diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block2_bf16_bf16.hpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block2_bf16_bf16.hpp | 176 |
1 files changed, 87 insertions, 89 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block2_bf16_bf16.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block2_bf16_bf16.hpp index f648ccf771..d9189258c1 100644 --- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block2_bf16_bf16.hpp +++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave2VL_block2_bf16_bf16.hpp @@ -22,32 +22,30 @@ * SOFTWARE. */ -#if defined(__ARM_FEATURE_SVE) +#if defined(ARM_COMPUTE_ENABLE_SME) template <> void interleave_block<2, 2, VLType::SME, false>( bfloat16 * &out, const bfloat16 * const *in, - size_t width, size_t height, size_t row_offset, bool first + size_t width, size_t height, size_t row_offset, bool ) { - ARM_COMPUTE_UNUSED(first); - __asm__ __volatile__( ".inst 0xd503477f // SMSTART ZA\n" - "cnth x21\n" - "mov x22, %x[width]\n" - "inch x22\n" + "cnth x22\n" + "mov x21, %x[width]\n" + "inch x21\n" "mov x20, %x[width]\n" - "sub x17, x21, #0x1\n" - "sub x22, x22, #0x1\n" + "sub x17, x22, #0x1\n" + "sub x21, x21, #0x1\n" "ands x17, x20, x17\n" "cntw x16\n" - "udiv x22, x22, x21\n" // n_passes = ceildiv(width, VL<T>) - "csel x17, x17, x21, NE\n" - "sub x13, x22, #0x1\n" + "udiv x21, x21, x22\n" // n_passes = ceildiv(width, VL<T>) + "csel x17, x17, x22, NE\n" + "sub x13, x21, #0x1\n" "add x17, x17, #0x1\n" "sub x15, x16, #0x2\n" - "lsl x21, %x[height], #0x1\n" // height * 2 + "lsl x22, %x[height], #0x1\n" // height * 2 "lsl x20, x16, #0x1\n" "mov x14, #0x0\n" "mov x11, %x[in]\n" @@ -57,15 +55,15 @@ void interleave_block<2, 2, VLType::SME, false>( "cntw x27, ALL, MUL #3\n" "ldr x26, [x10, #0x0]\n" "lsr x13, x13, #0x1\n" // n_loops = (n_passes - 1) / 2 - "and x25, x22, #0x1\n" // odd_tail = bool(n_passes & 0x1) + "and x25, x21, #0x1\n" // odd_tail = bool(n_passes & 0x1) "ldr x24, [x11, #0x8]\n" "lsr x17, x17, #0x1\n" "ptrue p13.s\n" - "ldr x23, [x10, #0x8]\n" - "whilelt p12.h, XZR, x21\n" - "whilelt p11.h, x20, x21\n" - "mov x22, %x[row_offset]\n" - "mov x21, %x[out]\n" + "ldr x21, [x10, #0x8]\n" + "whilelt p12.h, XZR, x22\n" + "whilelt p11.h, x20, x22\n" + "mov x23, %x[row_offset]\n" + "mov x22, %x[out]\n" "whilelt p10.h, x14, %x[width]\n" "whilelt p9.h, x14, %x[width]\n" "whilelt p8.h, x14, %x[width]\n" @@ -76,39 +74,39 @@ void interleave_block<2, 2, VLType::SME, false>( "1:" // K loop: Charge: Loop ".inst 0x25286581 // psel p1.h, p9.h/Z, p12.h[w12]\n" ".inst 0x25286160 // psel p0.h, p8.h/Z, p11.h[w12]\n" - ".inst 0xe0560520 // ld1h { za0h.h[x12] }, p1/Z, [x9, x22, LSL #1]\n" + ".inst 0xe0570520 // ld1h { za0h.h[x12] }, p1/Z, [x9, x23, LSL #1]\n" "ldr x9, [x11, #0x0]\n" - ".inst 0xe0560348 // ld1h { za1h.h[x12] }, p0/Z, [x26, x22, LSL #1]\n" + ".inst 0xe0570348 // ld1h { za1h.h[x12] }, p0/Z, [x26, x23, LSL #1]\n" ".inst 0x25686581 // psel p1.h, p9.h/Z, p12.h[w12, #2]\n" ".inst 0x25686160 // psel p0.h, p8.h/Z, p11.h[w12, #2]\n" "ldr x26, [x10, #0x0]\n" - ".inst 0xe0560702 // ld1h { za0h.h[x12, #2] }, p1/Z, [x24, x22, LSL #1]\n" + ".inst 0xe0570702 // ld1h { za0h.h[x12, #2] }, p1/Z, [x24, x23, LSL #1]\n" "ldr x24, [x11, #0x8]\n" "add x11, x11, #0x10\n" - ".inst 0xe05602ea // ld1h { za1h.h[x12, #2] }, p0/Z, [x23, x22, LSL #1]\n" + ".inst 0xe05702aa // ld1h { za1h.h[x12, #2] }, p0/Z, [x21, x23, LSL #1]\n" "add x12, x12, #0x4\n" "cmp x12, x15, LSL #1\n" - "ldr x23, [x10, #0x8]\n" + "ldr x21, [x10, #0x8]\n" "add x10, x10, #0x10\n" "blt 1b\n" "2:" // K loop: Charge: End ".inst 0x25286581 // psel p1.h, p9.h/Z, p12.h[w12]\n" ".inst 0x25286160 // psel p0.h, p8.h/Z, p11.h[w12]\n" - ".inst 0xe0560520 // ld1h { za0h.h[x12] }, p1/Z, [x9, x22, LSL #1]\n" - ".inst 0xe0560348 // ld1h { za1h.h[x12] }, p0/Z, [x26, x22, LSL #1]\n" + ".inst 0xe0570520 // ld1h { za0h.h[x12] }, p1/Z, [x9, x23, LSL #1]\n" + ".inst 0xe0570348 // ld1h { za1h.h[x12] }, p0/Z, [x26, x23, LSL #1]\n" ".inst 0x25686581 // psel p1.h, p9.h/Z, p12.h[w12, #2]\n" ".inst 0x25686160 // psel p0.h, p8.h/Z, p11.h[w12, #2]\n" "mov x11, %x[in]\n" "add x10, %x[in], x16, LSL #3\n" - ".inst 0xe0560702 // ld1h { za0h.h[x12, #2] }, p1/Z, [x24, x22, LSL #1]\n" + ".inst 0xe0570702 // ld1h { za0h.h[x12, #2] }, p1/Z, [x24, x23, LSL #1]\n" "ldr x9, [x11, #0x0]\n" - ".inst 0xe05602ea // ld1h { za1h.h[x12, #2] }, p0/Z, [x23, x22, LSL #1]\n" + ".inst 0xe05702aa // ld1h { za1h.h[x12, #2] }, p0/Z, [x21, x23, LSL #1]\n" "ldr x26, [x10, #0x0]\n" - "inch x22\n" + "inch x23\n" "inch x14\n" "ldr x24, [x11, #0x8]\n" "add x11, x11, #0x10\n" - "ldr x23, [x10, #0x8]\n" + "ldr x21, [x10, #0x8]\n" "add x10, x10, #0x10\n" "cbz x13, 8f\n" "mov x20, x13\n" @@ -121,60 +119,60 @@ void interleave_block<2, 2, VLType::SME, false>( "4:" // K loop: Main loop: First: Loop ".inst 0x25396581 // psel p1.h, p9.h/Z, p12.h[w13, #1]\n" ".inst 0x25396160 // psel p0.h, p8.h/Z, p11.h[w13, #1]\n" - ".inst 0xe0562521 // ld1h { za0h.h[x13, #1] }, p1/Z, [x9, x22, LSL #1]\n" + ".inst 0xe0572521 // ld1h { za0h.h[x13, #1] }, p1/Z, [x9, x23, LSL #1]\n" "ldr x9, [x11, #0x0]\n" - ".inst 0xe0562349 // ld1h { za1h.h[x13, #1] }, p0/Z, [x26, x22, LSL #1]\n" + ".inst 0xe0572349 // ld1h { za1h.h[x13, #1] }, p0/Z, [x26, x23, LSL #1]\n" ".inst 0x25796580 // psel p0.h, p9.h/Z, p12.h[w13, #3]\n" ".inst 0x25796162 // psel p2.h, p8.h/Z, p11.h[w13, #3]\n" "ldr x26, [x10, #0x0]\n" ".inst 