x[out_stride]\n" "bgt 2b\n" "3:" // Main row loop: Column loop skip "cmp %x[height], #0x1\n" |