diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_conv/pooling/kernels/sve_fp16_nhwc_avg_generic_depthfirst/generic.cpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_conv/pooling/kernels/sve_fp16_nhwc_avg_generic_depthfirst/generic.cpp | 148 |
1 files changed, 74 insertions, 74 deletions
diff --git a/src/core/NEON/kernels/arm_conv/pooling/kernels/sve_fp16_nhwc_avg_generic_depthfirst/generic.cpp b/src/core/NEON/kernels/arm_conv/pooling/kernels/sve_fp16_nhwc_avg_generic_depthfirst/generic.cpp index 594c65e18d..2bef44ea5c 100644 --- a/src/core/NEON/kernels/arm_conv/pooling/kernels/sve_fp16_nhwc_avg_generic_depthfirst/generic.cpp +++ b/src/core/NEON/kernels/arm_conv/pooling/kernels/sve_fp16_nhwc_avg_generic_depthfirst/generic.cpp @@ -57,68 +57,68 @@ void sve_fp16_nhwc_avg_generic_depthfirst_impl( "lsr x25, %x[n_valid_cells], #0x2\n" "mov z6.b, #0x0\n" "mov z5.b, #0x0\n" - "mov x20, %x[inptrs]\n" + "mov x24, %x[inptrs]\n" "mov z4.b, #0x0\n" "mov z3.b, #0x0\n" "cbz x25, 4f\n" - "ldp x24, x23, [x20, #0x0]\n" - "ldp x22, x21, [x20, #0x10]\n" + "ldp x23, x22, [x24, #0x0]\n" + "ldp x21, x20, [x24, #0x10]\n" "subs x25, x25, #0x1\n" - "add x20, x20, #0x20\n" - "ld1h { z2.h }, p3/Z, [x24, x9, LSL #1]\n" - "ld1h { z1.h }, p3/Z, [x23, x9, LSL #1]\n" - "ld1h { z0.h }, p3/Z, [x22, x9, LSL #1]\n" - "ld1h { z31.h }, p3/Z, [x21, x9, LSL #1]\n" - "ld1h { z30.h }, p2/Z, [x24, x28, LSL #1]\n" - "ld1h { z22.h }, p2/Z, [x23, x28, LSL #1]\n" - "ld1h { z29.h }, p2/Z, [x22, x28, LSL #1]\n" - "ld1h { z28.h }, p2/Z, [x21, x28, LSL #1]\n" - "ld1h { z27.h }, p1/Z, [x24, x27, LSL #1]\n" - "ld1h { z21.h }, p1/Z, [x23, x27, LSL #1]\n" - "ld1h { z26.h }, p1/Z, [x22, x27, LSL #1]\n" - "ld1h { z17.h }, p1/Z, [x21, x27, LSL #1]\n" - "ld1h { z25.h }, p0/Z, [x24, x26, LSL #1]\n" - "ld1h { z20.h }, p0/Z, [x23, x26, LSL #1]\n" - "ld1h { z24.h }, p0/Z, [x22, x26, LSL #1]\n" - "ld1h { z16.h }, p0/Z, [x21, x26, LSL #1]\n" + "add x24, x24, #0x20\n" + "ld1h { z2.h }, p3/Z, [x23, x9, LSL #1]\n" + "ld1h { z1.h }, p3/Z, [x22, x9, LSL #1]\n" + "ld1h { z0.h }, p3/Z, [x21, x9, LSL #1]\n" + "ld1h { z31.h }, p3/Z, [x20, x9, LSL #1]\n" + "ld1h { z30.h }, p2/Z, [x23, x28, LSL #1]\n" + "ld1h { z22.h }, p2/Z, [x22, x28, LSL #1]\n" + "ld1h { z29.h }, p2/Z, [x21, x28, LSL #1]\n" + "ld1h { z28.h }, p2/Z, [x20, x28, LSL #1]\n" + "ld1h { z27.h }, p1/Z, [x23, x27, LSL #1]\n" + "ld1h { z21.h }, p1/Z, [x22, x27, LSL #1]\n" + "ld1h { z26.h }, p1/Z, [x21, x27, LSL #1]\n" + "ld1h { z17.h }, p1/Z, [x20, x27, LSL #1]\n" + "ld1h { z25.h }, p0/Z, [x23, x26, LSL #1]\n" + "ld1h { z20.h }, p0/Z, [x22, x26, LSL #1]\n" + "ld1h { z24.h }, p0/Z, [x21, x26, LSL #1]\n" + "ld1h { z16.h }, p0/Z, [x20, x26, LSL #1]\n" "beq 3f\n" "2:" // 4-vectors of channels: 4 inputs loop "fadd z23.h, z2.h, z1.h\n" "fadd z19.h, z0.h, z31.h\n" - "ldp