diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_conv/pooling/kernels/sve_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst/generic.cpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_conv/pooling/kernels/sve_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst/generic.cpp | 50 |
1 files changed, 25 insertions, 25 deletions
diff --git a/src/core/NEON/kernels/arm_conv/pooling/kernels/sve_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst/generic.cpp b/src/core/NEON/kernels/arm_conv/pooling/kernels/sve_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst/generic.cpp index 7464349f9e..3c1858633b 100644 --- a/src/core/NEON/kernels/arm_conv/pooling/kernels/sve_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst/generic.cpp +++ b/src/core/NEON/kernels/arm_conv/pooling/kernels/sve_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst/generic.cpp @@ -86,22 +86,22 @@ void sve_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst_impl( "mov x4, #0x0\n" "ldr x20, [%x[args], %[offsetof_outptrs]]\n" "mov x5, #0x0\n" + "ldr x6, [%x[args], %[offsetof_inptrs]]\n" "mov x19, #0x4\n" - "ldp x6, x7, [x20, #0x0]\n" + "add x7, %x[args], %[offsetof_rescale]\n" + "ldp x8, x17, [x20, #0x0]\n" + "ldp x16, x15, [x20, #0x10]\n" "whilelt p0.h, XZR, x19\n" - "add x8, %x[args], %[offsetof_rescale]\n" - "ldp x17, x16, [x20, #0x10]\n" + "ldp x14, x13, [x6, #0x0]\n" "whilelt p1.h, x4, x3\n" - "ldr x15, [%x[args], %[offsetof_inptrs]]\n" - "ldp x14, x13, [x15, #0x0]\n" - "ldp x12, x11, [x15, #0x10]\n" - "ldp x10, x9, [x15, #0x20]\n" - "ldp x28, x27, [x15, #0x30]\n" - "ldp x26, x25, [x15, #0x40]\n" - "ldp x24, x23, [x15, #0x50]\n" - "ldp x22, x21, [x15, #0x60]\n" - "ldp x20, x19, [x15, #0x70]\n" - "ld1rqh { z7.h }, p0/Z, [x8]\n" + "ldp x12, x11, [x6, #0x10]\n" + "ldp x10, x9, [x6, #0x20]\n" + "ldp x28, x27, [x6, #0x30]\n" + "ldp x26, x25, [x6, #0x40]\n" + "ldp x24, x23, [x6, #0x50]\n" + "ldp x22, x21, [x6, #0x60]\n" + "ldp x20, x19, [x6, #0x70]\n" + "ld1rqh { z7.h }, p0/Z, [x7]\n" "ld1h { z8.h }, p1/Z, [x9, x4, LSL #1]\n" "ld1h { z6.h }, p1/Z, [x28, x4, LSL #1]\n" "ld1h { z5.h }, p1/Z, [x25, x4, LSL #1]\n" @@ -149,9 +149,9 @@ void sve_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst_impl( "ld1h { z28.h }, p1/Z, [x20, x4, LSL #1]\n" "fadd z16.h, z24.h, z22.h\n" "ld1h { z27.h }, p1/Z, [x14, x4, LSL #1]\n" - "fadd z19.h, z19.h, z21.h\n" + "fadd z19.h, z21.h, z19.h\n" "ld1h { z26.h }, p1/Z, [x11, x4, LSL #1]\n" - "fadd z18.h, z18.h, z21.h\n" + "fadd z18.h, z21.h, z18.h\n" "ld1h { z25.h }, p1/Z, [x22, x4, LSL #1]\n" "fadd z17.h, z17.h, z20.h\n" "ld1h { z24.h }, p1/Z, [x19, x4, LSL #1]\n" @@ -159,13 +159,13 @@ void sve_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst_impl( "fadd z16.h, z20.h, z16.h\n" "whilelt p1.h, x4, x3\n" "fmul z19.h, z19.h, z7.h[0]\n" - "st1h { z19.h }, p0, [x6, x5, LSL #1]\n" + "st1h { z19.h }, p0, [x8, x5, LSL #1]\n" "fmul z18.h, z18.h, z7.h[1]\n" "fmul z17.h, z17.h, z7.h[2]\n" - "st1h { z18.h }, p0, [x7, x5, LSL #1]\n" + "st1h { z18.h }, p0, [x17, x5, LSL #1]\n" "fmul z16.h, z16.h, z7.h[3]\n" - "st1h { z17.h }, p0, [x17, x5, LSL #1]\n" - "st1h { z16.h }, p0, [x16, x5, LSL #1]\n" + "st1h { z17.h }, p0, [x16, x5, LSL #1]\n" + "st1h { z16.h }, p0, [x15, x5, LSL #1]\n" "incw x5\n" "b.any 1b\n" "2:" // Vector: Tail @@ -183,18 +183,18 @@ void sve_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst_impl( "fadd z18.h, z26.h, z22.h\n" "fadd z17.h, z25.h, z23.h\n" "fadd z16.h, z24.h, z22.h\n" - "fadd z19.h, z19.h, z21.h\n" - "fadd z18.h, z18.h, z21.h\n" + "fadd z19.h, z21.h, z19.h\n" + "fadd z18.h, z21.h, z18.h\n" "fadd z17.h, z17.h, z20.h\n" "fadd z16.h, z20.h, z16.h\n" "fmul z19.h, z19.h, z7.h[0]\n" - "st1h { z19.h }, p0, [x6, x5, LSL #1]\n" + "st1h { z19.h }, p0, [x8, x5, LSL #1]\n" "fmul z18.h, z18.h, z7.h[1]\n" "fmul z17.h, z17.h, z7.h[2]\n" - "st1h { z18.h }, p0, [x7, x5, LSL #1]\n" + "st1h { z18.h }, p0, [x17, x5, LSL #1]\n" "fmul z16.h, z16.h, z7.h[3]\n" - "st1h { z17.h }, p0, [x17, x5, LSL #1]\n" - "st1h { z16.h }, p0, [x16, x5, LSL #1]\n" + "st1h { z17.h }, p0, [x16, x5, LSL #1]\n" + "st1h { z16.h }, p0, [x15, x5, LSL #1]\n" : : [args] "r" (&args), [offsetof_inptrs] "I" (offsetof(KernelArgs, inptrs)), [offsetof_n_channels] "I" (offsetof(KernelArgs, n_channels)), [offsetof_outptrs] "I" (offsetof(KernelArgs, outptrs)), [offsetof_rescale] "I" (offsetof(KernelArgs, rescale_vals)) : "cc", "memory", "p0", "p1", "x3", "x4", "x5", "x6", "x7", "x8", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31" |