diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst/generic.cpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst/generic.cpp | 378 |
1 files changed, 189 insertions, 189 deletions
diff --git a/src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst/generic.cpp b/src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst/generic.cpp index 292fd70fba..f3c596909f 100644 --- a/src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst/generic.cpp +++ b/src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst/generic.cpp @@ -43,46 +43,46 @@ void sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst_imp __asm__ __volatile__( "mov x15, #0x0\n" - "ptrue p2.b\n" + "whilelt p2.s, x15, %x[channel_multiplier]\n" "ldr x14, [%x[inptrs], #0x0]\n" "ldr x13, [%x[inptrs], #0x8]\n" + "ptrue p1.b\n" "ldr x12, [%x[inptrs], #0x10]\n" "ldr x11, [%x[inptrs], #0x18]\n" "mov x10, #0x0\n" "ldr x9, [%x[inptrs], #0x20]\n" "ldr x28, [%x[inptrs], #0x28]\n" - "whilelt p1.s, x15, %x[channel_multiplier]\n" + "ld1w { z16.s }, p2/Z, [%x[params]]\n" "ldp x27, x26, [%x[outptrs], #0x0]\n" - "ldp x25, x24, [%x[outptrs], #0x10]\n" - "ldp x23, x22, [%x[outptrs], #0x20]\n" - "ldp x21, x20, [%x[outptrs], #0x30]\n" - "ld1w { z16.s }, p1/Z, [%x[params]]\n" - "ld1rqw { z2.s }, p2/Z, [x14]\n" - "ld1rqw { z3.s }, p2/Z, [x14, #16]\n" - "ld1rqw { z4.s }, p2/Z, [x13]\n" - "ld1rqw { z5.s }, p2/Z, [x13, #16]\n" - "ld1rqw { z6.s }, p2/Z, [x12]\n" - "ld1rqw { z7.s }, p2/Z, [x12, #16]\n" - "ld1rqw { z8.s }, p2/Z, [x11]\n" "mov z25.d, z16.d\n" "mov z15.d, z16.d\n" - "ld1rqw { z9.s }, p2/Z, [x11, #16]\n" - "ld1rqw { z10.s }, p2/Z, [x9]\n" + "ldp x25, x24, [%x[outptrs], #0x10]\n" + "ldp x23, x22, [%x[outptrs], #0x20]\n" "mov z24.d, z16.d\n" "mov z14.d, z16.d\n" - "ld1rqw { z11.s }, p2/Z, [x9, #16]\n" - "ld1rqw { z12.s }, p2/Z, [x28]\n" + "ldp x21, x20, [%x[outptrs], #0x30]\n" + "ld1rqw { z2.s }, p1/Z, [x14]\n" "mov z26.d, z16.d\n" "mov z17.d, z16.d\n" - "ld1rqw { z13.s }, p2/Z, [x28, #16]\n" - "ld1rw { z23.s }, p2/Z, [%x[clamps]]\n" - "mov z21.d, z16.d\n" - "ld1rw { z22.s }, p2/Z, [%x[clamps], #4]\n" - "ld1w { z31.s }, p1/Z, [%x[params], #1, MUL VL]\n" - "ld1w { z30.s }, p1/Z, [%x[params], #2, MUL VL]\n" - "ld1w { z29.s }, p1/Z, [%x[params], #3, MUL VL]\n" - "ld1w { z28.s }, p1/Z, [%x[params], #4, MUL VL]\n" - "ld1w { z27.s }, p1/Z, [%x[params], #5, MUL VL]\n" + "ld1rqw { z3.s }, p1/Z, [x14, #16]\n" + "ld1rqw { z4.s }, p1/Z, [x13]\n" + "mov z23.d, z16.d\n" + "ld1rqw { z5.s }, p1/Z, [x13, #16]\n" + "ld1rqw { z6.s }, p1/Z, [x12]\n" + "ld1rqw { z7.s }, p1/Z, [x12, #16]\n" + "ld1rqw { z8.s }, p1/Z, [x11]\n" + "ld1rqw { z9.s }, p1/Z, [x11, #16]\n" + "ld1rqw { z10.s }, p1/Z, [x9]\n" + "ld1rqw { z11.s }, p1/Z, [x9, #16]\n" + "ld1rqw { z12.s }, p1/Z, [x28]\n" + "ld1rqw { z13.s }, p1/Z, [x28, #16]\n" + "ld1rw { z21.s }, p1/Z, [%x[clamps]]\n" + "ld1rw { z22.s }, p1/Z, [%x[clamps], #4]\n" + "ld1w { z31.s }, p2/Z, [%x[params], #1, MUL VL]\n" + "ld1w { z30.s }, p2/Z, [%x[params], #2, MUL VL]\n" + "ld1w { z29.s }, p2/Z, [%x[params], #3, MUL VL]\n" + "ld1w { z28.s }, p2/Z, [%x[params], #4, MUL VL]\n" + "ld1w { z27.s }, p2/Z, [%x[params], #5, MUL VL]\n" "addvl %x[params], %x[params], #6\n" "1:" // Output channel complete vector loop "fmla z16.s, z31.s, z2.s[0]\n" @@ -92,13 +92,13 @@ void sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst_imp "fmla z15.s, z31.s, z2.s[2]\n" "fmla z24.s, z31.s, z2.s[3]\n" "mov z1.d, z9.d\n" - "mov p0.b, p1.b\n" + "mov p0.b, p2.b\n" "fmla z14.s, z31.s, z4.s[0]\n" "fmla z26.s, z31.s, z4.s[1]\n" + "whilelt p2.s, x15, %x[channel_multiplier]\n" "fmla z17.s, z31.s, z4.s[2]\n" - "fmla z21.s, z31.s, z4.s[3]\n" - "ld1w { z20.s }, p2/Z, [%x[params]]\n" - "whilelt p1.s, x15, %x[channel_multiplier]\n" + "fmla z23.s, z31.s, z4.s[3]\n" + "ld1w { z20.s }, p1/Z, [%x[params]]\n" "fmla z16.s, z30.s, z2.s[1]\n" "fmla z25.s, z30.s, z2.s[2]\n" "fmla z15.s, z30.s, z2.s[3]\n" @@ -106,8 +106,8 @@ void sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst_imp "fmla z14.s, z30.s, z4.s[1]\n" "fmla z26.s, z30.s, z4.s[2]\n" "fmla z17.s, z30.s, z4.s[3]\n" - "fmla z21.s, z30.s, z5.s[0]\n" - "ld1w { z19.s }, p2/Z, [%x[params], #1, MUL VL]\n" + "fmla z23.s, z30.s, z5.s[0]\n" + "ld1w { z19.s }, p1/Z, [%x[params], #1, MUL VL]\n" "fmla z16.s, z29.s, z2.s[2]\n" "fmla z25.s, z29.s, z2.s[3]\n" "fmla z15.s, z29.s, z3.s[0]\n" @@ -115,8 +115,8 @@ void sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst_imp "fmla z14.s, z29.s, z4.s[2]\n" "fmla z26.s, z29.s, z4.s[3]\n" "fmla z17.s, z29.s, z5.s[0]\n" - "fmla z21.s, z29.s, z5.s[1]\n" - "ld1w { z18.s }, p2/Z, [%x[params], #2, MUL VL]\n" + "fmla z23.s, z29.s, z5.s[1]\n" + "ld1w { z18.s }, p1/Z, [%x[params], #2, MUL VL]\n" "fmla z16.s, z28.s, z2.s[3]\n" "fmla z25.s, z28.s, z3.s[0]\n" "fmla z15.s, z28.s, z3.s[1]\n" @@ -124,8 +124,8 @@ void sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst_imp "fmla z14.s, z28.s, z4.s[3]\n" "fmla z26.s, z28.s, z5.s[0]\n" "fmla