diff options
Diffstat (limited to 'src/core/NEON/kernels/convolution/winograd/output_transforms/sme_fp32_mopa_4x4_3x3.cpp')
| -rw-r--r-- | src/core/NEON/kernels/convolution/winograd/output_transforms/sme_fp32_mopa_4x4_3x3.cpp | 90 |
1 files changed, 45 insertions, 45 deletions
diff --git a/src/core/NEON/kernels/convolution/winograd/output_transforms/sme_fp32_mopa_4x4_3x3.cpp b/src/core/NEON/kernels/convolution/winograd/output_transforms/sme_fp32_mopa_4x4_3x3.cpp index 9d3c751d78..54749e6f28 100644 --- a/src/core/NEON/kernels/convolution/winograd/output_transforms/sme_fp32_mopa_4x4_3x3.cpp +++ b/src/core/NEON/kernels/convolution/winograd/output_transforms/sme_fp32_mopa_4x4_3x3.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2022 Arm Limited. + * Copyright (c) 2022-2023 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -112,17 +112,17 @@ void sme_fp32_mopa_4x4_3x3( "ld1rw { z12.s }, p5/Z, [%x[params], %[offsetof_Params_act_min]]\n" "ld1rw { z10.s }, p5/Z, [%x[params], %[offsetof_Params_act_max]]\n" "pfalse p8.b\n" - "ldr x19, [%x[params], %[offsetof_Params_inner_terms]]\n" + "ldr x8, [%x[params], %[offsetof_Params_inner_terms]]\n" "ld1w { z6.s }, p5/Z, [x20]\n" "ld1w { z7.s }, p5/Z, [x20, #1, MUL VL]\n" - "ld1rqw { z9.s }, p5/Z, [x19]\n" - "ld1rqw { z8.s }, p5/Z, [x19, #16]\n" - "ld1rqw { z15.s }, p5/Z, [x19, #32]\n" + "ld1rqw { z9.s }, p5/Z, [x8]\n" + "ld1rqw { z8.s }, p5/Z, [x8, #16]\n" + "ld1rqw { z15.s }, p5/Z, [x8, #32]\n" "fmul z11.s, z9.s, z6.s[0]\n" "fmul z5.s, z9.s, z6.s[1]\n" - "ld1rqw { z4.s }, p5/Z, [x19, #48]\n" - "ld1rqw { z3.s }, p5/Z, [x19, #64]\n" - "ld1rqw { z2.s }, p5/Z, [x19, #80]\n" + "ld1rqw { z4.s }, p5/Z, [x8, #48]\n" + "ld1rqw { z3.s }, p5/Z, [x8, #64]\n" + "ld1rqw { z2.s }, p5/Z, [x8, #80]\n" "cbz %x[bptr], 1f\n" "ptrue p8.s\n" "1:" // Set bias predicate: Done @@ -177,7 +177,7 @@ void sme_fp32_mopa_4x4_3x3( "add x20, %x[output], %x[output_col_stride], LSL #2\n" ".inst 0x809db562 // fmopa za2.s, p5/M, p5/M, z11.s, z29.s\n" "ld1w { z30.s }, p3/Z, [x21, x24, LSL #2]\n" - "add x19, %x[output], %x[output_row_stride], LSL #2\n" + "add x8, %x[output], %x[output_row_stride], LSL #2\n" ".inst 0x809cb563 // fmopa za3.s, p5/M, p5/M, z11.s, z28.s\n" "fmul z11.s, z9.s, z6.s[2]\n" "ld1w { z29.s }, p2/Z, [x21, x23, LSL #2]\n" @@ -567,25 +567,25 @@ void sme_fp32_mopa_4x4_3x3( "add x20, x20, %x[output_col_stride], LSL #2\n" "fmax z17.s, p5/M, z17.s, z12.s\n" "st1w { z28.s }, p0, [x20, x25, LSL #2]\n" - "add x20, x19, %x[output_col_stride], LSL #2\n" + "add x20, x8, %x[output_col_stride], LSL #2\n" "fmax z16.s, p5/M, z16.s, z12.s\n" - "st1w { z27.s }, p0, [x19, x25, LSL #2]\n" - "add