diff options
Diffstat (limited to 'src/cpu/kernels/activation/generic/sve/lut.cpp')
| -rw-r--r-- | src/cpu/kernels/activation/generic/sve/lut.cpp | 332 |
1 files changed, 166 insertions, 166 deletions
diff --git a/src/cpu/kernels/activation/generic/sve/lut.cpp b/src/cpu/kernels/activation/generic/sve/lut.cpp index a31c0020a6..b73c87e319 100644 --- a/src/cpu/kernels/activation/generic/sve/lut.cpp +++ b/src/cpu/kernels/activation/generic/sve/lut.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2022 Arm Limited. + * Copyright (c) 2022-2023 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -43,20 +43,20 @@ void substitute_bytes_sve( { __asm__ __volatile__( "ptrue p0.b\n" - "cntd x24\n" + "cntd x25\n" "addvl %x[table], %x[table], #8\n" "ld1b { z16.b }, p0/Z, [%x[table], #-8, MUL VL]\n" - "tbnz x24, #5, 1f\n" + "tbnz x25, #5, 1f\n" "ld1b { z17.b }, p0/Z, [%x[table], #-7, MUL VL]\n" - "tbnz x24, #4, 1f\n" + "tbnz x25, #4, 1f\n" "ld1b { z18.b }, p0/Z, [%x[table], #-6, MUL VL]\n" "ld1b { z19.b }, p0/Z, [%x[table], #-5, MUL VL]\n" - "tbnz x24, #3, 1f\n" + "tbnz x25, #3, 1f\n" "ld1b { z20.b }, p0/Z, [%x[table], #-4, MUL VL]\n" "ld1b { z21.b }, p0/Z, [%x[table], #-3, MUL VL]\n" "ld1b { z22.b }, p0/Z, [%x[table], #-2, MUL VL]\n" "ld1b { z23.b }, p0/Z, [%x[table], #-1, MUL VL]\n" - "tbnz x24, #2, 1f\n" + "tbnz x25, #2, 1f\n" "ld1b { z24.b }, p0/Z, [%x[table]]\n" "ld1b { z25.b }, p0/Z, [%x[table], #1, MUL VL]\n" "ld1b { z26.b }, p0/Z, [%x[table], #2, MUL VL]\n" @@ -66,16 +66,16 @@ void substitute_bytes_sve( "ld1b { z30.b }, p0/Z, [%x[table], #6, MUL VL]\n" "ld1b { z31.b }, p0/Z, [%x[table], #7, MUL VL]\n" "1:" // Table load done - "mov x23, #0x0\n" + "mov x24, #0x0\n" "2:" // string loop - "ldr x22, [%x[input], x23, LSL #0x3]\n" - "ldr x21, [%x[output], x23, LSL #0x3]\n" - "tbnz x24, #5, 14f\n" - "tbnz x24, #4, 11f\n" - "tbnz x24, #3, 8f\n" - "tbnz x24, #2, 5f\n" + "ldr x23, [%x[input], x24, LSL #0x3]\n" + "ldr x22, [%x[output], x24, LSL #0x3]\n" + "tbnz x25, #5, 14f\n" + "tbnz x25, #4, 11f\n" + "tbnz x25, #3, 8f\n" + "tbnz x25, #2, 5f\n" "mov z12.b, #0x10\n" - "mov x20, %x[string_length]\n" + "mov x21, %x[string_length]\n" "ptrue p5.b\n" "ptrue p4.b\n" "ptrue p3.b\n" @@ -83,30 +83,30 @@ void substitute_bytes_sve( "ptrue p1.b\n" "ptrue p0.b\n" "3:" // 16 rounds: width loop - "addvl x19, x20, #-6\n" - "cmp x19, XZR\n" + "addvl x20, x21, #-6\n" + "cmp x20, XZR\n" "bge 4f\n" - "mov x19, #0x0\n" - "addvl