diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_conv/pooling/kernels/sme_fp32_nhwc_avg_3x3_s1_output2x2_depthfirst/generic.cpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_conv/pooling/kernels/sme_fp32_nhwc_avg_3x3_s1_output2x2_depthfirst/generic.cpp | 146 |
1 files changed, 73 insertions, 73 deletions
diff --git a/src/core/NEON/kernels/arm_conv/pooling/kernels/sme_fp32_nhwc_avg_3x3_s1_output2x2_depthfirst/generic.cpp b/src/core/NEON/kernels/arm_conv/pooling/kernels/sme_fp32_nhwc_avg_3x3_s1_output2x2_depthfirst/generic.cpp index 602ef59159..cf69800522 100644 --- a/src/core/NEON/kernels/arm_conv/pooling/kernels/sme_fp32_nhwc_avg_3x3_s1_output2x2_depthfirst/generic.cpp +++ b/src/core/NEON/kernels/arm_conv/pooling/kernels/sme_fp32_nhwc_avg_3x3_s1_output2x2_depthfirst/generic.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2022 Arm Limited. + * Copyright (c) 2022-2023 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -82,97 +82,97 @@ void sme_fp32_nhwc_avg_3x3_s1_output2x2_depthfirst_impl( pad_left, pad_top, pad_right, pad_bottom); __asm__ __volatile__( - "ldr x20, [%x[args], %[offsetof_outptrs]]\n" + "ldr x21, [%x[args], %[offsetof_outptrs]]\n" ".inst 0xd503477f // SMSTART ZA\n" - "mov x4, #0x0\n" - "mov x19, #0x4\n" - "ldr x5, [%x[args], %[offsetof_inptrs]]\n" - "whilelt p0.s, XZR, x19\n" - "add x19, %x[args], %[offsetof_rescale]\n" - "ld1rqw { z4.s }, p0/Z, [x19]\n" - "ldr x6, [%x[args], %[offsetof_n_channels]]\n" - "whilelt p1.s, x4, x6\n" - "mov x7, #0x0\n" - "ldp x8, x17, [x20, #0x0]\n" - "ldp x16, x15, [x20, #0x10]\n" - "ldp x14, x13, [x5, #0x0]\n" - "ld1w { z3.s }, p1/Z, [x13, x4, LSL #2]\n" - "ldp x12, x11, [x5, #0x10]\n" - "ld1w { z2.s }, p1/Z, [x12, x4, LSL #2]\n" - "ldp x10, x9, [x5, #0x20]\n" - "ld1w { z1.s }, p1/Z, [x9, x4, LSL #2]\n" - "ldp x28, x27, [x5, #0x30]\n" - "ld1w { z0.s }, p1/Z, [x28, x4, LSL #2]\n" - "ldp x26, x25, [x5, #0x40]\n" - "ld1w { z31.s }, p1/Z, [x25, x4, LSL #2]\n" - "ldp x24, x23, [x5, #0x50]\n" - "ld1w { z30.s }, p1/Z, [x24, x4, LSL #2]\n" - "ldp x22, x21, [x5, #0x60]\n" - "ld1w { z29.s }, p1/Z, [x10, x4, LSL #2]\n" - "ldp x20, x19, [x5, #0x70]\n" - "ld1w { z28.s }, p1/Z, [x26, x4, LSL #2]\n" - "ld1w { z27.s }, p1/Z, [x27, x4, LSL #2]\n" - "ld1w { z22.s }, p1/Z, [x23, x4, LSL #2]\n" - "ld1w { z21.s }, p1/Z, [x21, x4, LSL #2]\n" - "ld1w { z20.s }, p1/Z, [x20, x4, LSL #2]\n" - "ld1w { z26.s }, p1/Z, [x14, x4, LSL #2]\n" - "ld1w { z25.s }, p1/Z, [x11, x4, LSL #2]\n" - "ld1w { z24.s }, p1/Z, [x22, x4, LSL #2]\n" - "ld1w { z23.s }, p1/Z, [x19, x4, LSL #2]\n" - "incw x4\n" - "whilelt p1.s, x4, x6\n" + "mov x3, #0x0\n" + "mov x20, #0x4\n" + "ldr x4, [%x[args], %[offsetof_inptrs]]\n" + "whilelt p0.s, XZR, x20\n" + "add x20, %x[args], %[offsetof_rescale]\n" + "ld1rqw { z4.s }, p0/Z, [x20]\n" + "ldr