diff options
author | Georgios Pinitas <georgios.pinitas@arm.com> | 2020-07-02 20:02:20 +0100 |
---|---|---|
committer | Georgios Pinitas <georgios.pinitas@arm.com> | 2020-07-06 16:51:32 +0000 |
commit | 5aa1a0b7ca5eed010e4b297a95b1c4851f741328 (patch) | |
tree | ba882de9e86589dfdd33937d538a89bbdf01c40e /src/core/NEON/kernels/arm_gemm/kernels/sve_interleaved_u8u32_dot_3VLx8 | |
parent | 42550c039105597ff6acd4e5efc0ee3c7c20b08e (diff) | |
download | ComputeLibrary-5aa1a0b7ca5eed010e4b297a95b1c4851f741328.tar.gz |
COMPID-3324: Clean GEMM kernels
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: I170de1671e061a78740caee31fb4a1b8642c1369
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3505
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/kernels/sve_interleaved_u8u32_dot_3VLx8')
-rw-r--r-- | src/core/NEON/kernels/arm_gemm/kernels/sve_interleaved_u8u32_dot_3VLx8/generic.cpp | 31 |
1 files changed, 16 insertions, 15 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/sve_interleaved_u8u32_dot_3VLx8/generic.cpp b/src/core/NEON/kernels/arm_gemm/kernels/sve_interleaved_u8u32_dot_3VLx8/generic.cpp index f4d33a9efa..6626f8463b 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/sve_interleaved_u8u32_dot_3VLx8/generic.cpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/sve_interleaved_u8u32_dot_3VLx8/generic.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018 Arm Limited. + * Copyright (c) 2019-2020 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -51,20 +51,20 @@ void sve_interleaved_u8u32_dot_3VLx8(const uint8_t *Apanel, const uint8_t *Bpane "mov z9.s, #0\n" "mov z10.s, #0\n" "mov z11.s, #0\n" - "mov z12.s, #0\n" "ld1rqb z0.b, p0/z, [%[a_ptr]]\n" - "mov z13.s, #0\n" + "mov z12.s, #0\n" "ld1b z4.b, p0/z, [%[b_ptr]]\n" - "mov z14.s, #0\n" + "mov z13.s, #0\n" "ld1rqb z1.b, p0/z, [%[a_ptr], #0x10]\n" - "mov z15.s, #0\n" + "mov z14.s, #0\n" "ld1b z5.b, p0/z, [%[b_ptr], #1, MUL VL]\n" - "mov z16.s, #0\n" + "mov z15.s, #0\n" "ld1rqb z2.b, p0/z, [%[a_ptr], #0x20]\n" - "mov z17.s, #0\n" + "mov z16.s, #0\n" "add %[a_ptr], %[a_ptr], #0x40\n" - "mov z18.s, #0\n" + "mov z17.s, #0\n" "addvl %[b_ptr], %[b_ptr], #3\n" + "mov z18.s, #0\n" "mov z19.s, #0\n" "mov z20.s, #0\n" "mov z21.s, #0\n" @@ -208,8 +208,8 @@ void sve_interleaved_u8u32_dot_3VLx8(const uint8_t *Apanel, const uint8_t *Bpane "udot z9.s, z4.b, z0.b[1]\n" "udot z10.s, z4.b, z0.b[2]\n" "udot