From 03b2971ac69a86f10a1566938d1a25afee15746c Mon Sep 17 00:00:00 2001 From: Viet-Hoa Do Date: Wed, 1 Jun 2022 11:47:14 +0100 Subject: Integrate SME2 kernels * Add SME/SME2 detection. * Integrate SME2 implementation for: - Normal convolution - Winograd - Depthwise convolution - Pooling Resolves: COMPMID-5700 Signed-off-by: Viet-Hoa Do Change-Id: I2f1ca1d05f8cfeee9309ed1c0a36096a4a6aad5c Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8692 Reviewed-by: Gunes Bayir Tested-by: Arm Jenkins Comments-Addressed: Arm Jenkins --- .../sme_interleave1VL_block4_u8_u8_summing.hpp | 252 +++++++++++++++++++++ 1 file changed, 252 insertions(+) create mode 100644 src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_u8_u8_summing.hpp (limited to 'src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_u8_u8_summing.hpp') diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_u8_u8_summing.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_u8_u8_summing.hpp new file mode 100644 index 0000000000..231d7ae213 --- /dev/null +++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_u8_u8_summing.hpp @@ -0,0 +1,252 @@ +/* + * Copyright (c) 2022 Arm Limited. + * + * SPDX-License-Identifier: MIT + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to + * deal in the Software without restriction, including without limitation the + * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or + * sell copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. + */ + +#if defined(__ARM_FEATURE_SVE) + +template <> +void interleave_block<1, 4, VLType::SME, true>( + uint8_t * &out, const uint8_t * const *in, + size_t width, size_t height, size_t row_offset, bool first +) +{ + __asm__ __volatile__( + ".inst 0xd503477f // SMSTART ZA\n" + "mov z18.b, #0x1\n" + "mov z17.s, #0x0\n" + "cntb x20\n" + "cntw x10\n" + "ptrue p1.b\n" + "mov x19, %x[width]\n" + "incb x19\n" + "sub x19, x19, #0x1\n" + "udiv x19, x19, x20\n" // n_passes = ceildiv(width, VL) + "sub x9, x19, #0x1\n" + "lsr x9, x9, #0x1\n" // n_loops = (n_passes - 1) / 2 + "and x28, x19, #0x1\n" // odd_tail = bool(n_passes & 0x1) + "mov x19, %x[width]\n" + "sub x27, x20, #0x1\n" + "ands x27, x19, x27\n" + "csel x27, x27, x20, NE\n" + "add x27, x27, #0x3\n" + "lsr x27, x27, #0x2\n" + "sub x26, x10, #0x2\n" + "ptrue p11.s\n" + "lsl x20, %x[height], #0x1\n" // height * 2 + "lsl x19, x10, #0x1\n" + "whilelt p9.b, XZR, x20\n" + "whilelt p8.b, x19, x20\n" + "zip1 p10.b, p9.b, p8.b\n" + "mov x25, %x[row_offset]\n" + "mov x24, %x[out]\n" + "mov x23, #0x0\n" + "whilelt p9.b, x23, %x[width]\n" + "whilelt p8.b, x23, %x[width]\n" + "cbnz %x[first], 1f\n" + "addvl x24, x24, #-1\n" + "ld1w { z17.s }, p1/Z, [x24]\n" + "1:" // K loop: Load row sums: End + "mov x22, %x[in]\n" + "ldr x21, [x22, #0x0]\n" + "ldr x20, [x22, #0x8]\n" + "add x22, x22, #0x10\n" + "mov x12, #0x0\n" + "cbz x26, 3f\n" + "2:" // K loop: Charge: Loop + ".inst 0x25246140 // dup p0.b, p8.b/Z, p10.b[w12]\n" + ".inst 0xe01902a0 // ld1b { za0h.b[x12] }, p0/Z, [x21, x25]\n" + ".inst 0x25646140 // dup p0.b, p8.b/Z, p10.b[w12, #4]\n" + "ldr x21, [x22, #0x0]\n" + ".inst 0xe0190284 // ld1b { za0h.b[x12, #4] }, p0/Z, [x20, x25]\n" + "ldr x20, [x22, #0x8]\n" + "add x22, x22, #0x10\n" + "add x12, x12, #0x8\n" + "cmp x12, x26, LSL #2\n" + "blt 