x19, x19, %x[output_row_stride], LSL #2\n" + "st1w { z27.s }, p0, [x8, x25, LSL #2]\n" + "add x8, x8, %x[output_row_stride], LSL #2\n" "st1w { z26.s }, p0, [x20, x25, LSL #2]\n" "add x20, x20, %x[output_col_stride], LSL #2\n" "st1w { z25.s }, p0, [x20, x25, LSL #2]\n" "add x20, x20, %x[output_col_stride], LSL #2\n" "st1w { z24.s }, p0, [x20, x25, LSL #2]\n" - "add x20, x19, %x[output_col_stride], LSL #2\n" - "st1w { z23.s }, p0, [x19, x25, LSL #2]\n" - "add x19, x19, %x[output_row_stride], LSL #2\n" + "add x20, x8, %x[output_col_stride], LSL #2\n" + "st1w { z23.s }, p0, [x8, x25, LSL #2]\n" + "add x8, x8, %x[output_row_stride], LSL #2\n" "st1w { z22.s }, p0, [x20, x25, LSL #2]\n" "add x20, x20, %x[output_col_stride], LSL #2\n" "st1w { z21.s }, p0, [x20, x25, LSL #2]\n" "add x20, x20, %x[output_col_stride], LSL #2\n" "st1w { z20.s }, p0, [x20, x25, LSL #2]\n" - "add x20, x19, %x[output_col_stride], LSL #2\n" - "st1w { z19.s }, p0, [x19, x25, LSL #2]\n" + "add x20, x8, %x[output_col_stride], LSL #2\n" + "st1w { z19.s }, p0, [x8, x25, LSL #2]\n" "st1w { z18.s }, p0, [x20, x25, LSL #2]\n" "add x20, x20, %x[output_col_stride], LSL #2\n" "st1w { z17.s }, p0, [x20, x25, LSL #2]\n" @@ -624,7 +624,7 @@ void sme_fp32_mopa_4x4_3x3( ".inst 0xc08254d4 // mova z20.s, p5/M, za1h.s[x14, #2]\n" "fmax z25.s, p5/M, z25.s, z12.s\n" "fmin z21.s, p5/M, z21.s, z10.s\n" - "add x19, %x[output], %x[output_row_stride], LSL #2\n" + "add x8, %x[output], %x[output_row_stride], LSL #2\n" ".inst 0xc08274f3 // mova z19.s, p5/M, za1h.s[XZR, #3]\n" "fmax z24.s, p5/M, z24.s, z12.s\n" "fmin z20.s, p5/M, z20.s, z10.s\n" @@ -648,25 +648,25 @@ void sme_fp32_mopa_4x4_3x3( "add x20, x20, %x[output_col_stride], LSL #2\n" "fmax z17.s, p5/M, z17.s, z12.s\n" "st1w { z28.s }, p0, [x20, x24, LSL #2]\n" - "add x20, x19, %x[output_col_stride], LSL #2\n" + "add x20, x8, %x[output_col_stride], LSL #2\n" "fmax z16.s, p5/M, z16.s, z12.s\n" - "st1w { z27.s }, p0, [x19, x24, LSL #2]\n" - "add x19, x19, %x[output_row_stride], LSL #2\n" + "st1w { z27.s }, p0, [x8, x24, LSL #2]\n" + "add x8, x8, %x[output_row_stride], LSL #2\n" "st1w { z26.s }, p0, [x20, x24, LSL #2]\n" "add x20, x20, %x[output_col_stride], LSL #2\n" "st1w { z25.s }, p0, [x20, x24, LSL #2]\n" "add x20, x20, %x[output_col_stride], LSL #2\n" "st1w { z24.s }, p0, [x20, x24, LSL #2]\n" - "add x20, x19, %x[output_col_stride], LSL #2\n" - "st1w { z23.s }, p0, [x19, x24, LSL #2]\n" - "add x19, x19, %x[output_row_stride], LSL #2\n" + "add x20, x8, %x[output_col_stride], LSL #2\n" + "st1w { z23.s }, p0, [x8, x24, LSL #2]\n" + "add x8, x8, %x[output_row_stride], LSL #2\n" "st1w { z22.s }, p0, [x20, x24, LSL #2]\n" "add