0x25307541 // psel p1.s, p13.s/Z, p10.s[w12]\n" - ".inst 0xe0562303 // ld1h { za0h.h[x13, #3] }, p0/Z, [x24, x22, LSL #1]\n" + ".inst 0xe0572303 // ld1h { za0h.h[x13, #3] }, p0/Z, [x24, x23, LSL #1]\n" "ldr x24, [x11, #0x8]\n" ".inst 0x25307540 // psel p0.s, p13.s/Z, p10.s[w12]\n" - ".inst 0xe0562aeb // ld1h { za1h.h[x13, #3] }, p2/Z, [x23, x22, LSL #1]\n" - "ldr x23, [x10, #0x8]\n" - ".inst 0xe0bf86a0 // st1w { za0v.s[x12] }, p1/Z, [x21, XZR, LSL #2]\n" + ".inst 0xe0572aab // ld1h { za1h.h[x13, #3] }, p2/Z, [x21, x23, LSL #1]\n" + "ldr x21, [x10, #0x8]\n" + ".inst 0xe0bf86c0 // st1w { za0v.s[x12] }, p1/Z, [x22, XZR, LSL #2]\n" ".inst 0x25707541 // psel p1.s, p13.s/Z, p10.s[w12, #1]\n" - ".inst 0xe0b082a4 // st1w { za1v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n" + ".inst 0xe0b082c4 // st1w { za1v.s[x12] }, p0/Z, [x22, x16, LSL #2]\n" ".inst 0x25707540 // psel p0.s, p13.s/Z, p10.s[w12, #1]\n" "add x11, x11, #0x10\n" - ".inst 0xe0bc86a1 // st1w { za0v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n" + ".inst 0xe0bc86c1 // st1w { za0v.s[x12, #1] }, p1/Z, [x22, x28, LSL #2]\n" "add x10, x10, #0x10\n" "add x13, x13, #0x4\n" - ".inst 0xe0bb82a5 // st1w { za1v.s[x12, #1] }, p0/Z, [x21, x27, LSL #2]\n" + ".inst 0xe0bb82c5 // st1w { za1v.s[x12, #1] }, p0/Z, [x22, x27, LSL #2]\n" "add x12, x12, #0x2\n" "cmp x12, x15\n" - "addvl x21, x21, #4\n" + "addvl x22, x22, #4\n" "blt 4b\n" "5:" // K loop: Main loop: First: Tail ".inst 0x25396581 // psel p1.h, p9.h/Z, p12.h[w13, #1]\n" ".inst 0x25396160 // psel p0.h, p8.h/Z, p11.h[w13, #1]\n" - ".inst 0xe0562521 // ld1h { za0h.h[x13, #1] }, p1/Z, [x9, x22, LSL #1]\n" - ".inst 0xe0562349 // ld1h { za1h.h[x13, #1] }, p0/Z, [x26, x22, LSL #1]\n" + ".inst 0xe0572521 // ld1h { za0h.h[x13, #1] }, p1/Z, [x9, x23, LSL #1]\n" + ".inst 0xe0572349 // ld1h { za1h.h[x13, #1] }, p0/Z, [x26, x23, LSL #1]\n" "mov x11, %x[in]\n" "add x10, %x[in], x16, LSL #3\n" "ldr x9, [x11, #0x0]\n" ".inst 0x25796580 // psel p0.h, p9.h/Z, p12.h[w13, #3]\n" ".inst 0x25796161 // psel p1.h, p8.h/Z, p11.h[w13, #3]\n" - ".inst 0xe0562303 // ld1h { za0h.h[x13, #3] }, p0/Z, [x24, x22, LSL #1]\n" + ".inst 0xe0572303 // ld1h { za0h.h[x13, #3] }, p0/Z, [x24, x23, LSL #1]\n" "ldr x26, [x10, #0x0]\n" ".inst 0x25307540 // psel p0.s, p13.s/Z, p10.s[w12]\n" - ".inst 0xe05626eb // ld1h { za1h.h[x13, #3] }, p1/Z, [x23, x22, LSL #1]\n" + ".inst 0xe05726ab // ld1h { za1h.h[x13, #3] }, p1/Z, [x21, x23, LSL #1]\n" "ldr x24, [x11, #0x8]\n" ".inst 0x25307542 // psel p2.s, p13.s/Z, p10.s[w12]\n" - "ldr x23, [x10, #0x8]\n" - ".inst 0xe0bf82a0 // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n" + "ldr x21, [x10, #0x8]\n" + ".inst 0xe0bf82c0 // st1w { za0v.s[x12] }, p0/Z, [x22, XZR, LSL #2]\n" ".inst 0x25707541 // psel p1.s, p13.s/Z, p10.s[w12, #1]\n" ".inst 0x25707540 // psel p0.s, p13.s/Z, p10.s[w12, #1]\n" - ".inst 0xe0b08aa4 // st1w { za1v.s[x12] }, p2/Z, [x21, x16, LSL #2]\n" + ".inst 0xe0b08ac4 // st1w { za1v.s[x12] }, p2/Z, [x22, x16, LSL #2]\n" "whilelt p10.h, x14, %x[width]\n" "inch x14\n" - ".inst 0xe0bc86a1 // st1w { za0v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n" + ".inst 0xe0bc86c1 // st1w { za0v.s[x12, #1] }, p1/Z, [x22, x28, LSL #2]\n" "add x11, x11, #0x10\n" "add x10, x10, #0x10\n" - ".inst 0xe0bb82a5 // st1w { za1v.s[x12, #1] }, p0/Z, [x21, x27, LSL #2]\n" - "addvl x21, x21, #4\n" - "inch x22\n" + ".inst 0xe0bb82c5 // st1w { za1v.s[x12, #1] }, p0/Z, [x22, x27, LSL #2]\n" + "addvl x22, x22, #4\n" + "inch x23\n" "whilelt p9.h, x14, %x[width]\n" "whilelt p8.h, x14, %x[width]\n" "mov x13, #0x0\n" @@ -183,61 +181,61 @@ void interleave_block<2, 2, VLType::SME, false>( "6:" // K loop: Main loop: Second: Loop ".inst 0x25296581 // psel p1.h, p9.h/Z, p12.h[w13]\n" ".inst 0x25296160 // psel p0.h, p8.h/Z, p11.h[w13]\n" - ".inst 0xe0562520 // ld1h { za0h.h[x13] }, p1/Z, [x9, x22, LSL #1]\n" + ".inst 0xe0572520 // ld1h { za0h.h[x13] }, p1/Z, [x9, x23, LSL #1]\n" "ldr x9, [x11, #0x0]\n" - ".inst 0xe0562348 // ld1h { za1h.h[x13] }, p0/Z, [x26, x22, LSL #1]\n" + ".inst 0xe0572348 // ld1h { za1h.h[x13] }, p0/Z, [x26, x23, LSL #1]\n" ".inst 0x25696580 // psel p0.h, p9.h/Z, p12.h[w13, #2]\n" ".inst 0x25696162 // psel p2.h, p8.h/Z, p11.h[w13, #2]\n" "ldr x26, [x10, #0x0]\n" ".inst 0x25307541 // psel p1.s, p13.s/Z, p10.s[w12]\n" - ".inst 0xe0562302 // ld1h { za0h.h[x13, #2] }, p0/Z, [x24, x22, LSL #1]\n" + ".inst 0xe0572302 // ld1h { za0h.h[x13, #2] }, p0/Z, [x24, x23, LSL #1]\n" "ldr x24, [x11, #0x8]\n" ".inst 0x25307540 // psel p0.s, p13.s/Z, p10.s[w12]\n" - ".inst 0xe0562aea // ld1h { za1h.h[x13, #2] }, p2/Z, [x23, x22, LSL #1]\n" - "ldr x23, [x10, #0x8]\n" - ".inst 0xe0bf86a8 // st1w { za2v.s[x12] }, p1/Z, [x21, XZR, LSL #2]\n" + ".inst 0xe0572aaa // ld1h { za1h.h[x13, #2] }, p2/Z, [x21, x23, LSL #1]\n" + "ldr x21, [x10, #0x8]\n" + ".inst 0xe0bf86c8 // st1w { za2v.s[x12] }, p1/Z, [x22, XZR, LSL #2]\n" ".inst 0x25707541 // psel p1.s, p13.s/Z, p10.s[w12, #1]\n" - ".inst 0xe0b082ac // st1w { za3v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n" + ".inst 0xe0b082cc // st1w { za3v.s[x12] }, p0/Z, [x22, x16, LSL #2]\n" ".inst 0x25707540 // psel p0.s, p13.s/Z, p10.s[w12, #1]\n" "add x11, x11, #0x10\n" - ".inst 0xe0bc86a9 // st1w { za2v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n" + ".inst 0xe0bc86c9 // st1w { za2v.s[x12, #1] }, p1/Z, [x22, x28, LSL #2]\n" "add x10, x10, #0x10\n" "add