x24, x23, [x20, #0x0]\n" - "ldp x22, x21, [x20, #0x10]\n" + "ldp x23, x22, [x24, #0x0]\n" + "ldp x21, x20, [x24, #0x10]\n" "fadd z22.h, z30.h, z22.h\n" "fadd z18.h, z29.h, z28.h\n" "subs x25, x25, #0x1\n" - "add x20, x20, #0x20\n" + "add x24, x24, #0x20\n" "fadd z21.h, z27.h, z21.h\n" "fadd z17.h, z26.h, z17.h\n" - "ld1h { z2.h }, p3/Z, [x24, x9, LSL #1]\n" - "ld1h { z1.h }, p3/Z, [x23, x9, LSL #1]\n" + "ld1h { z2.h }, p3/Z, [x23, x9, LSL #1]\n" + "ld1h { z1.h }, p3/Z, [x22, x9, LSL #1]\n" "fadd z20.h, z25.h, z20.h\n" "fadd z16.h, z24.h, z16.h\n" - "ld1h { z0.h }, p3/Z, [x22, x9, LSL #1]\n" - "ld1h { z31.h }, p3/Z, [x21, x9, LSL #1]\n" + "ld1h { z0.h }, p3/Z, [x21, x9, LSL #1]\n" + "ld1h { z31.h }, p3/Z, [x20, x9, LSL #1]\n" "fadd z19.h, z23.h, z19.h\n" "fadd z18.h, z22.h, z18.h\n" - "ld1h { z30.h }, p2/Z, [x24, x28, LSL #1]\n" - "ld1h { z22.h }, p2/Z, [x23, x28, LSL #1]\n" + "ld1h { z30.h }, p2/Z, [x23, x28, LSL #1]\n" + "ld1h { z22.h }, p2/Z, [x22, x28, LSL #1]\n" "fadd z17.h, z21.h, z17.h\n" "fadd z16.h, z20.h, z16.h\n" - "ld1h { z29.h }, p2/Z, [x22, x28, LSL #1]\n" - "ld1h { z28.h }, p2/Z, [x21, x28, LSL #1]\n" + "ld1h { z29.h }, p2/Z, [x21, x28, LSL #1]\n" + "ld1h { z28.h }, p2/Z, [x20, x28, LSL #1]\n" "fadd z6.h, z6.h, z19.h\n" "fadd z5.h, z5.h, z18.h\n" - "ld1h { z27.h }, p1/Z, [x24, x27, LSL #1]\n" - "ld1h { z21.h }, p1/Z, [x23, x27, LSL #1]\n" + "ld1h { z27.h }, p1/Z, [x23, x27, LSL #1]\n" + "ld1h { z21.h }, p1/Z, [x22, x27, LSL #1]\n" "fadd z4.h, z4.h, z17.h\n" "fadd z3.h, z3.h, z16.h\n" - "ld1h { z26.h }, p1/Z, [x22, x27, LSL #1]\n" - "ld1h { z17.h }, p1/Z, [x21, x27, LSL #1]\n" - "ld1h { z25.h }, p0/Z, [x24, x26, LSL #1]\n" - "ld1h { z20.h }, p0/Z, [x23, x26, LSL #1]\n" - "ld1h { z24.h }, p0/Z, [x22, x26, LSL #1]\n" - "ld1h { z16.h }, p0/Z, [x21, x26, LSL #1]\n" + "ld1h { z26.h }, p1/Z, [x21, x27, LSL #1]\n" + "ld1h { z17.h }, p1/Z, [x20, x27, LSL #1]\n" + "ld1h { z25.h }, p0/Z, [x23, x26, LSL #1]\n" + "ld1h { z20.h }, p0/Z, [x22, x26, LSL #1]\n" + "ld1h { z24.h }, p0/Z, [x21, x26, LSL #1]\n" + "ld1h { z16.h }, p0/Z, [x20, x26, LSL #1]\n" "bgt 2b\n" "3:" // 4-vectors of channels: 4 inputs tail "fadd z23.h, z2.h, z1.h\n" @@ -141,16 +141,16 @@ void sve_fp16_nhwc_avg_generic_depthfirst_impl( "ands x21, %x[n_valid_cells], #0x3\n" "beq 6f\n" "5:" // 4-vectors of channels: Single input loop - "ldr x24, [x20], #0x8\n" - "ld1h { z2.h }, p3/Z, [x24, x9, LSL #1]\n" + "ldr x20, [x24], #0x8\n" + "ld1h { z16.h }, p3/Z, [x20, x9, LSL #1]\n" "subs x21, x21, #0x1\n" - "fadd z6.h, z6.h, z2.h\n" - "ld1h { z30.h }, p2/Z, [x24, x28, LSL #1]\n" - "ld1h { z27.h }, p1/Z, [x24, x27, LSL #1]\n" - "fadd z5.h, z5.h, z30.h\n" - "fadd z4.h, z4.h, z27.h\n" - "ld1h { z25.h }, p0/Z, [x24, x26, LSL #1]\n" - "fadd z3.h, z3.h, z25.h\n" + "fadd z6.h, z6.h, z16.h\n" + "ld1h { z17.h }, p2/Z, [x20, x28, LSL #1]\n" + "ld1h { z16.h }, p1/Z, [x20, x27, LSL #1]\n" + "fadd z5.h, z5.h, z17.h\n" + "fadd z4.h, z4.h, z16.h\n" + "ld1h { z16.h }, p0/Z, [x20, x26, LSL #1]\n" + "fadd z3.h, z3.h, z16.h\n" "bgt 5b\n" "6:" // 4-vectors of channels: Single input loop: End "fmul z6.h, z6.h, z7.h\n" @@ -173,44 +173,44 @@ void sve_fp16_nhwc_avg_generic_depthfirst_impl( "8:" // Single vector of channels: Loop "lsr x25, %x[n_valid_cells], #0x2\n" "mov z6.b, #0x0\n" - "mov x20, %x[inptrs]\n" + "mov x24, %x[inptrs]\n" "cbz x25, 11f\n" - "ldp x24, x23, [x20, #0x0]\n" - "ldp x22, x21, [x20, #0x10]\n" + "ldp x23, x22, [x24, #0x0]\n" + "ldp x21, x20, [x24, #0x10]\n" "subs x25, x25, #0x1\n" - "add x20, x20, #0x20\n" - "ld1h { z2.h }, p3/Z, [x24, x9, LSL #1]\n" - "ld1h { z1.h }, p3/Z, [x23, x9, LSL #1]\n" - "ld1h { z0.h }, p3/Z, [x22, x9, LSL #1]\n" - "ld1h { z31.h }, p3/Z, [x21, x9, LSL #1]\n" + "add x24, x24, #0x20\n" + "ld1h { z2.h }, p3/Z, [x23, x9, LSL #1]\n" + "ld1h { z1.h }, p3/Z, [x22, x9, LSL #1]\n" + "ld1h { z0.h }, p3/Z, [x21, x9, LSL #1]\n" + "ld1h { z31.h }, p3/Z, [x20, x9, LSL #1]\n" "beq 10f\n" "9:" // Single vector of channels: Loop: 4 inputs loop - "fadd z23.h, z2.h, z1.h\n" - "fadd z19.h, z0.h, z31.h\n" - "ldp x24, x23, [x20, #0x0]\n" - "ldp x22, x21, [x20, #0x10]\n" - "fadd z19.h, z23.h, z19.h\n" + "fadd z17.h, z2.h, z1.h\n" + "fadd z16.h, z0.h, z31.h\n" + "ldp x23, x22, [x24, #0x0]\n" + "ldp x21, x20, [x24, #0x10]\n" + "fadd z16.h, z17.h, z16.h\n" "subs x25, x25, #0x1\n" - "fadd z6.h, z6.h, z19.h\n" - "add x20, x20, #0x20\n" - "ld1h { z2.h }, p3/Z, [x24, x9, LSL #1]\n" - "ld1h { z1.h }, p3/Z, [x23, x9, LSL #1]\n" - "ld1h { z0.h }, p3/Z, [x22, x9, LSL #1]\n" - "ld1h { z31.h }, p3/Z, [x21, x9, LSL #1]\n" + "fadd z6.h, z6.h, z16.h\n" + "add x24, x24, #0x20\n" + "ld1h { z2.h }, p3/Z, [x23, x9, LSL #1]\n" + "ld1h { z1.h }, p3/Z, [x22, x9, LSL #1]\n" + "ld1h { z0.h }, p3/Z, [x21, x9, LSL #1]\n" + "ld1h { z31.h }, p3/Z, [x20, x9, LSL #1]\n" "bgt 9b\n" "10:" // Single vector of channels: Loop: 4 inputs tail - "fadd z23.h, z2.h, z1.h\n" - "fadd z19.h, z0.h, z31.h\n" - "fadd z19.h, z23.h, z19.h\n" - "fadd z6.h, z6.h, z19.h\n" + "fadd z17.h, z2.h, z1.h\n" + "fadd z16.h, z0.h, z31.h\n" + "fadd z16.h, z17.h, z16.h\n" + "fadd z6.h, z6.h, z16.h\n" "11:" // Single vector of channels: Loop: After loop "ands x21, %x[n_valid_cells], #0x3\n" "beq 13f\n" "12:" // Single vector of channels: Loop: Single input loop - "ldr x24, [x20], #0x8\n" - "ld1h { z2.h }, p3/Z, [x24, x9, LSL #1]\n" + "ldr x20, [x24], #0x8\n" + "ld1h { z16.h }, p3/Z, [x20, x9, LSL #1]\n" "subs x21, x21, #0x1\n" - "fadd z6.h, z6.h, z2.h\n" + "fadd z6.h, z6.h, z16.h\n" "bgt 12b\n" "13:" // Single vector of channels: Loop: Single input loop: End "fmul z6.h, z6.h, z7.h\n" |