x20, x20, %x[output_col_stride], LSL #2\n" "st1w { z21.s }, p0, [x20, x24, LSL #2]\n" "add x20, x20, %x[output_col_stride], LSL #2\n" "st1w { z20.s }, p0, [x20, x24, LSL #2]\n" - "add x20, x19, %x[output_col_stride], LSL #2\n" - "st1w { z19.s }, p0, [x19, x24, LSL #2]\n" + "add x20, x8, %x[output_col_stride], LSL #2\n" + "st1w { z19.s }, p0, [x8, x24, LSL #2]\n" "st1w { z18.s }, p0, [x20, x24, LSL #2]\n" "add x20, x20, %x[output_col_stride], LSL #2\n" "st1w { z17.s }, p0, [x20, x24, LSL #2]\n" @@ -705,7 +705,7 @@ void sme_fp32_mopa_4x4_3x3( ".inst 0xc0825554 // mova z20.s, p5/M, za2h.s[x14, #2]\n" "fmax z25.s, p5/M, z25.s, z12.s\n" "fmin z21.s, p5/M, z21.s, z10.s\n" - "add x19, %x[output], %x[output_row_stride], LSL #2\n" + "add x8, %x[output], %x[output_row_stride], LSL #2\n" ".inst 0xc0827573 // mova z19.s, p5/M, za2h.s[XZR, #3]\n" "fmax z24.s, p5/M, z24.s, z12.s\n" "fmin z20.s, p5/M, z20.s, z10.s\n" @@ -729,25 +729,25 @@ void sme_fp32_mopa_4x4_3x3( "add x20, x20, %x[output_col_stride], LSL #2\n" "fmax z17.s, p5/M, z17.s, z12.s\n" "st1w { z28.s }, p0, [x20, x23, LSL #2]\n" - "add x20, x19, %x[output_col_stride], LSL #2\n" + "add x20, x8, %x[output_col_stride], LSL #2\n" "fmax z16.s, p5/M, z16.s, z12.s\n" - "st1w { z27.s }, p0, [x19, x23, LSL #2]\n" - "add x19, x19, %x[output_row_stride], LSL #2\n" + "st1w { z27.s }, p0, [x8, x23, LSL #2]\n" + "add x8, x8, %x[output_row_stride], LSL #2\n" "st1w { z26.s }, p0, [x20, x23, LSL #2]\n" "add x20, x20, %x[output_col_stride], LSL #2\n" "st1w { z25.s }, p0, [x20, x23, LSL #2]\n" "add x20, x20, %x[output_col_stride], LSL #2\n" "st1w { z24.s }, p0, [x20, x23, LSL #2]\n" - "add x20, x19, %x[output_col_stride], LSL #2\n" - "st1w { z23.s }, p0, [x19, x23, LSL #2]\n" - "add x19, x19, %x[output_row_stride], LSL #2\n" + "add x20, x8, %x[output_col_stride], LSL #2\n" + "st1w { z23.s }, p0, [x8, x23, LSL #2]\n" + "add x8, x8, %x[output_row_stride], LSL #2\n" "st1w { z22.s }, p0, [x20, x23, LSL #2]\n" "add x20, x20, %x[output_col_stride], LSL #2\n" "st1w { z21.s }, p0, [x20, x23, LSL #2]\n" "add x20, x20, %x[output_col_stride], LSL #2\n" "st1w { z20.s }, p0, [x20, x23, LSL #2]\n" - "add x20, x19, %x[output_col_stride], LSL #2\n" - "st1w { z19.s }, p0, [x19, x23, LSL #2]\n" + "add x20, x8, %x[output_col_stride], LSL #2\n" + "st1w { z19.s }, p0, [x8, x23, LSL #2]\n" "st1w { z18.s }, p0, [x20, x23, LSL #2]\n" "add x20, x20, %x[output_col_stride], LSL #2\n" "st1w { z17.s }, p0, [x20, x23, LSL #2]\n" @@ -790,7 +790,7 @@ void sme_fp32_mopa_4x4_3x3( ".inst 0xc08275f3 // mova z19.s, p5/M, za3h.s[XZR, #3]\n" "fmax z24.s, p5/M, z24.s, z12.s\n" "fmin z20.s, p5/M, z20.s, z10.s\n" - "add