x19, x19, #1\n" - "whilelt p5.b, XZR, x20\n" - "whilelt p4.b, x19, x20\n" - "addvl x19, x19, #1\n" - "whilelt p3.b, x19, x20\n" - "addvl x19, x19, #1\n" - "whilelt p2.b, x19, x20\n" - "addvl x19, x19, #1\n" - "whilelt p1.b, x19, x20\n" - "addvl x19, x19, #1\n" - "whilelt p0.b, x19, x20\n" + "mov x20, #0x0\n" + "addvl x20, x20, #1\n" + "whilelt p5.b, XZR, x21\n" + "whilelt p4.b, x20, x21\n" + "addvl x20, x20, #1\n" + "whilelt p3.b, x20, x21\n" + "addvl x20, x20, #1\n" + "whilelt p2.b, x20, x21\n" + "addvl x20, x20, #1\n" + "whilelt p1.b, x20, x21\n" + "addvl x20, x20, #1\n" + "whilelt p0.b, x20, x21\n" "4:" // 16 rounds: predicate OK - "ld1b { z11.b }, p5/Z, [x22]\n" - "ld1b { z10.b }, p4/Z, [x22, #1, MUL VL]\n" + "ld1b { z11.b }, p5/Z, [x23]\n" + "ld1b { z10.b }, p4/Z, [x23, #1, MUL VL]\n" "tbl z9.b, { z16.b }, z11.b\n" - "ld1b { z8.b }, p3/Z, [x22, #2, MUL VL]\n" - "ld1b { z7.b }, p2/Z, [x22, #3, MUL VL]\n" + "ld1b { z8.b }, p3/Z, [x23, #2, MUL VL]\n" + "ld1b { z7.b }, p2/Z, [x23, #3, MUL VL]\n" "sub z11.b, z11.b, z12.b\n" - "ld1b { z6.b }, p1/Z, [x22, #4, MUL VL]\n" - "ld1b { z5.b }, p0/Z, [x22, #5, MUL VL]\n" + "ld1b { z6.b }, p1/Z, [x23, #4, MUL VL]\n" + "ld1b { z5.b }, p0/Z, [x23, #5, MUL VL]\n" "tbl z4.b, { z16.b }, z10.b\n" "sub z10.b, z10.b, z12.b\n" "tbl z3.b, { z16.b }, z8.b\n" @@ -273,7 +273,7 @@ void substitute_bytes_sve( "sub z6.b, z6.b, z12.b\n" ".inst 0x05252fa0 // tbx z0.b, z29.b, z5.b\n" "sub z5.b, z5.b, z12.b\n" - "addvl x20, x20, #-6\n" + "addvl x21, x21, #-6\n" ".inst 0x052b2fc9 // tbx z9.b, z30.b, z11.b\n" "sub z11.b, z11.b, z12.b\n" ".inst 0x052a2fc4 // tbx z4.b, z30.b, z10.b\n" @@ -286,26 +286,26 @@ void substitute_bytes_sve( "sub z6.b, z6.b, z12.b\n" ".inst 0x05252fc0 // tbx z0.b, z30.b, z5.b\n" "sub z5.b, z5.b, z12.b\n" - "cmp x20, XZR\n" + "cmp x21, XZR\n" ".inst 0x052b2fe9 // tbx z9.b, z31.b, z11.b\n" ".inst 0x052a2fe4 // tbx z4.b, z31.b, z10.b\n" ".inst 0x05282fe3 // tbx z3.b, z31.b, z8.b\n" - "st1b { z9.b }, p5, [x21]\n" + "st1b { z9.b }, p5, [x22]\n" ".inst 0x05272fe2 // tbx z2.b, z31.b, z7.b\n" ".inst 0x05262fe1 // tbx z1.b, z31.b, z6.b\n" - "st1b { z4.b }, p4, [x21, #1, MUL VL]\n" + "st1b { z4.b }, p4, [x22, #1, MUL VL]\n" ".inst 0x05252fe0 // tbx z0.b, z31.b, z5.b\n" - "st1b { z3.b }, p3, [x21, #2, MUL VL]\n" + "st1b { z3.b }, p3, [x22, #2, MUL VL]\n" + "addvl x23, x23, #6\n" + "st1b { z2.b }, p2, [x22, #3, MUL VL]\n" + "st1b { z1.b }, p1, [x22, #4, MUL