diff options
author | Michael Tyler <michael.tyler@arm.com> | 2023-02-01 16:37:07 +0000 |
---|---|---|
committer | michael.tyler <michael.tyler@arm.com> | 2023-02-08 15:33:26 +0000 |
commit | 7d9a626aaba9837cb82d189a9c4f0bcef58825bb (patch) | |
tree | e3d8cfeb7f8539cca3a8bf2f1a8f412d25d89041 /src/core/NEON/kernels/arm_conv/pooling/kernels/sve_u8_nhwc_max_2x2_s1_output2x2_depthfirst/generic.cpp | |
parent | 4e2bbbbb23e6f4bd452f7f865e51228e1f51efec (diff) | |
download | ComputeLibrary-7d9a626aaba9837cb82d189a9c4f0bcef58825bb.tar.gz |
Update CPU kernels to remove x19 and w19
Resolves: COMPMID-5805
Change-Id: Idf720bbb136474810086f5089c5ed23b3f79835a
Signed-off-by: Michael Tyler <michael.tyler@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9081
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_conv/pooling/kernels/sve_u8_nhwc_max_2x2_s1_output2x2_depthfirst/generic.cpp')
-rw-r--r-- | src/core/NEON/kernels/arm_conv/pooling/kernels/sve_u8_nhwc_max_2x2_s1_output2x2_depthfirst/generic.cpp | 120 |
1 files changed, 60 insertions, 60 deletions
diff --git a/src/core/NEON/kernels/arm_conv/pooling/kernels/sve_u8_nhwc_max_2x2_s1_output2x2_depthfirst/generic.cpp b/src/core/NEON/kernels/arm_conv/pooling/kernels/sve_u8_nhwc_max_2x2_s1_output2x2_depthfirst/generic.cpp index 2a08610db6..70d308a585 100644 --- a/src/core/NEON/kernels/arm_conv/pooling/kernels/sve_u8_nhwc_max_2x2_s1_output2x2_depthfirst/generic.cpp +++ b/src/core/NEON/kernels/arm_conv/pooling/kernels/sve_u8_nhwc_max_2x2_s1_output2x2_depthfirst/generic.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2021-2022 Arm Limited. + * Copyright (c) 2021-2023 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -63,80 +63,80 @@ void sve_u8_nhwc_max_2x2_s1_output2x2_depthfirst_impl( pad_left, pad_top, pad_right, pad_bottom); __asm__ __volatile__( - "ldr x14, [%x[args], %[offsetof_n_channels]]\n" - "ldr x20, [%x[args], %[offsetof_outptrs]]\n" - "mov x13, #0x0\n" - "whilelt p2.b, x13, x14\n" - "ldr x19, [%x[args], %[offsetof_inptrs]]\n" - "ldp x12, x11, [x20, #0x0]\n" + "ldr x15, [%x[args], %[offsetof_n_channels]]\n" + "ldr x21, [%x[args], %[offsetof_outptrs]]\n" + "mov x14, #0x0\n" + "whilelt p2.b, x14, x15\n" + "ldr x20, [%x[args], %[offsetof_inptrs]]\n" + "ldp x13, x12, [x21, #0x0]\n" "ptrue p1.b\n" - "mov x10, #0x0\n" - "ldp x9, x28, [x20, #0x10]\n" - "ldp x27, x26, [x19, #0x0]\n" - "ldp x25, x24, [x19, #0x10]\n" - "ldp x23, x22, [x19, #0x20]\n" - "ldp x21, x20, [x19, #0x30]\n" - "ldr x19, [x19, #0x40]\n" - "ld1b { z31.b }, p2/Z, [x26, x13]\n" - "ld1b { z30.b }, p2/Z, [x23, x13]\n" - "ld1b { z29.b }, p2/Z, [x20, x13]\n" - "ld1b { z28.b }, p2/Z, [x24, x13]\n" - "ld1b { z27.b }, p2/Z, [x27, x13]\n" - "ld1b { z26.b }, p2/Z, [x22, x13]\n" - "ld1b { z25.b }, p2/Z, [x25, x13]\n" - "ld1b { z24.b }, p2/Z, [x21, x13]\n" - "ld1b { z23.b }, p2/Z, [x19, x13]\n" - "incw x13\n" - "whilelt p2.b, x13, x14\n" + "mov x11, #0x0\n" + "ldp x10, x9, [x21, #0x10]\n" + "ldp x28, x27, [x20, #0x0]\n" + "ldp x26, x25, [x20, #0x10]\n" + "ldp x24, x23, [x20, #0x20]\n" + "ldp x22, x21, [x20, #0x30]\n" + "ldr x20, [x20, #0x40]\n" + "ld1b { z31.b }, p2/Z, [x27, x14]\n" + "ld1b { z30.b }, p2/Z, [x24, x14]\n" + "ld1b { z29.b }, p2/Z, [x21, x14]\n" + "ld1b { z28.b }, p2/Z, [x25, x14]\n" + "ld1b { z27.b }, p2/Z, [x28, x14]\n" + "ld1b { z26.b }, p2/Z, [x26, x14]\n" + "ld1b { z25.b }, p2/Z, [x23, x14]\n" + "ld1b { z24.b }, p2/Z, [x22, x14]\n" + "ld1b { z23.b }, p2/Z, [x20, x14]\n" + "incw x14\n" + "whilelt p2.b, x14, x15\n" "b.none 2f\n" "1:" // Vector: Loop "movprfx z22, z31\n umax z22.b, p1/M, z22.b, z30.b\n" "movprfx z21, z30\n umax