aboutsummaryrefslogtreecommitdiff
path: root/src/core/NEON/kernels/arm_conv/pooling/kernels/sve_u8q_nhwc_avg_generic_depthfirst/generic.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/core/NEON/kernels/arm_conv/pooling/kernels/sve_u8q_nhwc_avg_generic_depthfirst/generic.cpp')
-rw-r--r--src/core/NEON/kernels/arm_conv/pooling/kernels/sve_u8q_nhwc_avg_generic_depthfirst/generic.cpp156
1 files changed, 78 insertions, 78 deletions
diff --git a/src/core/NEON/kernels/arm_conv/pooling/kernels/sve_u8q_nhwc_avg_generic_depthfirst/generic.cpp b/src/core/NEON/kernels/arm_conv/pooling/kernels/sve_u8q_nhwc_avg_generic_depthfirst/generic.cpp
index 36ac381004..e8339a2cd9 100644
--- a/src/core/NEON/kernels/arm_conv/pooling/kernels/sve_u8q_nhwc_avg_generic_depthfirst/generic.cpp
+++ b/src/core/NEON/kernels/arm_conv/pooling/kernels/sve_u8q_nhwc_avg_generic_depthfirst/generic.cpp
@@ -136,7 +136,7 @@ void sve_u8q_nhwc_avg_generic_depthfirst_impl(
"mov z13.d, z15.d\n"
"mov z12.d, z15.d\n"
"mov z11.d, z15.d\n"
- "mov x20, %x[inptrs]\n"
+ "mov x22, %x[inptrs]\n"
"mov z10.d, z15.d\n"
"mov z9.d, z15.d\n"
"mov z8.d, z15.d\n"
@@ -149,42 +149,42 @@ void sve_u8q_nhwc_avg_generic_depthfirst_impl(
"mov z1.d, z15.d\n"
"mov z0.d, z15.d\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 0x455e0bf7 // uaddlb z23.h, z31.b, z30.b\n"
".inst 0x455e0ff6 // uaddlt 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 0x455c0bb5 // uaddlb z21.h, z29.b, z28.b\n"
".inst 0x455c0fb4 // uaddlt 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 0x455a0b73 // uaddlb z19.h, z27.b, z26.b\n"
".inst 0x455a0f72 // uaddlt 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 0x45580b31 // uaddlb z17.h, z25.b, z24.b\n"
".inst 0x45580f30 // uaddlt 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 0x459749ef // uaddwb z15.s, z15.s, z23.h\n"
".inst 0x45974dce // uaddwt 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 0x459649ad // uaddwb z13.s, z13.s, z22.h\n"
".inst 0x45964d8c // uaddwt z12.s, z12.s, z22.h\n"
- "ld1b { z24.b }, p1/Z, [x21, x24]\n"
+ "ld1b { z24.b }, p1/Z, [x20, x24]\n"
".inst 0x4595496b // uaddwb z11.s, z11.s, z21.h\n"
".inst 0x45954d4a // uaddwt z10.s, z10.s, z21.h\n"
".inst 0x45944929 // uaddwb z9.s, z9.s, z20.h\n"
@@ -227,20 +227,20 @@ void sve_u8q_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 0x4508abf7 // ushllb z23.h, z31.b, #0x0\n"
- ".inst 0x4508aff6 // ushllt z22.h, z31.b, #0x0\n"
- "ld1b { z29.b }, p3/Z, [x22, x26]\n"
- "ld1b { z27.b }, p2/Z, [x22, x25]\n"
- ".inst 0x4508abb5 // ushllb z21.h, z29.b, #0x0\n"
