diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/transforms/sme_transpose_interleave_8VL.hpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_gemm/transforms/sme_transpose_interleave_8VL.hpp | 90 |
1 files changed, 45 insertions, 45 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/transforms/sme_transpose_interleave_8VL.hpp b/src/core/NEON/kernels/arm_gemm/transforms/sme_transpose_interleave_8VL.hpp index c49c5ba433..98e8bb20a5 100644 --- a/src/core/NEON/kernels/arm_gemm/transforms/sme_transpose_interleave_8VL.hpp +++ b/src/core/NEON/kernels/arm_gemm/transforms/sme_transpose_interleave_8VL.hpp @@ -39,51 +39,51 @@ void sme_transpose_interleave_8VL(uint16_t *out, const uint16_t *in, size_t widt "blt 4f\n" "1:" // Main row loop: Head "mov x25, %x[in]\n" - "mov x24, %x[out]\n" - "add x23, x25, %x[in_stride]\n" + "add x24, x25, %x[in_stride]\n" + "add %x[in], x24, %x[in_stride]\n" + "mov x23, %x[out]\n" "sub %x[height], %x[height], #0x2\n" - "add %x[in], x23, %x[in_stride]\n" "mov x22, %x[width]\n" "2:" // Main row loop: Column loop "mov x21, x22\n" - "mov x20, x24\n" "whilelt p0.h, XZR, x21\n" + "ld1h { z31.h }, p0/Z, [x25]\n" "dech x21\n" "whilelt p6.h, XZR, x21\n" + "ld1h { z30.h }, p6/Z, [x25, #1, MUL VL]\n" "dech x21\n" - "ld1h { z31.h }, p0/Z, [x25]\n" "whilelt p5.h, XZR, x21\n" + "ld1h { z29.h }, p5/Z, [x25, #2, MUL VL]\n" "dech x21\n" - "ld1h { z30.h }, p6/Z, [x25, #1, MUL VL]\n" "whilelt p4.h, XZR, x21\n" + "ld1h { z28.h }, p4/Z, [x25, #3, MUL VL]\n" "dech x21\n" - "ld1h { z29.h }, p5/Z, [x25, #2, MUL VL]\n" "whilelt p3.h, XZR, x21\n" + "ld1h { z27.h }, p3/Z, [x25, #4, MUL VL]\n" "dech x21\n" - "ld1h { z28.h }, p4/Z, [x25, #3, MUL VL]\n" "whilelt p2.h, XZR, x21\n" + "ld1h { z26.h }, p2/Z, [x25, #5, MUL VL]\n" "dech x21\n" - "ld1h { z27.h }, p3/Z, [x25, #4, MUL VL]\n" "whilelt p1.h, XZR, x21\n" - "dech x21\n" - "ld1h { z26.h }, p2/Z, [x25, #5, MUL VL]\n" - "dech x22, ALL, MUL #8\n" "ld1h { z25.h }, p1/Z, [x25, #6, MUL VL]\n" - "add x24, x24, %x[out_stride]\n" - "ld1h { z24.h }, p0/Z, [x23]\n" + "dech x21\n" + "mov x20, x23\n" + "ld1h { z24.h }, p0/Z, [x24]\n" "whilelt p0.h, XZR, x21\n" - "cmp x22, #0x0\n" + "dech x22, ALL, MUL #8\n" "ld1h { z23.h }, p0/Z, [x25, #7, MUL VL]\n" + "ld1h { z22.h }, p6/Z, [x24, #1, MUL VL]\n" + "cmp x22, #0x0\n" "addvl x25, x25, #8\n" - "ld1h { z22.h }, p6/Z, [x23, #1, MUL VL]\n" - "ld1h { z21.h }, p5/Z, [x23, #2, MUL VL]\n" - "ld1h { z20.h }, p4/Z, [x23, #3, MUL VL]\n" - "ld1h { z19.h }, p3/Z, [x23, #4, MUL VL]\n" - "ld1h { z18.h }, p2/Z, [x23, #5, MUL VL]\n" - "ld1h { z17.h }, p1/Z, [x23, #6, MUL VL]\n" - "ld1h { z16.h }, p0/Z, [x23, #7, MUL VL]\n" + "ld1h { z21.h }, p5/Z, [x24, #2, MUL VL]\n" + "add x23, x23, %x[out_stride]\n" + "ld1h { z20.h }, p4/Z, [x24, #3, MUL VL]\n" + "ld1h { z19.h }, p3/Z, [x24, #4, MUL VL]\n" + "ld1h { z18.h }, p2/Z, [x24, #5, MUL VL]\n" + "ld1h { z17.h }, p1/Z, [x24, #6, MUL VL]\n" + "ld1h { z16.h }, p0/Z, [x24, #7, MUL VL]\n" "st1h { z31.h }, p7, [x20]\n" - "addvl x23, x23, #8\n" + "addvl x24, x24, #8\n" "st1h { z30.h }, p7, [x20, #1, MUL VL]\n" "st1h { z29.h }, p7, [x20, #2, MUL VL]\n" "st1h { z28.h }, p7, [x20, #3, MUL VL]\n" @@ -109,47 +109,47 @@ void sme_transpose_interleave_8VL(uint16_t *out, const uint16_t *in, size_t widt "4:" // Main loop skip "5:" // Tail row loop: Head "mov x25, %x[in]\n" - "mov x24, %x[out]\n" "add %x[in], x25, %x[in_stride]\n" + "mov x23, %x[out]\n" "sub %x[height], %x[height], #0x1\n" "mov x21, %x[width]\n" "6:" // Tail row loop: Column loop "mov x20, x21\n" - "dech x21, ALL, MUL #8\n" - "whilelt p1.h, XZR, x20\n" + "whilelt p0.h, XZR, x20\n" + "ld1h { z23.h }, p0/Z, [x25]\n" "dech x20\n" "whilelt p0.h, XZR, x20\n" + "ld1h { z22.h }, p0/Z, [x25, #1, MUL VL]\n" "dech x20\n" - "ld1h { z23.h }, p1/Z, [x25]\n" - "whilelt p1.h, XZR, x20\n" + "whilelt p0.h, XZR, x20\n" + "ld1h { z21.h }, p0/Z, [x25, #2, MUL VL]\n" "dech x20\n" - "ld1h { z22.h }, p0/Z, [x25, #1, MUL VL]\n" "whilelt p0.h, XZR, x20\n" + "ld1h { z20.h }, p0/Z, [x25, #3, MUL VL]\n" "dech x20\n" - "ld1h { z21.h }, p1/Z, [x25, #2, MUL VL]\n" - "whilelt p1.h, XZR, x20\n" + "whilelt p0.h, XZR, x20\n" + "ld1h { z19.h }, p0/Z, [x25, #4, MUL VL]\n" "dech x20\n" - "ld1h { z20.h }, p0/Z, [x25, #3, MUL VL]\n" "whilelt p0.h, XZR, x20\n" + "ld1h { z18.h }, p0/Z, [x25, #5, MUL VL]\n" "dech x20\n" - "ld1h { z19.h }, p1/Z, [x25, #4, MUL VL]\n" - "whilelt p1.h, XZR, x20\n" + "whilelt p0.h, XZR, x20\n" + "ld1h { z17.h }, p0/Z, [x25, #6, MUL VL]\n" "dech x20\n" - "ld1h { z18.h }, p0/Z, [x25, #5, MUL VL]\n" + "dech x21, ALL, MUL #8\n" "whilelt p0.h, XZR, x20\n" "cmp x21, #0x0\n" - "ld1h { z17.h }, p1/Z, [x25, #6, MUL VL]\n" "ld1h { z16.h }, p0/Z, [x25, #7, MUL VL]\n" + "st1h { z23.h }, p7, [x23]\n" "addvl x25, x25, #8\n" - "st1h { z23.h }, p7, [x24]\n" - "st1h { z22.h }, p7, [x24, #1, MUL VL]\n" - "st1h { z21.h }, p7, [x24, #2, MUL VL]\n" - "st1h { z20.h }, p7, [x24, #3, MUL VL]\n" - "st1h { z19.h }, p7, [x24, #4, MUL VL]\n" - "st1h { z18.h }, p7, [x24, #5, MUL VL]\n" - "st1h { z17.h }, p7, [x24, #6, MUL VL]\n" - "st1h { z16.h }, p7, [x24, #7, MUL VL]\n" - "add x24, x24, %x[out_stride]\n" + "st1h { z22.h }, p7, [x23, #1, MUL VL]\n" + "st1h { z21.h }, p7, [x23, #2, MUL VL]\n" + "st1h { z20.h }, p7, [x23, #3, MUL VL]\n" + "st1h { z19.h }, p7, [x23, #4, MUL VL]\n" + "st1h { z18.h }, p7, [x23, #5, MUL VL]\n" + "st1h { z17.h }, p7, [x23, #6, MUL VL]\n" + "st1h { z16.h }, p7, [x23, #7, MUL VL]\n" + "add x23, x23, %x[out_stride]\n" "bgt 6b\n" "7:" // Tail row loop: Column loop skip "cmp %x[height], #0x1\n" |