diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_conv/pooling/kernels/sve_s8q_nhwc_avg_generic_depthfirst/generic.cpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_conv/pooling/kernels/sve_s8q_nhwc_avg_generic_depthfirst/generic.cpp | 136 |
1 files changed, 68 insertions, 68 deletions
diff --git a/src/core/NEON/kernels/arm_conv/pooling/kernels/sve_s8q_nhwc_avg_generic_depthfirst/generic.cpp b/src/core/NEON/kernels/arm_conv/pooling/kernels/sve_s8q_nhwc_avg_generic_depthfirst/generic.cpp index 7161c4f389..19a3b112ad 100644 --- a/src/core/NEON/kernels/arm_conv/pooling/kernels/sve_s8q_nhwc_avg_generic_depthfirst/generic.cpp +++ b/src/core/NEON/kernels/arm_conv/pooling/kernels/sve_s8q_nhwc_avg_generic_depthfirst/generic.cpp @@ -128,7 +128,7 @@ void sve_s8q_nhwc_avg_generic_depthfirst_impl( "lsr x23, %x[n_valid_cells], #0x1\n" "mov z15.s, #0x0\n" "mov z14.s, #0x0\n" - "mov x20, %x[inptrs]\n" + "mov x22, %x[inptrs]\n" "mov z13.s, #0x0\n" "mov z12.s, #0x0\n" "mov z11.s, #0x0\n" @@ -144,42 +144,42 @@ void sve_s8q_nhwc_avg_generic_depthfirst_impl( "mov z1.s, #0x0\n" "mov z0.s, #0x0\n" "cbz x23, 4f\n" - "ldp x22, x21, [x20, #0x0]\n" + "ldp x21, x20, [x22, #0x0]\n" "subs x23, x23, #0x1\n" - "add x20, x20, #0x10\n" - "ld1b { z31.b }, p4/Z, [x22, x27]\n" - "ld1b { z30.b }, p4/Z, [x21, x27]\n" - "ld1b { z29.b }, p3/Z, [x22, x26]\n" - "ld1b { z28.b }, p3/Z, [x21, x26]\n" - "ld1b { z27.b }, p2/Z, [x22, x25]\n" - "ld1b { z26.b }, p2/Z, [x21, x25]\n" - "ld1b { z25.b }, p1/Z, [x22, x24]\n" - "ld1b { z24.b }, p1/Z, [x21, x24]\n" + "add x22, x22, #0x10\n" + "ld1b { z31.b }, p4/Z, [x21, x27]\n" + "ld1b { z30.b }, p4/Z, [x20, x27]\n" + "ld1b { z29.b }, p3/Z, [x21, x26]\n" + "ld1b { z28.b }, p3/Z, [x20, x26]\n" + "ld1b { z27.b }, p2/Z, [x21, x25]\n" + "ld1b { z26.b }, p2/Z, [x20, x25]\n" + "ld1b { z25.b }, p1/Z, [x21, x24]\n" + "ld1b { z24.b }, p1/Z, [x20, x24]\n" "beq 3f\n" "2:" // 4-vectors of channels: 2 inputs loop ".inst 0x455e03f7 // saddlb z23.h, z31.b, z30.b\n" ".inst 0x455e07f6 // saddlt z22.h, z31.b, z30.b\n" - "ldp x22, x21, [x20, #0x0]\n" + "ldp x21, x20, [x22, #0x0]\n" "subs x23, x23, #0x1\n" ".inst 0x455c03b5 // saddlb z21.h, z29.b, z28.b\n" ".inst 0x455c07b4 // saddlt z20.h, z29.b, z28.b\n" - "add x20, x20, #0x10\n" - "ld1b { z31.b }, p4/Z, [x22, x27]\n" + "add x22, x22, #0x10\n" + "ld1b { z31.b }, p4/Z, [x21, x27]\n" ".inst 0x455a0373 // saddlb z19.h, z27.b, z26.b\n" ".inst 0x455a0772 // saddlt z18.h, z27.b, z26.b\n" - "ld1b { z30.b }, p4/Z, [x21, x27]\n" - "ld1b { z29.b }, p3/Z, [x22, x26]\n" + "ld1b { z30.b }, p4/Z, [x20, x27]\n" + "ld1b { z29.b }, p3/Z, [x21, x26]\n" ".inst 0x45580331 // saddlb z17.h, z25.b, z24.b\n" ".inst 0x45580730 // saddlt z16.h, z25.b, z24.b\n" - "ld1b { z28.b }, p3/Z, [x21, x26]\n" - "ld1b { z27.b }, p2/Z, [x22, x25]\n" + "ld1b { z28.b }, p3/Z, [x20, x26]\n" + "ld1b { z27.b }, p2/Z, [x21, x25]\n" ".inst 0x459741ef // saddwb z15.s, z15.s, z23.h\n" ".inst 0x459745ce // saddwt z14.s, z14.s, z23.h\n" - "ld1b { z26.b }, p2/Z, [x21, x25]\n" - "ld1b { z25.b }, p1/Z, [x22, x24]\n" + "ld1b { z26.b }, p2/Z, [x20, x25]\n" + "ld1b { z25.b }, p1/Z, [x21, x24]\n" ".inst 0x459641ad // saddwb z13.s, z13.s, z22.h\n" ".inst 0x4596458c // saddwt z12.s, z12.s, z22.h\n" - "ld1b { z24.b }, p1/Z, [x21, x24]\n" + "ld1b { z24.b }, p1/Z, [x20, x24]\n" ".inst 0x4595416b // saddwb z11.s, z11.s, z21.h\n" ".inst 0x4595454a // saddwt z10.s, z10.s, z21.h\n" ".inst 0x45944129 // saddwb z9.s, z9.s, z20.h\n" @@ -222,20 +222,20 @@ void sve_s8q_nhwc_avg_generic_depthfirst_impl( "ands x21, %x[n_valid_cells], #0x1\n" "beq 6f\n" "5:" // 4-vectors of channels: Single input loop - "ldr x22, [x20], #0x8\n" - "ld1b { z31.b }, p4/Z, [x22, x27]\n" - ".inst 0x4508a3f7 // sshllb z23.h, z31.b, #0x0\n" - ".inst 0x4508a7f6 // sshllt z22.h, z31.b, #0x0\n" - "ld1b { z29.b }, p3/Z, [x22, x26]\n" - "ld1b { z27.b }, p2/Z, [x22, x25]\n" - ".inst 0x4508a3b5 // sshllb z21.h, z29.b, #0x0\n" - ".inst 0x4508a7b4 // sshllt z20.h, z29.b, #0x0\n" - "ld1b { z25.b }, p1/Z, [x22, x24]\n" - ".inst 0x4508a373 // sshllb z19.h, z27.b, #0x0\n" - ".inst 0x4508a772 // sshllt z18.h, z27.b, #0x0\n" + "ldr x20, [x22], #0x8\n" + "ld1b { z16.b }, p4/Z, [x20, x27]\n" + ".inst 0x4508a217 // sshllb z23.h, z16.b, #0x0\n" + ".inst 0x4508a616 // sshllt z22.h, z16.b, #0x0\n" + "ld1b { z16.b }, p3/Z, [x20, x26]\n" + "ld1b { z17.b }, p2/Z, [x20, x25]\n" + ".inst 0x4508a215 // sshllb z21.h, z16.b, #0x0\n" + ".inst 0x4508a614 // sshllt z20.h, z16.b, #0x0\n" + "ld1b { z16.b }, p1/Z, [x20, x24]\n" + ".inst 0x4508a233 // sshllb z19.h, z17.b, #0x0\n" + ".inst 0x4508a632 // sshllt z18.h, z17.b, #0x0\n" "subs x21, x21, #0x1\n" - ".inst 0x4508a331 // sshllb z17.h, z25.b, #0x0\n" - ".inst 0x4508a730 // sshllt z16.h, z25.b, #0x0\n" + ".inst 0x4508a211 // sshllb z17.h, z16.b, #0x0\n" + ".inst 0x4508a610 // sshllt z16.h, z16.b, #0x0\n" ".inst 0x459741ef // saddwb z15.s, z15.s, z23.h\n" ".inst 0x459745ce // saddwt z14.s, z14.s, z23.h\n" ".inst 0x459641ad // saddwb z13.s, z13.s, z22.h\n" @@ -368,57 +368,57 @@ void sve_s8q_nhwc_avg_generic_depthfirst_impl( "lsr x23, %x[n_valid_cells], #0x1\n" "mov z15.s, #0x0\n" "mov z14.s, #0x0\n" - "mov x20, %x[inptrs]\n" + "mov x22, %x[inptrs]\n" "mov z13.s, #0x0\n" "mov z12.s, #0x0\n" "cbz x23, 11f\n" - "ldp x22, x21, [x20, #0x0]\n" + "ldp x21, x20, [x22, #0x0]\n" "subs x23, x23, #0x1\n" - "add x20, x20, #0x10\n" - "ld1b { z31.b }, p4/Z, [x22, x27]\n" - "ld1b { z30.b }, p4/Z, [x21, x27]\n" + "add x22, x22, #0x10\n" + "ld1b { z31.b }, p4/Z, [x21, x27]\n" + "ld1b { z30.b }, p4/Z, [x20, x27]\n" "beq 10f\n" "9:" // Single vector of channels: Loop: 2 inputs loop - ".inst 0x455e03f7 // saddlb z23.h, z31.b, z30.b\n" - ".inst 0x455e07f6 // saddlt z22.h, z31.b, z30.b\n" - "ldp x22, x21, [x20, #0x0]\n" + ".inst 0x455e03f1 // saddlb z17.h, z31.b, z30.b\n" + ".inst 0x455e07f0 // saddlt z16.h, z31.b, z30.b\n" + "ldp x21, x20, [x22, #0x0]\n" "subs x23, x23, #0x1\n" - ".inst 0x459741ef // saddwb z15.s, z15.s, z23.h\n" - ".inst 0x459745ce // saddwt z14.s, z14.s, z23.h\n" - "add