- ".inst 0x4508afb4 // ushllt z20.h, z29.b, #0x0\n"
- "ld1b { z25.b }, p1/Z, [x22, x24]\n"
- ".inst 0x4508ab73 // ushllb z19.h, z27.b, #0x0\n"
- ".inst 0x4508af72 // ushllt z18.h, z27.b, #0x0\n"
+ "ldr x20, [x22], #0x8\n"
+ "ld1b { z16.b }, p4/Z, [x20, x27]\n"
+ ".inst 0x4508aa17 // ushllb z23.h, z16.b, #0x0\n"
+ ".inst 0x4508ae16 // ushllt z22.h, z16.b, #0x0\n"
+ "ld1b { z16.b }, p3/Z, [x20, x26]\n"
+ "ld1b { z17.b }, p2/Z, [x20, x25]\n"
+ ".inst 0x4508aa15 // ushllb z21.h, z16.b, #0x0\n"
+ ".inst 0x4508ae14 // ushllt z20.h, z16.b, #0x0\n"
+ "ld1b { z16.b }, p1/Z, [x20, x24]\n"
+ ".inst 0x4508aa33 // ushllb z19.h, z17.b, #0x0\n"
+ ".inst 0x4508ae32 // ushllt z18.h, z17.b, #0x0\n"
"subs x21, x21, #0x1\n"
- ".inst 0x4508ab31 // ushllb z17.h, z25.b, #0x0\n"
- ".inst 0x4508af30 // ushllt z16.h, z25.b, #0x0\n"
+ ".inst 0x4508aa11 // ushllb z17.h, z16.b, #0x0\n"
+ ".inst 0x4508ae10 // ushllt z16.h, z16.b, #0x0\n"
".inst 0x459749ef // uaddwb z15.s, z15.s, z23.h\n"
".inst 0x45974dce // uaddwt z14.s, z14.s, z23.h\n"
".inst 0x459649ad // uaddwb z13.s, z13.s, z22.h\n"
@@ -393,55 +393,55 @@ void sve_u8q_nhwc_avg_generic_depthfirst_impl(
"mov z14.d, z15.d\n"
"mov z13.d, z15.d\n"
"mov z12.d, z15.d\n"
- "mov x20, %x[inptrs]\n"
+ "mov x22, %x[inptrs]\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 0x455e0bf7 // uaddlb z23.h, z31.b, z30.b\n"
- ".inst 0x455e0ff6 // uaddlt z22.h, z31.b, z30.b\n"
- "ldp x22, x21, [x20, #0x0]\n"
+ ".inst 0x455e0bf1 // uaddlb z17.h, z31.b, z30.b\n"
+ ".inst 0x455e0ff0 // uaddlt z16.h, z31.b, z30.b\n"
+ "ldp x21, x20, [x22, #0x0]\n"
"subs x23, x23, #0x1\n"
- ".inst 0x459749ef // uaddwb z15.s, z15.s, z23.h\n"
- ".inst 0x45974dce // uaddwt z14.s, z14.s, z23.h\n"
- "add x20, x20, #0x10\n"
- "ld1b { z31.b }, p4/Z, [x22, x27]\n"
- ".inst 0x459649ad // uaddwb z13.s, z13.s, z22.h\n"
- ".inst 0x45964d8c // uaddwt z12.s, z12.s, z22.h\n"
- "ld1b { z30.b }, p4/Z, [x21, x27]\n"
+ ".inst 0x459149ef // uaddwb z15.s, z15.s, z17.h\n"
+ ".inst 0x45914dce // uaddwt z14.s, z14.s, z17.h\n"
+ "add x22, x22, #0x10\n"
+ "ld1b { z31.b }, p4/Z, [x21, x27]\n"
+ ".inst 0x459049ad // uaddwb z13.s, z13.s, z16.h\n"
+ ".inst 0x45904d8c // uaddwt 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 0x455e0bf7 // uaddlb z23.h, z31.b, z30.b\n"
- ".inst 0x455e0ff6 // uaddlt z22.h, z31.b, z30.b\n"
- ".inst 0x459749ef // uaddwb z15.s, z15.s, z23.h\n"
- ".inst 0x45974dce // uaddwt z14.s, z14.s, z23.h\n"
- ".inst 0x459649ad // uaddwb z13.s, z13.s, z22.h\n"
- ".inst 0x45964d8c // uaddwt z12.s, z12.s, z22.h\n"