2b\n" + "3:" // K loop: Charge: End + ".inst 0x25246140 // dup p0.b, p8.b/Z, p10.b[w12]\n" + ".inst 0xe01902a0 // ld1b { za0h.b[x12] }, p0/Z, [x21, x25]\n" + ".inst 0x25646140 // dup p0.b, p8.b/Z, p10.b[w12, #4]\n" + "mov x22, %x[in]\n" + ".inst 0xe0190284 // ld1b { za0h.b[x12, #4] }, p0/Z, [x20, x25]\n" + "ldr x21, [x22, #0x0]\n" + "ldr x20, [x22, #0x8]\n" + "add x22, x22, #0x10\n" + "incb x25\n" + "incb x23\n" + "cbz x9, 9f\n" + "mov x19, x9\n" + "4:" // K loop: Main loop + "whilelt p8.b, x23, %x[width]\n" + "mov x13, #0x0\n" + "mov x12, #0x0\n" + "cbz x26, 6f\n" + "5:" // K loop: Main loop: First: Loop + ".inst 0x25356140 // dup p0.b, p8.b/Z, p10.b[w13, #2]\n" + ".inst 0xe01922a2 // ld1b { za0h.b[x13, #2] }, p0/Z, [x21, x25]\n" + ".inst 0x25756140 // dup p0.b, p8.b/Z, p10.b[w13, #6]\n" + "ldr x21, [x22, #0x0]\n" + ".inst 0xe0192286 // ld1b { za0h.b[x13, #6] }, p0/Z, [x20, x25]\n" + "ldr x20, [x22, #0x8]\n" + "add x22, x22, #0x10\n" + ".inst 0xc0828410 // mova z16.s, p1/M, za0v.s[x12]\n" + ".inst 0x25306d20 // dup p0.s, p11.s/Z, p9.s[w12]\n" + ".inst 0xe0bf8300 // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n" + ".inst 0x25706d20 // dup p0.s, p11.s/Z, p9.s[w12, #1]\n" + "add x13, x13, #0x8\n" + ".inst 0xe0aa8301 // st1w { za0v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n" + "udot z17.s, z16.b, z18.b\n" + ".inst 0xc0828430 // mova z16.s, p1/M, za0v.s[x12, #1]\n" + "addvl x24, x24, #2\n" + "add x12, x12, #0x2\n" + "cmp x12, x26\n" + "udot z17.s, z16.b, z18.b\n" + "blt 5b\n" + "6:" // K loop: Main loop: First: Tail + "mov x22, %x[in]\n" + ".inst 0x25356140 // dup p0.b, p8.b/Z, p10.b[w13, #2]\n" + ".inst 0xe01922a2 // ld1b { za0h.b[x13, #2] }, p0/Z, [x21, x25]\n" + ".inst 0x25756140 // dup p0.b, p8.b/Z, p10.b[w13, #6]\n" + "ldr x21, [x22, #0x0]\n" + ".inst 0xe0192286 // ld1b { za0h.b[x13, #6] }, p0/Z, [x20, x25]\n" + "ldr x20, [x22, #0x8]\n" + "add x22, x22, #0x10\n" + ".inst 0xc0828410 // mova z16.s, p1/M, za0v.s[x12]\n" + ".inst 0x25306d20 // dup p0.s, p11.s/Z, p9.s[w12]\n" + ".inst 0xe0bf8300 // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n" + ".inst 0x25706d20 // dup p0.s, p11.s/Z, p9.s[w12, #1]\n" + "whilelt p9.b, x23, %x[width]\n" + ".inst 0xe0aa8301 // st1w { za0v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n" + "udot z17.s, z16.b, z18.b\n" + ".inst 0xc0828430 // mova z16.s, p1/M, za0v.s[x12, #1]\n" + "addvl x24, x24, #2\n" + "incb x23\n" + "incb x25\n" + "udot z17.s, z16.b, z18.b\n" + "whilelt p8.b, x23, %x[width]\n" + "mov x13, #0x0\n" + "mov x12, #0x0\n" + "cbz x26, 8f\n" + "7:" // K loop: Main loop: Second: Loop + ".inst 0x25256140 // dup p0.b, p8.b/Z, p10.b[w13]\n" + ".inst 0xe01922a0 // ld1b { za0h.b[x13] }, p0/Z, [x21, x25]\n" + ".inst 0x25656140 // dup p0.b, p8.b/Z, p10.b[w13, #4]\n" + "ldr x21, [x22, #0x0]\n" + ".inst 0xe0192284 // ld1b { za0h.b[x13, #4] }, p0/Z, [x20, x25]\n" + "ldr x20, [x22, #0x8]\n" + "add x22, x22, #0x10\n" + ".inst 0xc0828510 // mova z16.s, p1/M, za2v.s[x12]\n" + ".inst 0x25306d20 // dup p0.s, p11.s/Z, p9.s[w12]\n" + ".inst 0xe0bf8308 // st1w { za2v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n" + ".inst 0x25706d20 // dup p0.s, p11.s/Z, p9.s[w12, #1]\n" + "add x13, x13, #0x8\n" + ".inst 0xe0aa8309 // st1w { za2v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n" + "udot