VL]\n" + "st1b { z0.b }, p0, [x22, #5, MUL VL]\n" "addvl x22, x22, #6\n" - "st1b { z2.b }, p2, [x21, #3, MUL VL]\n" - "st1b { z1.b }, p1, [x21, #4, MUL VL]\n" - "st1b { z0.b }, p0, [x21, #5, MUL VL]\n" - "addvl x21, x21, #6\n" "bgt 3b\n" "b 17f\n" "5:" // 256 bits "mov z12.b, #0x20\n" - "mov x20, %x[string_length]\n" + "mov x21, %x[string_length]\n" "ptrue p5.b\n" "ptrue p4.b\n" "ptrue p3.b\n" @@ -313,30 +313,30 @@ void substitute_bytes_sve( "ptrue p1.b\n" "ptrue p0.b\n" "6:" // 8 rounds: width loop - "addvl x19, x20, #-6\n" - "cmp x19, XZR\n" + "addvl x20, x21, #-6\n" + "cmp x20, XZR\n" "bge 7f\n" - "mov x19, #0x0\n" - "addvl x19, x19, #1\n" - "whilelt p5.b, XZR, x20\n" - "whilelt p4.b, x19, x20\n" - "addvl x19, x19, #1\n" - "whilelt p3.b, x19, x20\n" - "addvl x19, x19, #1\n" - "whilelt p2.b, x19, x20\n" - "addvl x19, x19, #1\n" - "whilelt p1.b, x19, x20\n" - "addvl x19, x19, #1\n" - "whilelt p0.b, x19, x20\n" + "mov x20, #0x0\n" + "addvl x20, x20, #1\n" + "whilelt p5.b, XZR, x21\n" + "whilelt p4.b, x20, x21\n" + "addvl x20, x20, #1\n" + "whilelt p3.b, x20, x21\n" + "addvl x20, x20, #1\n" + "whilelt p2.b, x20, x21\n" + "addvl x20, x20, #1\n" + "whilelt p1.b, x20, x21\n" + "addvl x20, x20, #1\n" + "whilelt p0.b, x20, x21\n" "7:" // 8 rounds: predicate OK - "ld1b { z11.b }, p5/Z, [x22]\n" - "ld1b { z10.b }, p4/Z, [x22, #1, MUL VL]\n" + "ld1b { z11.b }, p5/Z, [x23]\n" + "ld1b { z10.b }, p4/Z, [x23, #1, MUL VL]\n" "tbl z9.b, { z16.b }, z11.b\n" - "ld1b { z8.b }, p3/Z, [x22, #2, MUL VL]\n" - "ld1b { z7.b }, p2/Z, [x22, #3, MUL VL]\n" + "ld1b { z8.b }, p3/Z, [x23, #2, MUL VL]\n" + "ld1b { z7.b }, p2/Z, [x23, #3, MUL VL]\n" "sub z11.b, z11.b, z12.b\n" - "ld1b { z6.b }, p1/Z, [x22, #4, MUL VL]\n" - "ld1b { z5.b }, p0/Z, [x22, #5, MUL VL]\n" + "ld1b { z6.b }, p1/Z, [x23, #4, MUL VL]\n" + "ld1b { z5.b }, p0/Z, [x23, #5, MUL VL]\n" "tbl z4.b, { z16.b }, z10.b\n" "sub z10.b, z10.b, z12.b\n" "tbl z3.b, { z16.b }, z8.b\n" @@ -407,7 +407,7 @@ void substitute_bytes_sve( "sub z6.b, z6.b, z12.b\n" ".inst 0x05252ea0 // tbx z0.b, z21.b, z5.b\n" "sub z5.b, z5.b, z12.b\n" - "addvl x20, x20, #-6\n" + "addvl x21, x21, #-6\n" ".inst 0x052b2ec9 // tbx z9.b, z22.b, z11.b\n" "sub z11.b, z11.b, z12.b\n" ".inst 0x052a2ec4 // tbx z4.b, z22.b, z10.b\n" @@ -420,26 +420,26 @@ void substitute_bytes_sve( "sub z6.b, z6.b, z12.b\n" ".inst 0x05252ec0 // tbx z0.b, z22.b, z5.b\n" "sub z5.b, z5.b, z12.b\n" - "cmp