x5, [%x[args], %[offsetof_n_channels]]\n" + "whilelt p1.s, x3, x5\n" + "mov x6, #0x0\n" + "ldp x7, x8, [x21, #0x0]\n" + "ldp x17, x16, [x21, #0x10]\n" + "ldp x15, x14, [x4, #0x0]\n" + "ld1w { z3.s }, p1/Z, [x14, x3, LSL #2]\n" + "ldp x13, x12, [x4, #0x10]\n" + "ld1w { z2.s }, p1/Z, [x13, x3, LSL #2]\n" + "ldp x11, x10, [x4, #0x20]\n" + "ld1w { z1.s }, p1/Z, [x10, x3, LSL #2]\n" + "ldp x9, x28, [x4, #0x30]\n" + "ld1w { z0.s }, p1/Z, [x9, x3, LSL #2]\n" + "ldp x27, x26, [x4, #0x40]\n" + "ld1w { z31.s }, p1/Z, [x26, x3, LSL #2]\n" + "ldp x25, x24, [x4, #0x50]\n" + "ld1w { z30.s }, p1/Z, [x25, x3, LSL #2]\n" + "ldp x23, x22, [x4, #0x60]\n" + "ld1w { z29.s }, p1/Z, [x11, x3, LSL #2]\n" + "ldp x21, x20, [x4, #0x70]\n" + "ld1w { z28.s }, p1/Z, [x27, x3, LSL #2]\n" + "ld1w { z27.s }, p1/Z, [x28, x3, LSL #2]\n" + "ld1w { z22.s }, p1/Z, [x24, x3, LSL #2]\n" + "ld1w { z21.s }, p1/Z, [x22, x3, LSL #2]\n" + "ld1w { z20.s }, p1/Z, [x21, x3, LSL #2]\n" + "ld1w { z26.s }, p1/Z, [x15, x3, LSL #2]\n" + "ld1w { z25.s }, p1/Z, [x12, x3, LSL #2]\n" + "ld1w { z24.s }, p1/Z, [x23, x3, LSL #2]\n" + "ld1w { z23.s }, p1/Z, [x20, x3, LSL #2]\n" + "incw x3\n" + "whilelt p1.s, x3, x5\n" "b.none 2f\n" "1:" // Vector: Loop "fadd z17.s, z1.s, z0.s\n" "fadd z16.s, z31.s, z30.s\n" - "ld1w { z1.s }, p1/Z, [x9, x4, LSL #2]\n" - "whilelt p0.s, x7, x6\n" + "ld1w { z1.s }, p1/Z, [x10, x3, LSL #2]\n" + "whilelt p0.s, x6, x5\n" "fadd z19.s, z17.s, z16.s\n" "fadd z18.s, z3.s, z2.s\n" - "ld1w { z0.s }, p1/Z, [x28, x4, LSL #2]\n" + "ld1w { z0.s }, p1/Z, [x9, x3, LSL #2]\n" "fadd z17.s, z29.s, z28.s\n" "fadd z22.s, z27.s, z22.s\n" - "ld1w { z31.s }, p1/Z, [x25, x4, LSL #2]\n" + "ld1w { z31.s }, p1/Z, [x26, x3, LSL #2]\n" "fadd z16.s, z21.s, z20.s\n" "fadd z21.s, z18.s, z19.s\n" - "ld1w { z30.s }, p1/Z, [x24, x4, LSL #2]\n" + "ld1w { z30.s }, p1/Z, [x25, x3, LSL #2]\n" "fadd z20.s, z16.s, z19.s\n" "fadd z19.s, z26.s, z17.s\n" - "ld1w { z3.s }, p1/Z, [x13, x4, LSL #2]\n" + "ld1w { z3.s }, p1/Z, [x14, x3, LSL #2]\n" "fadd z18.s, z25.s, z22.s\n" "fadd z17.s, z24.s, z17.s\n" - "ld1w { z2.s }, p1/Z, [x12, x4, LSL #2]\n" + "ld1w { z2.s }, p1/Z, [x13, x3, LSL #2]\n" "fadd z16.s, z23.s, z22.s\n" - "fadd z19.s, z19.s, z21.s\n" - "ld1w { z29.s }, p1/Z, [x10, x4, LSL #2]\n" - "fadd z18.s, z18.s, z21.s\n" + "fadd z19.s, z21.s, z19.s\n" + "ld1w { z29.s }, p1/Z, [x11, x3, LSL #2]\n" + "fadd z18.s, z21.s, z18.s\n" "fadd z17.s, z17.s, z20.s\n" - "ld1w { z28.s }, p1/Z, [x26, x4, LSL #2]\n" + "ld1w { z28.s }, p1/Z, [x27, x3, LSL #2]\n" "fadd z16.s, z16.s, z20.s\n" - "ld1w { z27.s }, p1/Z, [x27, x4, LSL #2]\n" + "ld1w { z27.s }, p1/Z, [x28, x3, LSL #2]\n" "fmul z19.s, z19.s, z4.s[0]\n" - "ld1w { z22.s }, p1/Z, [x23, x4, LSL #2]\n" + "ld1w { z22.s }, p1/Z, [x24, x3, LSL #2]\n" "fmul z18.s, z18.s, z4.s[1]\n" "fmul