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 --- .../sme2_interleave2VL_block2_fp32_bf16.hpp | 187 +++++++++++++++++++++ 1 file changed, 187 insertions(+) create mode 100644 src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme2_interleave2VL_block2_fp32_bf16.hpp (limited to 'src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme2_interleave2VL_block2_fp32_bf16.hpp') diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme2_interleave2VL_block2_fp32_bf16.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme2_interleave2VL_block2_fp32_bf16.hpp new file mode 100644 index 0000000000..e712eca3ff --- /dev/null +++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme2_interleave2VL_block2_fp32_bf16.hpp @@ -0,0 +1,187 @@ +/* + * 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<2, 2, VLType::SME, false>( + bfloat16 * &out, const float * const *in, + size_t width, size_t height, size_t row_offset, bool first +) +{ + ARM_COMPUTE_UNUSED(first); + + __asm__ __volatile__( + ".inst 0xd503477f // SMSTART ZA\n" + "cntw x21, ALL, MUL #2\n" + "cntw x28\n" + "sub x27, %x[width], #0x1\n" + "cntw x20, ALL, MUL #2\n" + "sub x19, x21, #0x1\n" + ".inst 0x25207815 // ptrue pn13.b\n" + "whilelt p12.s, XZR, %x[height]\n" + "whilelt p11.s, x28, %x[height]\n" + "add x27, x27, x20\n" + "ands x26, %x[width], x19\n" + "udiv x27, x27, x20\n" + "csel x26, x26, x21, NE\n" + "mov x25, #0x0\n" + "and x24, x27, #0x1\n" + "sub x27, x27, #0x1\n" + "add x26, x26, #0x1\n" + "mov x19, %x[width]\n" + "mov x23, %x[in]\n" + "ptrue p0.b\n" + "mov x22, %x[outptr_raw]\n" + "mov x21, %x[row_offset]\n" + "lsr x27, x27, #0x1\n" + "lsr x26, x26, #0x1\n" + "mov x12, #0x0\n" + ".inst 0x25b34732 // whilelt pn10.s, x25, x19, VLx2\n" + "add x20, x23, x28, LSL #3\n" + "1:" // Width loop: Preamble: Loop + "ldr x19, [x23], #0x8\n" + ".inst 0x25306989 // psel p9.s, p10.s/Z, p12.s[w12]\n" + ".inst 0x25306968 // psel p8.s, p10.s/Z, p11.s[w12]\n" + ".inst 0xa0154678 // ld1w { z24.s-z25.s }, pn9.s/Z, [x19, x21, LSL #2]\n" + "ldr x19, [x20], #0x8\n" + ".inst 0xa0154276 // ld1w { z22.s-z23.s }, pn8.s/Z, [x19, x21, LSL #2]\n" + ".inst 0xc160e318 // bfcvt z24.h, { z24.s-z25.s }\n" + ".inst 0xc160e2d6 // bfcvt z22.h, { z22.s-z23.s }\n" + ".inst 0xc0800300 // mova za0h.s[x12], p0/M, z24.s\n" + ".inst 0xc08002c4 // mova za1h.s[x12], p0/M, z22.s\n" + "add x12, x12, #0x1\n" + "cmp x12, x28\n" + "blt 1b\n" + "incw x21, ALL, MUL #2\n" + "incw x25, ALL, MUL #2\n" + "cbz x27, 5f\n" + "2:" // Width loop + "mov x19, %x[width]\n" + "mov x23, %x[in]\n" + "mov x12, #0x0\n" + ".inst 0x25b34732 // whilelt pn10.s, x25, x19, VLx2\n" + "add x20, x23, x28, LSL #3\n" + "3:" // Width loop: Odd: Loop + "ldr x19, [x23], #0x8\n" + ".inst 0x25306989 // psel p9.s, p10.s/Z, p12.s[w12]\n" + ".inst 0x25306968 // psel p8.s, p10.s/Z, p11.s[w12]\n" + ".inst 0xa0154676 // ld1w { z22.s-z23.s }, pn9.s/Z, [x19, x21, LSL #2]\n" + "ldr x19, [x20], #0x8\n" + ".inst 0xa015426a // ld1w { z10.s-z11.s }, pn8.s/Z, [x19, x21, LSL #2]\n" + ".inst 0xc160e2d6 // bfcvt z22.h, { z22.s-z23.s }\n" + ".inst 0xc160e14a // bfcvt z10.h, { z10.s-z11.s }\n" + ".inst 0xc08002c8 // mova za2h.s[x12], p0/M, z22.s\n" + ".inst 0xc080014c // mova za3h.s[x12], p0/M, z10.s\n" + ".inst 0xc0828008 // mova z8.s, p0/M, za0v.s[x12]\n" + ".inst 0xc0828089 // mova