diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/transforms/sme_transpose_interleave_1VL_1x4.hpp')
-rw-r--r-- | src/core/NEON/kernels/arm_gemm/transforms/sme_transpose_interleave_1VL_1x4.hpp | 76 |
1 files changed, 38 insertions, 38 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/transforms/sme_transpose_interleave_1VL_1x4.hpp b/src/core/NEON/kernels/arm_gemm/transforms/sme_transpose_interleave_1VL_1x4.hpp index ff8cfc7efe..d67e353f18 100644 --- a/src/core/NEON/kernels/arm_gemm/transforms/sme_transpose_interleave_1VL_1x4.hpp +++ b/src/core/NEON/kernels/arm_gemm/transforms/sme_transpose_interleave_1VL_1x4.hpp @@ -43,65 +43,65 @@ void sme_transpose_interleave_1VL_1x4(uint8_t *out, const uint8_t *in, size_t wi "ptrue p1.b\n" "1:" // Main row loop: Head "mov x26, %x[in]\n" - "cmp %x[height], #0x3\n" "add x25, x26, %x[in_stride]\n" - "mov x24, %x[width]\n" - "add x23, x25, %x[in_stride]\n" - "cntb x22\n" - "add x21, x23, %x[in_stride]\n" - "csel x23, x23, %x[pad_row], GE\n" - "add %x[in], x21, %x[in_stride]\n" - "csel x21, x21, %x[pad_row], GT\n" + "add x24, x25, %x[in_stride]\n" + "add x23, x24, %x[in_stride]\n" + "cmp %x[height], #0x3\n" + "add %x[in], x23, %x[in_stride]\n" + "csel x23, x23, %x[pad_row], GT\n" + "csel x24, x24, %x[pad_row], GE\n" "cmp %x[height], #0x1\n" - "mov x20, %x[out]\n" + "mov x22, %x[width]\n" + "cntb x21\n" "csel x25, x25, %x[pad_row], GT\n" - "cmp x24, x22\n" + "cmp x22, x21\n" + "mov x20, %x[out]\n" "sub %x[height], %x[height], #0x4\n" "blt 3f\n" "2:" // Main row loop: Unroll column loop - "ld1b { z20.b }, p1/Z, [x26]\n" - "sub x24, x24, x22\n" + "ld1b { z17.b }, p1/Z, [x26]\n" + "sub x22, x22, x21\n" + "cmp x22, x21\n" + "ld1b { z18.b }, p1/Z, [x25]\n" "addvl x26, x26, #1\n" - "ld1b { z19.b }, p1/Z, [x25]\n" - "cmp x24, x22\n" "addvl x25, x25, #1\n" - "ld1b { z17.b }, p1/Z, [x23]\n" + "ld1b { z16.b }, p1/Z, [x24]\n" + "zip1 z20.b, z17.b, z16.b\n" + "zip2 z19.b, z17.b, z16.b\n" + "addvl x24, x24, #1\n" + "ld1b { z16.b }, p1/Z, [x23]\n" + "zip1 z17.b, z18.b, z16.b\n" + "zip2 z18.b, z18.b, z16.b\n" "addvl x23, x23, #1\n" - "ld1b { z16.b }, p1/Z, [x21]\n" - "addvl x21, x21, #1\n" - "zip1 z18.b, z20.b, z17.b\n" - "zip2 z20.b, z20.b, z17.b\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 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, [x20]\n" + "zip1 z16.b, z20.b, z17.b\n" + "st1b { z16.b }, p1, [x20]\n" "add x20, x20, %x[out_stride]\n" - "st1b { z18.b }, p1, [x20]\n" + "zip2 z16.b, z20.b, z17.b\n" + "st1b { z16.b }, p1, [x20]\n" "add x20, x20, %x[out_stride]\n" + "zip1 z17.b, z19.b, z18.b\n" + "zip2 z16.b, z19.b, z18.b\n" "st1b { z17.b }, p1, [x20]\n" "add x20, x20, %x[out_stride]\n" "st1b { z16.b }, p1, [x20]\n" "add x20, x20, %x[out_stride]\n" "bge 2b\n" "3:" // Main row loop: Unroll column loop skip - "cbz x24, 5f\n" + "cbz x22, 5f\n" "4:" // Main row loop: Column loop - "whilelt p0.b, XZR, x24\n" - "decw x24\n" - "ld1b { z19.b }, p0/Z, [x26]\n" - "cmp x24, #0x0\n" - "incd x26, ALL, MUL #2\n" + "whilelt p0.b, XZR, x22\n" + "ld1b { z17.b }, p0/Z, [x26]\n" + "decw x22\n" "ld1b { z18.b }, p0/Z, [x25]\n" + "cmp x22, #0x0\n" + "incd x26, ALL, MUL #2\n" + "ld1b { z16.b }, p0/Z, [x24]\n" + "zip1 z17.b, z17.b, z16.b\n" "incd x25, ALL, MUL #2\n" - "ld1b { z17.b }, p0/Z, [x23]\n" - "incd x23, ALL, MUL #2\n" - "ld1b { z16.b }, p0/Z, [x21]\n" - "incd x21, ALL, MUL #2\n" - "zip1 z17.b, z19.b, z17.b\n" + "incd x24, ALL, MUL #2\n" + "ld1b { z16.b }, p0/Z, [x23]\n" "zip1 z16.b, z18.b, z16.b\n" + "incd x23, ALL, MUL #2\n" "zip1 z16.b, z17.b, z16.b\n" "st1b { z16.b }, p1, [x20]\n" "add x20, x20, %x[out_stride]\n" |