z17.s, z16.b, z18.b\n" + ".inst 0xc0828530 // mova z16.s, p1/M, za2v.s[x12, #1]\n" + "addvl x24, x24, #2\n" + "add x12, x12, #0x2\n" + "cmp x12, x26\n" + "udot z17.s, z16.b, z18.b\n" + "blt 7b\n" + "8:" // K loop: Main loop: Second: Tail + "mov x22, %x[in]\n" + ".inst 0x25256140 // dup p0.b, p8.b/Z, p10.b[w13]\n" + ".inst 0xe01922a0 // ld1b { za0h.b[x13] }, p0/Z, [x21, x25]\n" + ".inst 0x25656140 // dup p0.b, p8.b/Z, p10.b[w13, #4]\n" + "ldr x21, [x22, #0x0]\n" + ".inst 0xe0192284 // ld1b { za0h.b[x13, #4] }, p0/Z, [x20, x25]\n" + "ldr x20, [x22, #0x8]\n" + "add x22, x22, #0x10\n" + ".inst 0xc0828510 // mova z16.s, p1/M, za2v.s[x12]\n" + ".inst 0x25306d20 // dup p0.s, p11.s/Z, p9.s[w12]\n" + ".inst 0xe0bf8308 // st1w { za2v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n" + ".inst 0x25706d20 // dup p0.s, p11.s/Z, p9.s[w12, #1]\n" + "whilelt p9.b, x23, %x[width]\n" + ".inst 0xe0aa8309 // st1w { za2v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n" + "udot z17.s, z16.b, z18.b\n" + ".inst 0xc0828530 // mova z16.s, p1/M, za2v.s[x12, #1]\n" + "addvl x24, x24, #2\n" + "incb x23\n" + "incb x25\n" + "udot z17.s, z16.b, z18.b\n" + "subs x19, x19, #0x1\n" + "bgt 4b\n" + "9:" // K loop: Tails + "cbnz x28, 12f\n" + "mov x22, %x[in]\n" + "whilelt p8.b, x23, %x[width]\n" + "mov x13, #0x0\n" + "mov x12, #0x0\n" + "10:" // K loop: Tails: Even: First + ".inst 0xc0828410 // mova z16.s, p1/M, za0v.s[x12]\n" + ".inst 0x25306d20 // dup p0.s, p11.s/Z, p9.s[w12]\n" + ".inst 0xe0bf8300 // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n" + ".inst 0x25356140 // dup p0.b, p8.b/Z, p10.b[w13, #2]\n" + "addvl x24, x24, #1\n" + "ldr x21, [x22, #0x0]\n" + ".inst 0xe01922a2 // ld1b { za0h.b[x13, #2] }, p0/Z, [x21, x25]\n" + "udot z17.s, z16.b, z18.b\n" + "add x22, x22, #0x8\n" + "add x13, x13, #0x4\n" + "add x12, x12, #0x1\n" + "cmp x12, x10\n" + "blt 10b\n" + "whilelt p9.b, x23, %x[width]\n" + "whilelt p8.b, x23, %x[width]\n" + "mov x19, #0x0\n" + "mov x12, #0x0\n" + "11:" // K loop: Tails: Even: Second + ".inst 0xc0828510 // mova z16.s, p1/M, za2v.s[x12]\n" + ".inst 0x25306d20 // dup p0.s, p11.s/Z, p9.s[w12]\n" + ".inst 0xe0bf8308 // st1w { za2v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n" + "addvl x24, x24, #1\n" + "add x19, x19, #0x4\n" + "add x12, x12, #0x1\n" + "udot z17.s, z16.b, z18.b\n" + "cmp x12, x27\n" + "blt 11b\n" + "whilelt p9.b, x23, %x[width]\n" + "b 14f\n" + "12:" // K loop: Tails: Odd + "mov x12, #0x0\n" + "13:" // K loop: Tails: Odd: Loop + ".inst 0xc0828410 // mova z16.s, p1/M, za0v.s[x12]\n" + ".inst 0x25306d20 // dup p0.s, p11.s/Z, p9.s[w12]\n" + ".inst 0xe0bf8300 // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n" + "addvl x24, x24, #1\n" + "add x12, x12, #0x1\n" + "cmp x12, x27\n" + "udot z17.s, z16.b, z18.b\n" + "blt 13b\n" + "14:" // K loop: End + "st1w { z17.s }, p1, [x24]\n" + "addvl x24, x24, #1\n" + "mov %x[out], x24\n" + ".inst 0xd503467f // SMSTOP\n" + : [out] "+&r" (out) + : [first] "r" (first), [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width) + : "cc", "memory", "p0", "p1", "p8", "p9", "p10", "p11", "x9", "x10", "x12", "x13", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31" + ); +} + +#endif // defined(__ARM_FEATURE_SVE) -- cgit v1.2.1