z9.s, p0/M, za1v.s[x12]\n" + "add x12, x12, #0x1\n" + "cmp x12, x28\n" + ".inst 0xa06056c8 // st1w { z8.s-z9.s }, pn13.b, [x22]\n" + "addvl x22, x22, #2\n" + "blt 3b\n" + "incw x25, ALL, MUL #2\n" + "mov x19, %x[width]\n" + "mov x23, %x[in]\n" + "incw x21, ALL, MUL #2\n" + "mov x12, #0x0\n" + ".inst 0x25b34732 // whilelt pn10.s, x25, x19, VLx2\n" + "add x20, x23, x28, LSL #3\n" + "4:" // Width loop: Even: Loop + "ldr x19, [x23], #0x8\n" + ".inst 0x25306989 // psel p9.s, p10.s/Z, p12.s[w12]\n" + ".inst 0x25306968 // psel p8.s, p10.s/Z, p11.s[w12]\n" + ".inst 0xa015467a // ld1w { z26.s-z27.s }, pn9.s/Z, [x19, x21, LSL #2]\n" + "ldr x19, [x20], #0x8\n" + ".inst 0xa015427e // ld1w { z30.s-z31.s }, pn8.s/Z, [x19, x21, LSL #2]\n" + ".inst 0xc160e35a // bfcvt z26.h, { z26.s-z27.s }\n" + ".inst 0xc160e3de // bfcvt z30.h, { z30.s-z31.s }\n" + ".inst 0xc0800340 // mova za0h.s[x12], p0/M, z26.s\n" + ".inst 0xc08003c4 // mova za1h.s[x12], p0/M, z30.s\n" + ".inst 0xc0828106 // mova z6.s, p0/M, za2v.s[x12]\n" + ".inst 0xc082818e // mova z14.s, p0/M, za3v.s[x12]\n" + "add x12, x12, #0x1\n" + "cmp x12, x28\n" + ".inst 0xa16056c6 // st1w { z6.s, z14.s }, pn13.b, [x22]\n" + "addvl x22, x22, #2\n" + "blt 4b\n" + "subs x27, x27, #0x1\n" + "incw x21, ALL, MUL #2\n" + "incw x25, ALL, MUL #2\n" + "bgt 2b\n" + "5:" // Width loop: Tails + "cbnz x24, 8f\n" + "mov x19, %x[width]\n" + "mov x23, %x[in]\n" + "mov x12, #0x0\n" + ".inst 0x25b34732 // whilelt pn10.s, x25, x19, VLx2\n" + "add x20, x23, x28, LSL #3\n" + "6:" // Width loop: Tails: Even: Odd: Loop + "ldr x19, [x23], #0x8\n" + ".inst 0x25306989 // psel p9.s, p10.s/Z, p12.s[w12]\n" + ".inst 0x25306968 // psel p8.s, p10.s/Z, p11.s[w12]\n" + ".inst 0xa015466c // ld1w { z12.s-z13.s }, pn9.s/Z, [x19, x21, LSL #2]\n" + "ldr x19, [x20], #0x8\n" + ".inst 0xa015426e // ld1w { z14.s-z15.s }, pn8.s/Z, [x19, x21, LSL #2]\n" + ".inst 0xc160e18c // bfcvt z12.h, { z12.s-z13.s }\n" + ".inst 0xc160e1ce // bfcvt z14.h, { z14.s-z15.s }\n" + ".inst 0xc0800188 // mova za2h.s[x12], p0/M, z12.s\n" + ".inst 0xc08001cc // mova za3h.s[x12], p0/M, z14.s\n" + ".inst 0xc0828007 // mova z7.s, p0/M, za0v.s[x12]\n" + ".inst 0xc082808f // mova z15.s, p0/M, za1v.s[x12]\n" + "add x12, x12, #0x1\n" + "cmp x12, x28\n" + ".inst 0xa16056c7 // st1w { z7.s, z15.s }, pn13.b, [x22]\n" + "addvl x22, x22, #2\n" + "blt 6b\n" + "mov x12, #0x0\n" + "7:" // Width loop: Tails: Even: Even: Loop + ".inst 0xc082810e // mova z14.s, p0/M, za2v.s[x12]\n" + ".inst 0xc082818f // mova z15.s, p0/M, za3v.s[x12]\n" + "add x12, x12, #0x1\n" + "cmp x12, x26\n" + ".inst 0xa06056ce // st1w { z14.s-z15.s }, pn13.b, [x22]\n" + "addvl x22, x22, #2\n" + "blt 7b\n" + "b 10f\n" + "8:" // Width loop: Tails: Odd + "mov x12, #0x0\n" + "9:" // Width loop: Tails: Odd: Loop + ".inst 0xc0828014 // mova z20.s, p0/M, za0v.s[x12]\n" + ".inst 0xc0828095 // mova z21.s, p0/M, za1v.s[x12]\n" + "add x12, x12, #0x1\n" + "cmp x12, x26\n" + ".inst 0xa06056d4 // st1w { z20.s-z21.s }, pn13.b, [x22]\n" + "addvl x22, x22, #2\n" + "blt 9b\n" + "10:" // End + "mov %x[outptr_raw], x22\n" + ".inst 0xd503467f // SMSTOP\n" + : [outptr_raw] "+&r" (out) + : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width) + : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x12", "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