z17.s, z28.s, z5.s[1]\n" - "fmla z21.s, z28.s, z5.s[2]\n" - "ld1w { z30.s }, p2/Z, [%x[params], #3, MUL VL]\n" + "fmla z23.s, z28.s, z5.s[2]\n" + "ld1w { z28.s }, p1/Z, [%x[params], #3, MUL VL]\n" "fmla z16.s, z27.s, z3.s[0]\n" "fmla z25.s, z27.s, z3.s[1]\n" "fmla z15.s, z27.s, z3.s[2]\n" @@ -133,8 +133,8 @@ void sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst_imp "fmla z14.s, z27.s, z5.s[0]\n" "fmla z26.s, z27.s, z5.s[1]\n" "fmla z17.s, z27.s, z5.s[2]\n" - "fmla z21.s, z27.s, z5.s[3]\n" - "ld1w { z27.s }, p2/Z, [%x[params], #4, MUL VL]\n" + "fmla z23.s, z27.s, z5.s[3]\n" + "ld1w { z27.s }, p1/Z, [%x[params], #4, MUL VL]\n" "fmla z16.s, z20.s, z4.s[0]\n" "fmla z25.s, z20.s, z4.s[1]\n" "fmla z15.s, z20.s, z4.s[2]\n" @@ -142,8 +142,8 @@ void sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst_imp "fmla z14.s, z20.s, z6.s[0]\n" "fmla z26.s, z20.s, z6.s[1]\n" "fmla z17.s, z20.s, z6.s[2]\n" - "fmla z21.s, z20.s, z6.s[3]\n" - "ld1w { z20.s }, p2/Z, [%x[params], #5, MUL VL]\n" + "fmla z23.s, z20.s, z6.s[3]\n" + "ld1w { z20.s }, p1/Z, [%x[params], #5, MUL VL]\n" "fmla z16.s, z19.s, z4.s[1]\n" "fmla z25.s, z19.s, z4.s[2]\n" "fmla z15.s, z19.s, z4.s[3]\n" @@ -151,8 +151,8 @@ void sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst_imp "fmla z14.s, z19.s, z6.s[1]\n" "fmla z26.s, z19.s, z6.s[2]\n" "fmla z17.s, z19.s, z6.s[3]\n" - "fmla z21.s, z19.s, z7.s[0]\n" - "ld1w { z19.s }, p2/Z, [%x[params], #6, MUL VL]\n" + "fmla z23.s, z19.s, z7.s[0]\n" + "ld1w { z19.s }, p1/Z, [%x[params], #6, MUL VL]\n" "fmla z16.s, z18.s, z4.s[2]\n" "fmla z25.s, z18.s, z4.s[3]\n" "fmla z15.s, z18.s, z5.s[0]\n" @@ -160,18 +160,18 @@ void sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst_imp "fmla z14.s, z18.s, z6.s[2]\n" "fmla z26.s, z18.s, z6.s[3]\n" "fmla z17.s, z18.s, z7.s[0]\n" - "fmla z21.s, z18.s, z7.s[1]\n" - "ld1w { z29.s }, p2/Z, [%x[params], #7, MUL VL]\n" + "fmla z23.s, z18.s, z7.s[1]\n" + "ld1w { z18.s }, p1/Z, [%x[params], #7, MUL VL]\n" "addvl %x[params], %x[params], #16\n" - "fmla z16.s, z30.s, z4.s[3]\n" - "fmla z25.s, z30.s, z5.s[0]\n" - "fmla z15.s, z30.s, z5.s[1]\n" - "fmla z24.s, z30.s, z5.s[2]\n" - "fmla z14.s, z30.s, z6.s[3]\n" - "fmla z26.s, z30.s, z7.s[0]\n" - "fmla z17.s, z30.s, z7.s[1]\n" - "fmla z21.s, z30.s, z7.s[2]\n" - "ld1w { z18.s }, p2/Z, [%x[params], #-8, MUL VL]\n" + "fmla z16.s, z28.s, z4.s[3]\n" + "fmla z25.s, z28.s, z5.s[0]\n" + "fmla z15.s, z28.s, z5.s[1]\n" + "fmla z24.s, z28.s, z5.s[2]\n" + "fmla