z17.s, z17.s, z4.s[2]\n" - "ld1w { z21.s }, p1/Z, [x21, x4, LSL #2]\n" + "ld1w { z21.s }, p1/Z, [x22, x3, LSL #2]\n" "fmul z16.s, z16.s, z4.s[3]\n" - "st1w { z19.s }, p0, [x8, x7, LSL #2]\n" - "ld1w { z20.s }, p1/Z, [x20, x4, LSL #2]\n" - "st1w { z18.s }, p0, [x17, x7, LSL #2]\n" - "ld1w { z26.s }, p1/Z, [x14, x4, LSL #2]\n" - "st1w { z17.s }, p0, [x16, x7, LSL #2]\n" - "ld1w { z25.s }, p1/Z, [x11, x4, LSL #2]\n" - "st1w { z16.s }, p0, [x15, x7, LSL #2]\n" - "incw x7\n" - "ld1w { z24.s }, p1/Z, [x22, x4, LSL #2]\n" - "ld1w { z23.s }, p1/Z, [x19, x4, LSL #2]\n" - "incw x4\n" - "whilelt p1.s, x4, x6\n" + "st1w { z19.s }, p0, [x7, x6, LSL #2]\n" + "ld1w { z20.s }, p1/Z, [x21, x3, LSL #2]\n" + "st1w { z18.s }, p0, [x8, x6, LSL #2]\n" + "ld1w { z26.s }, p1/Z, [x15, x3, LSL #2]\n" + "st1w { z17.s }, p0, [x17, x6, LSL #2]\n" + "ld1w { z25.s }, p1/Z, [x12, x3, LSL #2]\n" + "st1w { z16.s }, p0, [x16, x6, LSL #2]\n" + "incw x6\n" + "ld1w { z24.s }, p1/Z, [x23, x3, LSL #2]\n" + "ld1w { z23.s }, p1/Z, [x20, x3, LSL #2]\n" + "incw x3\n" + "whilelt p1.s, x3, x5\n" "b.any 1b\n" "2:" // Vector: Tail "fadd z17.s, z1.s, z0.s\n" "fadd z16.s, z31.s, z30.s\n" - "whilelt p0.s, x7, x6\n" + "whilelt p0.s, x6, x5\n" "fadd z19.s, z17.s, z16.s\n" "fadd z18.s, z3.s, z2.s\n" "fadd z17.s, z29.s, z28.s\n" @@ -184,22 +184,22 @@ void sme_fp32_nhwc_avg_3x3_s1_output2x2_depthfirst_impl( "fadd z18.s, z25.s, z22.s\n" "fadd z17.s, z24.s, z17.s\n" "fadd z16.s, z23.s, z22.s\n" - "fadd z19.s, z19.s, z21.s\n" - "fadd z18.s, z18.s, z21.s\n" + "fadd z19.s, z21.s, z19.s\n" + "fadd z18.s, z21.s, z18.s\n" "fadd z17.s, z17.s, z20.s\n" "fadd z16.s, z16.s, z20.s\n" "fmul z19.s, z19.s, z4.s[0]\n" - "st1w { z19.s }, p0, [x8, x7, LSL #2]\n" + "st1w { z19.s }, p0, [x7, x6, LSL #2]\n" "fmul z18.s, z18.s, z4.s[1]\n" "fmul z17.s, z17.s, z4.s[2]\n" - "st1w { z18.s }, p0, [x17, x7, LSL #2]\n" + "st1w { z18.s }, p0, [x8, x6, LSL #2]\n" "fmul z16.s, z16.s, z4.s[3]\n" - "st1w { z17.s }, p0, [x16, x7, LSL #2]\n" - "st1w { z16.s }, p0, [x15, x7, LSL #2]\n" + "st1w { z17.s }, p0, [x17, x6, LSL #2]\n" + "st1w { z16.s }, p0, [x16, x6, LSL #2]\n" ".inst 0xd503467f // SMSTOP\n" : : [args] "r" (&args), [offsetof_inptrs] "I" (offsetof(KernelArgs, inptrs)), [offsetof_n_channels] "I" (offsetof(KernelArgs, n_channels)), [offsetof_outptrs] "I" (offsetof(KernelArgs, outptrs)), [offsetof_rescale] "I" (offsetof(KernelArgs, rescale_vals)) - : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x4", "x5", "x6", "x7", "x8", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "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" + : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x3", "x4", "x5", "x6", "x7", "x8", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "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" ); } |