x20, XZR\n" + "cmp x21, XZR\n" ".inst 0x052b2ee9 // tbx z9.b, z23.b, z11.b\n" ".inst 0x052a2ee4 // tbx z4.b, z23.b, z10.b\n" ".inst 0x05282ee3 // tbx z3.b, z23.b, z8.b\n" - "st1b { z9.b }, p5, [x21]\n" + "st1b { z9.b }, p5, [x22]\n" ".inst 0x05272ee2 // tbx z2.b, z23.b, z7.b\n" ".inst 0x05262ee1 // tbx z1.b, z23.b, z6.b\n" - "st1b { z4.b }, p4, [x21, #1, MUL VL]\n" + "st1b { z4.b }, p4, [x22, #1, MUL VL]\n" ".inst 0x05252ee0 // tbx z0.b, z23.b, z5.b\n" - "st1b { z3.b }, p3, [x21, #2, MUL VL]\n" + "st1b { z3.b }, p3, [x22, #2, MUL VL]\n" + "addvl x23, x23, #6\n" + "st1b { z2.b }, p2, [x22, #3, MUL VL]\n" + "st1b { z1.b }, p1, [x22, #4, MUL VL]\n" + "st1b { z0.b }, p0, [x22, #5, MUL VL]\n" "addvl x22, x22, #6\n" - "st1b { z2.b }, p2, [x21, #3, MUL VL]\n" - "st1b { z1.b }, p1, [x21, #4, MUL VL]\n" - "st1b { z0.b }, p0, [x21, #5, MUL VL]\n" - "addvl x21, x21, #6\n" "bgt 6b\n" "b 17f\n" "8:" // 512 bits "mov z12.b, #0x40\n" - "mov x20, %x[string_length]\n" + "mov x21, %x[string_length]\n" "ptrue p5.b\n" "ptrue p4.b\n" "ptrue p3.b\n" @@ -447,30 +447,30 @@ void substitute_bytes_sve( "ptrue p1.b\n" "ptrue p0.b\n" "9:" // 4 rounds: width loop - "addvl x19, x20, #-6\n" - "cmp x19, XZR\n" + "addvl x20, x21, #-6\n" + "cmp x20, XZR\n" "bge 10f\n" - "mov x19, #0x0\n" - "addvl x19, x19, #1\n" - "whilelt p5.b, XZR, x20\n" - "whilelt p4.b, x19, x20\n" - "addvl x19, x19, #1\n" - "whilelt p3.b, x19, x20\n" - "addvl x19, x19, #1\n" - "whilelt p2.b, x19, x20\n" - "addvl x19, x19, #1\n" - "whilelt p1.b, x19, x20\n" - "addvl x19, x19, #1\n" - "whilelt p0.b, x19, x20\n" + "mov x20, #0x0\n" + "addvl x20, x20, #1\n" + "whilelt p5.b, XZR, x21\n" + "whilelt p4.b, x20, x21\n" + "addvl x20, x20, #1\n" + "whilelt p3.b, x20, x21\n" + "addvl x20, x20, #1\n" + "whilelt p2.b, x20, x21\n" + "addvl x20, x20, #1\n" + "whilelt p1.b, x20, x21\n" + "addvl x20, x20, #1\n" + "whilelt p0.b, x20, x21\n" "10:" // 4 rounds: predicate OK - "ld1b { z11.b }, p5/Z, [x22]\n" - "ld1b { z10.b }, p4/Z, [x22, #1, MUL VL]\n" + "ld1b { z11.b }, p5/Z, [x23]\n" + "ld1b { z10.b }, p4/Z, [x23, #1, MUL VL]\n" "tbl z9.b, { z16.b }, z11.b\n" - "ld1b { z8.b }, p3/Z, [x22, #2, MUL VL]\n" - "ld1b { z7.b }, p2/Z, [x22, #3, MUL VL]\n" + "ld1b { z8.b }, p3/Z, [x23, #2, MUL VL]\n" + "ld1b { z7.b }, p2/Z, [x23, #3, MUL VL]\n" "sub z11.b, z11.b, z12.b\n" - "ld1b { z6.b }, p1/Z, [x22, #4, MUL VL]\n" - "ld1b { z5.b }, p0/Z, [x22, #5, MUL VL]\n" + "ld1b { z6.b }, p1/Z, [x23, #4, MUL VL]\n" + "ld1b { z5.b }, p0/Z, [x23, #5, MUL VL]\n" "tbl z4.b, { z16.b }, z10.b\n" "sub z10.b, z10.b, z12.b\n" "tbl z3.b, { z16.b }, z8.b\n" @@ -493,7 +493,7 @@ void substitute_bytes_sve( "sub z6.b, z6.b, z12.b\n" ".inst 0x05252e20 // tbx z0.b, z17.b, z5.b\n" "sub z5.b, z5.b, z12.b\n" - "addvl x20, x20, #-6\n" + "addvl x21, x21, #-6\n" ".inst 0x052b2e49 // tbx z9.b, z18.b, z11.b\n" "sub z11.b, z11.b, z12.b\n" ".inst 0x052a2e44 // tbx z4.b, z18.b, z10.b\n" @@ -506,26 +506,26 @@ void substitute_bytes_sve( "sub z6.b, z6.b, z12.b\n" ".inst 0x05252e40 // tbx z0.b, z18.b, z5.b\n" "sub z5.b, z5.b, z12.b\n" - "cmp x20, XZR\n" + "cmp x21, XZR\n" ".inst 0x052b2e69 // tbx z9.b, z19.b, z11.b\n" ".inst 0x052a2e64 // tbx z4.b, z19.b, z10.b\n" ".inst 0x05282e63 // tbx z3.b, z19.b, z8.b\n" - "st1b { z9.b }, p5, [x21]\n" + "st1b { z9.b }, p5, [x22]\n" ".inst 0x05272e62 // tbx z2.b, z19.b, z7.b\n" ".inst 0x05262e61 // tbx z1.b, z19.b, z6.b\n" - "st1b { z4.b }, p4, [x21, #1, MUL VL]\n" + "st1b { z4.b }, p4, [x22, #1, MUL VL]\n" ".inst 0x05252e60 // tbx z0.b, z19.b, z5.b\n" - "st1b { z3.b }, p3, [x21, #2, MUL VL]\n" + "st1b { z3.b }, p3, [x22, #2, MUL VL]\n" + "addvl x23, x23, #6\n" + "st1b { z2.b }, p2, [x22, #3, MUL VL]\n" + "st1b { z1.b }, p1, [x22, #4, MUL VL]\n" + "st1b { z0.b }, p0, [x22, #5, MUL VL]\n" "addvl x22, x22, #6\n" - "st1b { z2.b }, p2, [x21, #3, MUL VL]\n" - "st1b { z1.b }, p1, [x21, #4, MUL VL]\n" - "st1b { z0.b }, p0, [x21, #5, MUL VL]\n" - "addvl x21, x21, #6\n" "bgt 9b\n" "b 17f\n" "11:" // 1024 bits "mov z12.b, #0x80\n" - "mov x20, %x[string_length]\n" + "mov x21, %x[string_length]\n" "ptrue p5.b\n" "ptrue p4.b\n" "ptrue p3.b\n" @@ -533,30 +533,30 @@ void substitute_bytes_sve( "ptrue p1.b\n" "ptrue p0.b\n" "12:" // 2 rounds: width loop - "addvl x19, x20, #-6\n" - "cmp x19, XZR\n" + "addvl x20, x21, #-6\n" + "cmp x20, XZR\n" "bge 13f\n" - "mov x19, #0x0\n" - "addvl x19, x19, #1\n" - "whilelt p5.b, XZR, x20\n" - "whilelt p4.b, x19, x20\n" - "addvl x19, x19, #1\n" - "whilelt p3.b, x19, x20\n" - "addvl x19, x19, #1\n" - "whilelt p2.b, x19, x20\n" - "addvl x19, x19, #1\n" - "whilelt p1.b, x19, x20\n" - "addvl x19, x19, #1\n" - "whilelt p0.b, x19, x20\n" + "mov x20, #0x0\n" + "addvl x20, x20, #1\n" + "whilelt p5.b, XZR, x21\n" + "whilelt p4.b, x20, x21\n" + "addvl x20, x20, #1\n" + "whilelt p3.b, x20, x21\n" + "addvl x20, x20, #1\n" + "whilelt p2.b, x20, x21\n" + "addvl