x20, x20, #0x10\n" - "ld1b { z31.b }, p4/Z, [x22, x27]\n" - ".inst 0x459641ad // saddwb z13.s, z13.s, z22.h\n" - ".inst 0x4596458c // saddwt z12.s, z12.s, z22.h\n" - "ld1b { z30.b }, p4/Z, [x21, x27]\n" + ".inst 0x459141ef // saddwb z15.s, z15.s, z17.h\n" + ".inst 0x459145ce // saddwt z14.s, z14.s, z17.h\n" + "add x22, x22, #0x10\n" + "ld1b { z31.b }, p4/Z, [x21, x27]\n" + ".inst 0x459041ad // saddwb z13.s, z13.s, z16.h\n" + ".inst 0x4590458c // saddwt z12.s, z12.s, z16.h\n" + "ld1b { z30.b }, p4/Z, [x20, x27]\n" "bgt 9b\n" "10:" // Single vector of channels: Loop: 2 inputs tail - ".inst 0x455e03f7 // saddlb z23.h, z31.b, z30.b\n" - ".inst 0x455e07f6 // saddlt z22.h, z31.b, z30.b\n" - ".inst 0x459741ef // saddwb z15.s, z15.s, z23.h\n" - ".inst 0x459745ce // saddwt z14.s, z14.s, z23.h\n" - ".inst 0x459641ad // saddwb z13.s, z13.s, z22.h\n" - ".inst 0x4596458c // saddwt z12.s, z12.s, z22.h\n" + ".inst 0x455e03f1 // saddlb z17.h, z31.b, z30.b\n" + ".inst 0x455e07f0 // saddlt z16.h, z31.b, z30.b\n" + ".inst 0x459141ef // saddwb z15.s, z15.s, z17.h\n" + ".inst 0x459145ce // saddwt z14.s, z14.s, z17.h\n" + ".inst 0x459041ad // saddwb z13.s, z13.s, z16.h\n" + ".inst 0x4590458c // saddwt z12.s, z12.s, z16.h\n" "11:" // Single vector of channels: Loop: After loop "ands x21, %x[n_valid_cells], #0x1\n" "beq 13f\n" "12:" // Single vector of channels: Loop: Single input loop - "ldr x22, [x20], #0x8\n" - "ld1b { z31.b }, p4/Z, [x22, x27]\n" - ".inst 0x4508a3f7 // sshllb z23.h, z31.b, #0x0\n" - ".inst 0x4508a7f6 // sshllt z22.h, z31.b, #0x0\n" + "ldr x20, [x22], #0x8\n" + "ld1b { z16.b }, p4/Z, [x20, x27]\n" + ".inst 0x4508a211 // sshllb z17.h, z16.b, #0x0\n" + ".inst 0x4508a610 // sshllt z16.h, z16.b, #0x0\n" "subs x21, x21, #0x1\n" - ".inst 0x459741ef // saddwb z15.s, z15.s, z23.h\n" - ".inst 0x459745ce // saddwt z14.s, z14.s, z23.h\n" - ".inst 0x459641ad // saddwb z13.s, z13.s, z22.h\n" - ".inst 0x4596458c // saddwt z12.s, z12.s, z22.h\n" + ".inst 0x459141ef // saddwb z15.s, z15.s, z17.h\n" + ".inst 0x459145ce // saddwt z14.s, z14.s, z17.h\n" + ".inst 0x459041ad // saddwb z13.s, z13.s, z16.h\n" + ".inst 0x4590458c // saddwt z12.s, z12.s, z16.h\n" "bgt 12b\n" "13:" // Single vector of channels: Loop: Single input loop: End - "ld1rw { z18.s }, p0/Z, [%x[left_shift]]\n" + "ld1rw { z16.s }, p0/Z, [%x[left_shift]]\n" "ld1rw { z17.s }, p0/Z, [%x[combined_rescale_value]]\n" - ".inst 0x4482824f // srshl z15.s, p0/M, z15.s, z18.s\n" - ".inst 0x4482824e // srshl z14.s, p0/M, z14.s, z18.s\n" - ".inst 0x4482824d // srshl z13.s, p0/M, z13.s, z18.s\n" - ".inst 0x4482824c // srshl z12.s, p0/M, z12.s, z18.s\n" + ".inst 0x4482820f // srshl z15.s, p0/M, z15.s, z16.s\n" + ".inst 0x4482820e // srshl z14.s, p0/M, z14.s, z16.s\n" + ".inst 0x4482820d // srshl z13.s, p0/M, z13.s, z16.s\n" + ".inst 0x4482820c // srshl z12.s, p0/M, z12.s, z16.s\n" "ld1rw { z16.s }, p0/Z, [%x[right_shift]]\n" ".inst 0x04b175ef // sqrdmulh z15.s, z15.s, z17.s\n" ".inst 0x04b175ce // sqrdmulh z14.s, z14.s, z17.s\n" |