z14.s, z28.s, z6.s[3]\n" + "fmla z26.s, z28.s, z7.s[0]\n" + "fmla z17.s, z28.s, z7.s[1]\n" + "fmla z23.s, z28.s, z7.s[2]\n" + "ld1w { z30.s }, p1/Z, [%x[params], #-8, MUL VL]\n" "fmla z16.s, z27.s, z5.s[0]\n" "fmla z25.s, z27.s, z5.s[1]\n" "fmla z15.s, z27.s, z5.s[2]\n" @@ -179,8 +179,8 @@ void sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst_imp "fmla z14.s, z27.s, z7.s[0]\n" "fmla z26.s, z27.s, z7.s[1]\n" "fmla z17.s, z27.s, z7.s[2]\n" - "fmla z21.s, z27.s, z7.s[3]\n" - "ld1w { z27.s }, p2/Z, [%x[params], #-7, MUL VL]\n" + "fmla z23.s, z27.s, z7.s[3]\n" + "ld1w { z27.s }, p1/Z, [%x[params], #-7, MUL VL]\n" "fmla z16.s, z20.s, z6.s[0]\n" "fmla z25.s, z20.s, z6.s[1]\n" "fmla z15.s, z20.s, z6.s[2]\n" @@ -188,8 +188,8 @@ void sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst_imp "fmla z14.s, z20.s, z0.s[0]\n" "fmla z26.s, z20.s, z0.s[1]\n" "fmla z17.s, z20.s, z0.s[2]\n" - "fmla z21.s, z20.s, z0.s[3]\n" - "ld1w { z20.s }, p2/Z, [%x[params], #-6, MUL VL]\n" + "fmla z23.s, z20.s, z0.s[3]\n" + "ld1w { z20.s }, p1/Z, [%x[params], #-6, MUL VL]\n" "fmla z16.s, z19.s, z6.s[1]\n" "fmla z25.s, z19.s, z6.s[2]\n" "fmla z15.s, z19.s, z6.s[3]\n" @@ -197,26 +197,26 @@ void sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst_imp "fmla z14.s, z19.s, z0.s[1]\n" "fmla z26.s, z19.s, z0.s[2]\n" "fmla z17.s, z19.s, z0.s[3]\n" - "fmla z21.s, z19.s, z1.s[0]\n" - "ld1w { z19.s }, p2/Z, [%x[params], #-5, MUL VL]\n" - "fmla z16.s, z29.s, z6.s[2]\n" - "fmla z25.s, z29.s, z6.s[3]\n" - "fmla z15.s, z29.s, z7.s[0]\n" - "fmla z24.s, z29.s, z7.s[1]\n" - "fmla z14.s, z29.s, z0.s[2]\n" - "fmla z26.s, z29.s, z0.s[3]\n" - "fmla z17.s, z29.s, z1.s[0]\n" - "fmla z21.s, z29.s, z1.s[1]\n" - "ld1w { z30.s }, p2/Z, [%x[params], #-4, MUL VL]\n" - "fmla z16.s, z18.s, z6.s[3]\n" - "fmla z25.s, z18.s, z7.s[0]\n" - "fmla z15.s, z18.s, z7.s[1]\n" - "fmla z24.s, z18.s, z7.s[2]\n" - "fmla z14.s, z18.s, z0.s[3]\n" - "fmla z26.s, z18.s, z1.s[0]\n" - "fmla z17.s, z18.s, z1.s[1]\n" - "fmla z21.s, z18.s, z1.s[2]\n" - "ld1w { z28.s }, p2/Z, [%x[params], #-3, MUL VL]\n" + "fmla z23.s, z19.s, z1.s[0]\n" + "ld1w { z19.s }, p1/Z, [%x[params], #-5, MUL VL]\n" + "fmla z16.s, z18.s, z6.s[2]\n" + "fmla z25.s, z18.s, z6.s[3]\n" + "fmla z15.s, z18.s, z7.s[0]\n" + "fmla z24.s, z18.s, z7.s[1]\n" + "fmla z14.s, z18.s, z0.s[2]\n" + "fmla z26.s, z18.s, z0.s[3]\n" + "fmla z17.s, z18.s, z1.s[0]\n" + "fmla z23.s, z18.s, z1.s[1]\n" + "ld1w { z18.s }, p1/Z, [%x[params], #-4, MUL VL]\n" + "fmla