x19, %x[output], %x[output_row_stride], LSL #2\n" + "add x8, %x[output], %x[output_row_stride], LSL #2\n" ".inst 0xc08235f2 // mova z18.s, p5/M, za3h.s[x13, #3]\n" "fmax z23.s, p5/M, z23.s, z12.s\n" "fmin z19.s, p5/M, z19.s, z10.s\n" @@ -817,24 +817,24 @@ void sme_fp32_mopa_4x4_3x3( "add x20, x20, %x[output_col_stride], LSL #2\n" "fmax z16.s, p5/M, z16.s, z12.s\n" "st1w { z28.s }, p0, [x20, x22, LSL #2]\n" - "add x20, x19, %x[output_col_stride], LSL #2\n" - "st1w { z27.s }, p0, [x19, x22, LSL #2]\n" - "add x19, x19, %x[output_row_stride], LSL #2\n" + "add x20, x8, %x[output_col_stride], LSL #2\n" + "st1w { z27.s }, p0, [x8, x22, LSL #2]\n" + "add x8, x8, %x[output_row_stride], LSL #2\n" "st1w { z26.s }, p0, [x20, x22, LSL #2]\n" "add x20, x20, %x[output_col_stride], LSL #2\n" "st1w { z25.s }, p0, [x20, x22, LSL #2]\n" "add x20, x20, %x[output_col_stride], LSL #2\n" "st1w { z24.s }, p0, [x20, x22, LSL #2]\n" - "add x20, x19, %x[output_col_stride], LSL #2\n" - "st1w { z23.s }, p0, [x19, x22, LSL #2]\n" - "add x19, x19, %x[output_row_stride], LSL #2\n" + "add x20, x8, %x[output_col_stride], LSL #2\n" + "st1w { z23.s }, p0, [x8, x22, LSL #2]\n" + "add x8, x8, %x[output_row_stride], LSL #2\n" "st1w { z22.s }, p0, [x20, x22, LSL #2]\n" "add x20, x20, %x[output_col_stride], LSL #2\n" "st1w { z21.s }, p0, [x20, x22, LSL #2]\n" "add x20, x20, %x[output_col_stride], LSL #2\n" "st1w { z20.s }, p0, [x20, x22, LSL #2]\n" - "add x20, x19, %x[output_col_stride], LSL #2\n" - "st1w { z19.s }, p0, [x19, x22, LSL #2]\n" + "add x20, x8, %x[output_col_stride], LSL #2\n" + "st1w { z19.s }, p0, [x8, x22, LSL #2]\n" "st1w { z18.s }, p0, [x20, x22, LSL #2]\n" "add x20, x20, %x[output_col_stride], LSL #2\n" "st1w { z17.s }, p0, [x20, x22, LSL #2]\n" @@ -880,7 +880,7 @@ void sme_fp32_mopa_4x4_3x3( ".inst 0xd503467f // SMSTOP\n" : : [bptr] "r" (bptr), [inptr] "r" (inptr), [matrix_stride] "r" (matrix_stride), [n_channels] "r" (n_channels), [offsetof_Params_act_max] "I" (offsetof(Params, act_max)), [offsetof_Params_act_min] "I" (offsetof(Params, act_min)), [offsetof_Params_inner_terms] "I" (offsetof(Params, inner_terms)), [offsetof_Params_outer_terms] "I" (offsetof(Params, outer_terms)), [output] "r" (output), [output_col_stride] "r" (output_col_stride), [output_row_stride] "r" (output_row_stride), [params] "r" (¶ms) - : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p8", "x12", "x13", "x14", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "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", "p8", "x12", "x13", "x14", "x8", "x20", "x21", "x22", "x23", "x24", "x25", "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" ); } |