x20, x20, #1\n" + "whilelt p1.b, x20, x21\n" + "addvl x20, x20, #1\n" + "whilelt p0.b, x20, x21\n" "13:" // 2 rounds: predicate OK - "ld1b { z11.b }, p5/Z, [x22]\n" - "ld1b { z10.b }, p4/Z, [x22, #1, MUL VL]\n" - "addvl x20, x20, #-6\n" - "ld1b { z8.b }, p3/Z, [x22, #2, MUL VL]\n" - "ld1b { z7.b }, p2/Z, [x22, #3, MUL VL]\n" + "ld1b { z11.b }, p5/Z, [x23]\n" + "ld1b { z10.b }, p4/Z, [x23, #1, MUL VL]\n" + "addvl x21, x21, #-6\n" + "ld1b { z8.b }, p3/Z, [x23, #2, MUL VL]\n" + "ld1b { z7.b }, p2/Z, [x23, #3, MUL VL]\n" "tbl z9.b, { z16.b }, z11.b\n" - "ld1b { z6.b }, p1/Z, [x22, #4, MUL VL]\n" - "ld1b { z5.b }, p0/Z, [x22, #5, MUL VL]\n" + "ld1b { z6.b }, p1/Z, [x23, #4, MUL VL]\n" + "ld1b { z5.b }, p0/Z, [x23, #5, MUL VL]\n" "sub z11.b, z11.b, z12.b\n" "tbl z4.b, { z16.b }, z10.b\n" "sub z10.b, z10.b, z12.b\n" @@ -568,25 +568,25 @@ void substitute_bytes_sve( "sub z6.b, z6.b, z12.b\n" "tbl z0.b, { z16.b }, z5.b\n" "sub z5.b, z5.b, z12.b\n" - "cmp x20, XZR\n" + "cmp x21, XZR\n" ".inst 0x052b2e29 // tbx z9.b, z17.b, z11.b\n" ".inst 0x052a2e24 // tbx z4.b, z17.b, z10.b\n" ".inst 0x05282e23 // tbx z3.b, z17.b, z8.b\n" - "st1b { z9.b }, p5, [x21]\n" + "st1b { z9.b }, p5, [x22]\n" ".inst 0x05272e22 // tbx z2.b, z17.b, z7.b\n" ".inst 0x05262e21 // tbx z1.b, z17.b, z6.b\n" - "st1b { z4.b }, p4, [x21, #1, MUL VL]\n" + "st1b { z4.b }, p4, [x22, #1, MUL VL]\n" ".inst 0x05252e20 // tbx z0.b, z17.b, z5.b\n" - "st1b { z3.b }, p3, [x21, #2, MUL VL]\n" + "st1b { z3.b }, p3, [x22, #2, MUL VL]\n" + "addvl x23, x23, #6\n" + "st1b { z2.b }, p2, [x22, #3, MUL VL]\n" + "st1b { z1.b }, p1, [x22, #4, MUL VL]\n" + "st1b { z0.b }, p0, [x22, #5, MUL VL]\n" "addvl x22, x22, #6\n" - "st1b { z2.b }, p2, [x21, #3, MUL VL]\n" - "st1b { z1.b }, p1, [x21, #4, MUL VL]\n" - "st1b { z0.b }, p0, [x21, #5, MUL VL]\n" - "addvl x21, x21, #6\n" "bgt 12b\n" "b 17f\n" "14:" // 2048 bits - "mov x20, %x[string_length]\n" + "mov x21, %x[string_length]\n" "ptrue p5.b\n" "ptrue p4.b\n" "ptrue p3.b\n" @@ -594,52 +594,52 @@ void substitute_bytes_sve( "ptrue p1.b\n" "ptrue p0.b\n" "15:" // 1 rounds: width loop - "addvl x19, x20, #-6\n" - "cmp x19, XZR\n" + "addvl x20, x21, #-6\n" + "cmp x20, XZR\n" "bge 16f\n" - "mov x19, #0x0\n" - "addvl x19, x19, #1\n" - "whilelt p5.b, XZR, x20\n" - "whilelt p4.b, x19, x20\n" - "addvl x19, x19, #1\n" - "whilelt p3.b, x19, x20\n" - "addvl x19, x19, #1\n" - "whilelt p2.b, x19, x20\n" - "addvl x19, x19, #1\n" - "whilelt