z11.s, z4.b, z0.b[3]\n" - "udot z20.s, z4.b, z1.b[0]\n" "st1w z8.s, p0, [%[c_ptr]]\n" + "udot z20.s, z4.b, z1.b[0]\n" "udot z21.s, z4.b, z1.b[1]\n" "udot z22.s, z4.b, z1.b[2]\n" "udot z23.s, z4.b, z1.b[3]\n" @@ -217,8 +217,8 @@ void sve_interleaved_u8u32_dot_3VLx8(const uint8_t *Apanel, const uint8_t *Bpane "udot z13.s, z5.b, z0.b[1]\n" "udot z14.s, z5.b, z0.b[2]\n" "udot z15.s, z5.b, z0.b[3]\n" - "udot z24.s, z5.b, z1.b[0]\n" "st1w z12.s, p0, [%[c_ptr], #1, MUL VL]\n" + "udot z24.s, z5.b, z1.b[0]\n" "udot z25.s, z5.b, z1.b[1]\n" "udot z26.s, z5.b, z1.b[2]\n" "udot z27.s, z5.b, z1.b[3]\n" @@ -226,10 +226,11 @@ void sve_interleaved_u8u32_dot_3VLx8(const uint8_t *Apanel, const uint8_t *Bpane "udot z17.s, z6.b, z0.b[1]\n" "udot z18.s, z6.b, z0.b[2]\n" "udot z19.s, z6.b, z0.b[3]\n" - "udot z28.s, z6.b, z1.b[0]\n" "st1w z16.s, p0, [%[c_ptr], #2, MUL VL]\n" + "udot z28.s, z6.b, z1.b[0]\n" "udot z29.s, z6.b, z1.b[1]\n" "udot z30.s, z6.b, z1.b[2]\n" + "st1w z9.s, p0, [%[c_ptr], #3, MUL VL]\n" "udot z31.s, z6.b, z1.b[3]\n" "b 4f\n" "3:\n" @@ -267,8 +268,8 @@ void sve_interleaved_u8u32_dot_3VLx8(const uint8_t *Apanel, const uint8_t *Bpane "udot z9.s, z4.b, z2.b[1]\n" "udot z10.s, z4.b, z2.b[2]\n" "udot z11.s, z4.b, z2.b[3]\n" - "udot z20.s, z4.b, z3.b[0]\n" "st1w z8.s, p0, [%[c_ptr]]\n" + "udot z20.s, z4.b, z3.b[0]\n" "udot z21.s, z4.b, z3.b[1]\n" "udot z22.s, z4.b, z3.b[2]\n" "udot z23.s, z4.b, z3.b[3]\n" @@ -276,8 +277,8 @@ void sve_interleaved_u8u32_dot_3VLx8(const uint8_t *Apanel, const uint8_t *Bpane "udot z13.s, z5.b, z2.b[1]\n" "udot z14.s, z5.b, z2.b[2]\n" "udot z15.s, z5.b, z2.b[3]\n" - "udot z24.s, z5.b, z3.b[0]\n" "st1w z12.s, p0, [%[c_ptr], #1, MUL VL]\n" + "udot z24.s, z5.b, z3.b[0]\n" "udot z25.s, z5.b, z3.b[1]\n" "udot z26.s, z5.b, z3.b[2]\n" "udot z27.s, z5.b, z3.b[3]\n" @@ -285,13 +286,13 @@ void sve_interleaved_u8u32_dot_3VLx8(const uint8_t *Apanel, const uint8_t *Bpane "udot z17.s, z6.b, z2.b[1]\n" "udot z18.s, z6.b, z2.b[2]\n" "udot z19.s, z6.b, z2.b[3]\n" - "udot z28.s, z6.b, z3.b[0]\n" "st1w z16.s, p0, [%[c_ptr], #2, MUL VL]\n" + "udot z28.s, z6.b, z3.b[0]\n" "udot z29.s, z6.b, z3.b[1]\n" "udot z30.s, z6.b, z3.b[2]\n" + "st1w z9.s, p0, [%[c_ptr], #3, MUL VL]\n" "udot z31.s, z6.b, z3.b[3]\n" "4:\n" - "st1w z9.s, p0, [%[c_ptr], #3, MUL VL]\n" "st1w z13.s, p0, [%[c_ptr], #4, MUL VL]\n" "st1w z17.s, p0, [%[c_ptr], #5, MUL VL]\n" "st1w z10.s, p0, [%[c_ptr], #6, MUL VL]\n" |