+ ".inst 0x455e0bf1 // uaddlb z17.h, z31.b, z30.b\n"
+ ".inst 0x455e0ff0 // uaddlt z16.h, z31.b, z30.b\n"
+ ".inst 0x459149ef // uaddwb z15.s, z15.s, z17.h\n"
+ ".inst 0x45914dce // uaddwt z14.s, z14.s, z17.h\n"
+ ".inst 0x459049ad // uaddwb z13.s, z13.s, z16.h\n"
+ ".inst 0x45904d8c // uaddwt 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 0x4508abf7 // ushllb z23.h, z31.b, #0x0\n"
- ".inst 0x4508aff6 // ushllt z22.h, z31.b, #0x0\n"
+ "ldr x20, [x22], #0x8\n"
+ "ld1b { z16.b }, p4/Z, [x20, x27]\n"
+ ".inst 0x4508aa11 // ushllb z17.h, z16.b, #0x0\n"
+ ".inst 0x4508ae10 // ushllt z16.h, z16.b, #0x0\n"
"subs x21, x21, #0x1\n"
- ".inst 0x459749ef // uaddwb z15.s, z15.s, z23.h\n"
- ".inst 0x45974dce // uaddwt z14.s, z14.s, z23.h\n"
- ".inst 0x459649ad // uaddwb z13.s, z13.s, z22.h\n"
- ".inst 0x45964d8c // uaddwt z12.s, z12.s, z22.h\n"
+ ".inst 0x459149ef // uaddwb z15.s, z15.s, z17.h\n"
+ ".inst 0x45914dce // uaddwt z14.s, z14.s, z17.h\n"
+ ".inst 0x459049ad // uaddwb z13.s, z13.s, z16.h\n"
+ ".inst 0x45904d8c // uaddwt 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 { z17.s }, p0/Z, [%x[left_shift]]\n"
"ld1rw { z16.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 0x4482822f // srshl z15.s, p0/M, z15.s, z17.s\n"
+ ".inst 0x4482822e // srshl z14.s, p0/M, z14.s, z17.s\n"
+ ".inst 0x4482822d // srshl z13.s, p0/M, z13.s, z17.s\n"
+ ".inst 0x4482822c // srshl z12.s, p0/M, z12.s, z17.s\n"
"ld1rw { z17.s }, p0/Z, [%x[right_shift]]\n"
".inst 0x04b075ef // sqrdmulh z15.s, z15.s, z16.s\n"
".inst 0x04b075ce // sqrdmulh z14.s, z14.s, z16.s\n"
@@ -457,17 +457,17 @@ void sve_u8q_nhwc_avg_generic_depthfirst_impl(
"add z14.s, z14.s, z16.s\n"
"add z13.s, z13.s, z16.s\n"
"add z12.s, z12.s, z16.s\n"
- "mov z16.s, #0x0\n"
- "smax z15.s, p0/M, z15.s, z16.s\n"
- "smax z14.s, p0/M, z14.s, z16.s\n"
- "mov z18.s, #0xff\n"
- "smax z13.s, p0/M, z13.s, z16.s\n"
- "smax z12.s, p0/M, z12.s, z16.s\n"
- "smin z15.s, p0/M, z15.s, z18.s\n"
- "smin z14.s, p0/M, z14.s, z18.s\n"
+ "mov z17.s, #0x0\n"
+ "smax z15.s, p0/M, z15.s, z17.s\n"
+ "smax z14.s, p0/M, z14.s, z17.s\n"
+ "mov z16.s, #0xff\n"
+ "smax z13.s, p0/M, z13.s, z17.s\n"
+ "smax z12.s, p0/M, z12.s, z17.s\n"
+ "smin z15.s, p0/M, z15.s, z16.s\n"
+ "smin z14.s, p0/M, z14.s, z16.s\n"
"trn1 z17.h, z15.h, z14.h\n"
- "smin z13.s, p0/M, z13.s, z18.s\n"
- "smin z12.s, p0/M, z12.s, z18.s\n"
+ "smin z13.s, p0/M, z13.s, z16.s\n"
+ "smin z12.s, p0/M, z12.s, z16.s\n"
"trn1 z16.h, z13.h, z12.h\n"
"trn1 z16.b, z17.b, z16.b\n"
"st1b { z16.b }, p4, [%x[outptr], x27]\n"