p1.b, x19, x20\n" - "addvl x19, x19, #1\n" - "whilelt p0.b, x19, x20\n" + "mov x20, #0x0\n" + "addvl x20, x20, #1\n" + "whilelt p5.b, XZR, x21\n" + "whilelt p4.b, x20, x21\n" + "addvl x20, x20, #1\n" + "whilelt p3.b, x20, x21\n" + "addvl x20, x20, #1\n" + "whilelt p2.b, x20, x21\n" + "addvl x20, x20, #1\n" + "whilelt p1.b, x20, x21\n" + "addvl x20, x20, #1\n" + "whilelt p0.b, x20, x21\n" "16:" // 1 rounds: predicate OK - "addvl x20, x20, #-6\n" - "ld1b { z11.b }, p5/Z, [x22]\n" - "ld1b { z10.b }, p4/Z, [x22, #1, MUL VL]\n" - "ld1b { z8.b }, p3/Z, [x22, #2, MUL VL]\n" - "ld1b { z7.b }, p2/Z, [x22, #3, MUL VL]\n" - "cmp x20, XZR\n" - "ld1b { z6.b }, p1/Z, [x22, #4, MUL VL]\n" - "ld1b { z5.b }, p0/Z, [x22, #5, MUL VL]\n" + "addvl x21, x21, #-6\n" + "ld1b { z11.b }, p5/Z, [x23]\n" + "ld1b { z10.b }, p4/Z, [x23, #1, MUL VL]\n" + "ld1b { z8.b }, p3/Z, [x23, #2, MUL VL]\n" + "ld1b { z7.b }, p2/Z, [x23, #3, MUL VL]\n" + "cmp x21, XZR\n" + "ld1b { z6.b }, p1/Z, [x23, #4, MUL VL]\n" + "ld1b { z5.b }, p0/Z, [x23, #5, MUL VL]\n" "tbl z9.b, { z16.b }, z11.b\n" "tbl z4.b, { z16.b }, z10.b\n" "tbl z3.b, { z16.b }, z8.b\n" - "st1b { z9.b }, p5, [x21]\n" + "st1b { z9.b }, p5, [x22]\n" "tbl z2.b, { z16.b }, z7.b\n" "tbl z1.b, { z16.b }, z6.b\n" - "st1b { z4.b }, p4, [x21, #1, MUL VL]\n" + "st1b { z4.b }, p4, [x22, #1, MUL VL]\n" "tbl z0.b, { z16.b }, z5.b\n" - "st1b { z3.b }, p3, [x21, #2, MUL VL]\n" + "st1b { z3.b }, p3, [x22, #2, MUL VL]\n" + "addvl x23, x23, #6\n" + "st1b { z2.b }, p2, [x22, #3, MUL VL]\n" + "st1b { z1.b }, p1, [x22, #4, MUL VL]\n" + "st1b { z0.b }, p0, [x22, #5, MUL VL]\n" "addvl x22, x22, #6\n" - "st1b { z2.b }, p2, [x21, #3, MUL VL]\n" - "st1b { z1.b }, p1, [x21, #4, MUL VL]\n" - "st1b { z0.b }, p0, [x21, #5, MUL VL]\n" - "addvl x21, x21, #6\n" "bgt 15b\n" "17:" // SVE body done - "add x23, x23, #0x1\n" - "cmp x23, %x[num_strings]\n" + "add x24, x24, #0x1\n" + "cmp x24, %x[num_strings]\n" "bne 2b\n" : [table] "+&r"(table) : [input] "r"(input), [num_strings] "r"(num_strings), [output] "r"(output), [string_length] "r"(string_length) - : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "x19", "x20", "x21", "x22", "x23", "x24", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"); + : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "x20", "x21", "x22", "x23", "x24", "x25", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"); } #endif // __aarch64__ } // namespace |