x13, x13, #0x4\n" - ".inst 0xe0bb82ad // st1w { za3v.s[x12, #1] }, p0/Z, [x21, x27, LSL #2]\n" + ".inst 0xe0bb82cd // st1w { za3v.s[x12, #1] }, p0/Z, [x22, x27, LSL #2]\n" "add x12, x12, #0x2\n" "cmp x12, x15\n" - "addvl x21, x21, #4\n" + "addvl x22, x22, #4\n" "blt 6b\n" "7:" // K loop: Main loop: Second: Tail ".inst 0x25296581 // psel p1.h, p9.h/Z, p12.h[w13]\n" ".inst 0x25296160 // psel p0.h, p8.h/Z, p11.h[w13]\n" - ".inst 0xe0562520 // ld1h { za0h.h[x13] }, p1/Z, [x9, x22, LSL #1]\n" - ".inst 0xe0562348 // ld1h { za1h.h[x13] }, p0/Z, [x26, x22, LSL #1]\n" + ".inst 0xe0572520 // ld1h { za0h.h[x13] }, p1/Z, [x9, x23, LSL #1]\n" + ".inst 0xe0572348 // ld1h { za1h.h[x13] }, p0/Z, [x26, x23, LSL #1]\n" "mov x11, %x[in]\n" "add x10, %x[in], x16, LSL #3\n" "ldr x9, [x11, #0x0]\n" ".inst 0x25696580 // psel p0.h, p9.h/Z, p12.h[w13, #2]\n" ".inst 0x25696161 // psel p1.h, p8.h/Z, p11.h[w13, #2]\n" - ".inst 0xe0562302 // ld1h { za0h.h[x13, #2] }, p0/Z, [x24, x22, LSL #1]\n" + ".inst 0xe0572302 // ld1h { za0h.h[x13, #2] }, p0/Z, [x24, x23, LSL #1]\n" "ldr x26, [x10, #0x0]\n" ".inst 0x25307540 // psel p0.s, p13.s/Z, p10.s[w12]\n" - ".inst 0xe05626ea // ld1h { za1h.h[x13, #2] }, p1/Z, [x23, x22, LSL #1]\n" + ".inst 0xe05726aa // ld1h { za1h.h[x13, #2] }, p1/Z, [x21, x23, LSL #1]\n" "ldr x24, [x11, #0x8]\n" ".inst 0x25307542 // psel p2.s, p13.s/Z, p10.s[w12]\n" - "ldr x23, [x10, #0x8]\n" - ".inst 0xe0bf82a8 // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n" + "ldr x21, [x10, #0x8]\n" + ".inst 0xe0bf82c8 // st1w { za2v.s[x12] }, p0/Z, [x22, XZR, LSL #2]\n" ".inst 0x25707541 // psel p1.s, p13.s/Z, p10.s[w12, #1]\n" ".inst 0x25707540 // psel p0.s, p13.s/Z, p10.s[w12, #1]\n" - ".inst 0xe0b08aac // st1w { za3v.s[x12] }, p2/Z, [x21, x16, LSL #2]\n" + ".inst 0xe0b08acc // st1w { za3v.s[x12] }, p2/Z, [x22, x16, LSL #2]\n" "whilelt p10.h, x14, %x[width]\n" "subs x20, x20, #0x1\n" - ".inst 0xe0bc86a9 // st1w { za2v.s[x12, #1] }, p1/Z, [x21, x28, LSL #2]\n" + ".inst 0xe0bc86c9 // st1w { za2v.s[x12, #1] }, p1/Z, [x22, x28, LSL #2]\n" "add x11, x11, #0x10\n" "add x10, x10, #0x10\n" - ".inst 0xe0bb82ad // st1w { za3v.s[x12, #1] }, p0/Z, [x21, x27, LSL #2]\n" - "addvl x21, x21, #4\n" + ".inst 0xe0bb82cd // st1w { za3v.s[x12, #1] }, p0/Z, [x22, x27, LSL #2]\n" + "addvl x22, x22, #4\n" "inch x14\n" - "inch x22\n" + "inch x23\n" "bgt 3b\n" "8:" // K loop: Tails "cbnz x25, 11f\n" @@ -248,51 +246,51 @@ void interleave_block<2, 2, VLType::SME, false>( "mov x12, #0x0\n" "9:" // K loop: Tails: Even: First ".inst 0x25307540 // psel p0.s, p13.s/Z, p10.s[w12]\n" - ".inst 0xe0bf82a0 // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n" + ".inst 0xe0bf82c0 // st1w { za0v.s[x12] }, p0/Z, [x22, XZR, LSL #2]\n" ".inst 0x25307540 // psel p0.s, p13.s/Z, p10.s[w12]\n" - ".inst 0xe0b082a4 // st1w { za1v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n" - "ldr x9, [x11, #0x0]\n" + ".inst 0xe0b082c4 // st1w { za1v.s[x12] }, p0/Z, [x22, x16, LSL #2]\n" + "ldr x21, [x11, #0x0]\n" "add x12, x12, #0x1\n" ".inst 0x25396581 // psel p1.h, p9.h/Z, p12.h[w13, #1]\n" - "ldr x26, [x11, x16, LSL #0x3]\n" + "ldr x20, [x11, x16, LSL #0x3]\n" ".inst 0x25396160 // psel p0.h, p8.h/Z, p11.h[w13, #1]\n" "cmp x12, x16\n" - ".inst 0xe0562521 // ld1h { za0h.h[x13, #1] }, p1/Z, [x9, x22, LSL #1]\n" - ".inst 0xe0562349 // ld1h { za1h.h[x13, #1] }, p0/Z, [x26, x22, LSL #1]\n" + ".inst 0xe05726a1 // ld1h { za0h.h[x13, #1] }, p1/Z, [x21, x23, LSL #1]\n" + ".inst 0xe0572289 // ld1h { za1h.h[x13, #1] }, p0/Z, [x20, x23, LSL #1]\n" "add x11, x11, #0x8\n" - "addvl x21, x21, #2\n" + "addvl x22, x22, #2\n" "add x13, x13, #0x2\n" "blt 9b\n" "whilelt p10.h, x14, %x[width]\n" - "whilelt p9.h, x14, %x[width]\n" + "whilelt p8.h, x14, %x[width]\n" "whilelt p8.h, x14, %x[width]\n" "mov x20, #0x0\n" "mov x12, #0x0\n" "10:" // K loop: Tails: Even: Second ".inst 0x25307540 // psel p0.s, p13.s/Z, p10.s[w12]\n" - ".inst 0xe0bf82a8 // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n" + ".inst 0xe0bf82c8 // st1w { za2v.s[x12] }, p0/Z, [x22, XZR, LSL #2]\n" ".inst 0x25307540 // psel p0.s, p13.s/Z, p10.s[w12]\n" - ".inst 0xe0b082ac // st1w { za3v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n" + ".inst 0xe0b082cc // st1w { za3v.s[x12] }, p0/Z, [x22, x16, LSL #2]\n" "add x12, x12, #0x1\n" "cmp x12, x17\n" - "addvl x21, x21, #2\n" + "addvl x22, x22, #2\n" "add x20, x20, #0x2\n" "blt 10b\n" - "whilelt p10.h, x14, %x[width]\n" + "whilelt p8.h, x14, %x[width]\n" "b 13f\n" "11:" // K loop: Tails: Odd "mov x12, #0x0\n" "12:" // K loop: Tails: Odd: Loop ".inst 0x25307540 // psel p0.s, p13.s/Z, p10.s[w12]\n" - ".inst 0xe0bf82a0 // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n" + ".inst 0xe0bf82c0 // st1w { za0v.s[x12] }, p0/Z, [x22, XZR, LSL #2]\n" ".inst 0x25307540 // psel p0.s, p13.s/Z, p10.s[w12]\n" - ".inst 0xe0b082a4 // st1w { za1v.s[x12] }, p0/Z, [x21, x16, LSL #2]\n" + ".inst 0xe0b082c4 // st1w { za1v.s[x12] }, p0/Z, [x22, x16, LSL #2]\n" "add x12, x12, #0x1\n" "cmp x12, x17\n" - "addvl x21, x21, #2\n" + "addvl x22, x22, #2\n" "blt 12b\n" "13:" // K loop: End - "mov %x[out], x21\n" + "mov %x[out], x22\n" ".inst 0xd503467f // SMSTOP\n" : [out] "+&r" (out) : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width) @@ -300,4 +298,4 @@ void interleave_block<2, 2, VLType::SME, false>( ); } -#endif // defined(__ARM_FEATURE_SVE) +#endif // defined(ARM_COMPUTE_ENABLE_SME) |