z21.b, p1/M, z21.b, z29.b\n" - "ld1b { z31.b }, p2/Z, [x26, x13]\n" - "ld1b { z30.b }, p2/Z, [x23, x13]\n" + "ld1b { z31.b }, p2/Z, [x27, x14]\n" + "ld1b { z30.b }, p2/Z, [x24, x14]\n" "movprfx z20, z28\n umax z20.b, p1/M, z20.b, z27.b\n" - "movprfx z17, z26\n umax z17.b, p1/M, z17.b, z25.b\n" - "ld1b { z29.b }, p2/Z, [x20, x13]\n" - "ld1b { z27.b }, p2/Z, [x27, x13]\n" - "movprfx z19, z24\n umax z19.b, p1/M, z19.b, z28.b\n" - "movprfx z18, z26\n umax z18.b, p1/M, z18.b, z23.b\n" - "ld1b { z28.b }, p2/Z, [x24, x13]\n" - "ld1b { z26.b }, p2/Z, [x22, x13]\n" - "ld1b { z25.b }, p2/Z, [x25, x13]\n" - "ld1b { z24.b }, p2/Z, [x21, x13]\n" - "whilelt p0.b, x10, x14\n" + "movprfx z19, z26\n umax z19.b, p1/M, z19.b, z25.b\n" + "ld1b { z29.b }, p2/Z, [x21, x14]\n" + "ld1b { z27.b }, p2/Z, [x28, x14]\n" + "movprfx z17, z28\n umax z17.b, p1/M, z17.b, z24.b\n" + "movprfx z18, z25\n umax z18.b, p1/M, z18.b, z23.b\n" + "ld1b { z28.b }, p2/Z, [x25, x14]\n" + "ld1b { z26.b }, p2/Z, [x26, x14]\n" + "ld1b { z25.b }, p2/Z, [x23, x14]\n" + "ld1b { z24.b }, p2/Z, [x22, x14]\n" + "whilelt p0.b, x11, x15\n" "movprfx z16, z22\n umax z16.b, p1/M, z16.b, z20.b\n" - "ld1b { z23.b }, p2/Z, [x19, x13]\n" - "incw x13\n" - "whilelt p2.b, x13, x14\n" - "st1b { z16.b }, p0, [x12, x10]\n" - "movprfx z16, z17\n umax z16.b, p1/M, z16.b, z22.b\n" - "movprfx z17, z21\n umax z17.b, p1/M, z17.b, z19.b\n" - "st1b { z16.b }, p0, [x11, x10]\n" - "movprfx z16, z21\n umax z16.b, p1/M, z16.b, z18.b\n" - "st1b { z17.b }, p0, [x9, x10]\n" - "st1b { z16.b }, p0, [x28, x10]\n" - "incw x10\n" + "ld1b { z23.b }, p2/Z, [x20, x14]\n" + "incw x14\n" + "whilelt p2.b, x14, x15\n" + "st1b { z16.b }, p0, [x13, x11]\n" + "movprfx z16, z19\n umax z16.b, p1/M, z16.b, z22.b\n" + "umax z17.b, p1/M, z17.b, z21.b\n" + "st1b { z16.b }, p0, [x12, x11]\n" + "movprfx z16, z18\n umax z16.b, p1/M, z16.b, z21.b\n" + "st1b { z17.b }, p0, [x10, x11]\n" + "st1b { z16.b }, p0, [x9, x11]\n" + "incw x11\n" "b.any 1b\n" "2:" // Vector: Tail "movprfx z22, z31\n umax z22.b, p1/M, z22.b, z30.b\n" "movprfx z21, z30\n umax z21.b, p1/M, z21.b, z29.b\n" "movprfx z20, z28\n umax z20.b, p1/M, z20.b, z27.b\n" - "movprfx z17, z26\n umax z17.b, p1/M, z17.b, z25.b\n" - "movprfx z19, z24\n umax z19.b, p1/M, z19.b, z28.b\n" - "movprfx z18, z26\n umax z18.b, p1/M, z18.b, z23.b\n" - "whilelt p0.b, x10, x14\n" + "movprfx z19, z26\n umax z19.b, p1/M, z19.b, z25.b\n" + "movprfx z17, z28\n umax z17.b, p1/M, z17.b, z24.b\n" + "movprfx z18, z25\n umax z18.b, p1/M, z18.b, z23.b\n" + "whilelt p0.b, x11, x15\n" "movprfx z16, z22\n umax z16.b, p1/M, z16.b, z20.b\n" - "st1b { z16.b }, p0, [x12, x10]\n" - "movprfx z16, z17\n umax z16.b, p1/M, z16.b, z22.b\n" - "movprfx z17, z21\n umax z17.b, p1/M, z17.b, z19.b\n" - "st1b { z16.b }, p0, [x11, x10]\n" - "movprfx z16, z21\n umax z16.b, p1/M, z16.b, z18.b\n" - "st1b { z17.b }, p0, [x9, x10]\n" - "st1b { z16.b }, p0, [x28, x10]\n" + "st1b { z16.b }, p0, [x13, x11]\n" + "movprfx z16, z19\n umax z16.b, p1/M, z16.b, z22.b\n" + "umax z17.b, p1/M, z17.b, z21.b\n" + "st1b { z16.b }, p0, [x12, x11]\n" + "movprfx z16, z18\n umax z16.b, p1/M, z16.b, z21.b\n" + "st1b { z17.b }, p0, [x10, x11]\n" + "st1b { z16.b }, p0, [x9, x11]\n" : : [args] "r" (&args), [offsetof_inptrs] "I" (offsetof(KernelArgs, inptrs)), [offsetof_n_channels] "I" (offsetof(KernelArgs, n_channels)), [offsetof_outptrs] "I" (offsetof(KernelArgs, outptrs)) - : "cc", "memory", "p0", "p1", "p2", "x9", "x10", "x11", "x12", "x13", "x14", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31" + : "cc", "memory", "p0", "p1", "p2", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31" ); } |