z16.s, z30.s, z6.s[3]\n" + "fmla z25.s, z30.s, z7.s[0]\n" + "fmla z15.s, z30.s, z7.s[1]\n" + "fmla z24.s, z30.s, z7.s[2]\n" + "fmla z14.s, z30.s, z0.s[3]\n" + "fmla z26.s, z30.s, z1.s[0]\n" + "fmla z17.s, z30.s, z1.s[1]\n" + "fmla z23.s, z30.s, z1.s[2]\n" + "ld1w { z31.s }, p1/Z, [%x[params], #-3, MUL VL]\n" "fmla z16.s, z27.s, z7.s[0]\n" "fmla z25.s, z27.s, z7.s[1]\n" "fmla z15.s, z27.s, z7.s[2]\n" @@ -224,8 +224,8 @@ void sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst_imp "fmla z14.s, z27.s, z1.s[0]\n" "fmla z26.s, z27.s, z1.s[1]\n" "fmla z17.s, z27.s, z1.s[2]\n" - "fmla z21.s, z27.s, z1.s[3]\n" - "ld1w { z27.s }, p2/Z, [%x[params], #-2, MUL VL]\n" + "fmla z23.s, z27.s, z1.s[3]\n" + "ld1w { z27.s }, p1/Z, [%x[params], #-2, MUL VL]\n" "fmla z16.s, z20.s, z0.s[0]\n" "fmla z25.s, z20.s, z0.s[1]\n" "fmla z15.s, z20.s, z0.s[2]\n" @@ -234,47 +234,47 @@ void sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst_imp "fmla z14.s, z20.s, z0.s[0]\n" "fmla z26.s, z20.s, z0.s[1]\n" "fmla z17.s, z20.s, z0.s[2]\n" - "fmla z21.s, z20.s, z0.s[3]\n" + "fmla z23.s, z20.s, z0.s[3]\n" "mov z0.d, z8.d\n" - "ld1w { z31.s }, p2/Z, [%x[params], #-1, MUL VL]\n" + "ld1w { z20.s }, p1/Z, [%x[params], #-1, MUL VL]\n" "fmla z16.s, z19.s, z0.s[1]\n" "fmla z25.s, z19.s, z0.s[2]\n" "fmla z15.s, z19.s, z0.s[3]\n" "fmla z24.s, z19.s, z1.s[0]\n" + "mov z1.d, z10.d\n" + "mov z0.d, z11.d\n" + "fmla z14.s, z19.s, z1.s[1]\n" + "fmla z26.s, z19.s, z1.s[2]\n" + "fmla z17.s, z19.s, z1.s[3]\n" + "fmla z23.s, z19.s, z0.s[0]\n" + "mov z1.d, z8.d\n" + "ld1w { z19.s }, p1/Z, [%x[params]]\n" + "mov z0.d, z9.d\n" + "fmla z16.s, z18.s, z1.s[2]\n" + "fmla z25.s, z18.s, z1.s[3]\n" + "fmla z15.s, z18.s, z0.s[0]\n" + "fmla z24.s, z18.s, z0.s[1]\n" + "mov z1.d, z10.d\n" + "mov z0.d, z11.d\n" + "fmla z14.s, z18.s, z1.s[2]\n" + "fmla z26.s, z18.s, z1.s[3]\n" + "fmla z17.s, z18.s, z0.s[0]\n" + "fmla z23.s, z18.s, z0.s[1]\n" + "mov z1.d, z8.d\n" + "ld1w { z18.s }, p1/Z, [%x[params], #1, MUL VL]\n" + "mov z0.d, z9.d\n" + "fmla z16.s, z31.s, z1.s[3]\n" + "fmla z25.s, z31.s, z0.s[0]\n" + "fmla z15.s, z31.s, z0.s[1]\n" + "fmla z24.s, z31.s, z0.s[2]\n" "mov z0.d, z10.d\n" "mov z1.d, z11.d\n" - "fmla z14.s, z19.s, z0.s[1]\n" - "fmla z26.s, z19.s, z0.s[2]\n" - "fmla z17.s, z19.s, z0.s[3]\n" - "mov z0.d, z8.d\n" - "fmla z21.s, z19.s, z1.s[0]\n" - "ld1w { z20.s }, p2/Z, [%x[params]]\n" - "mov z1.d, z9.d\n" - "fmla z16.s, z30.s, z0.s[2]\n" - "fmla z25.s, z30.s, z0.s[3]\n" - "mov z0.d, z10.d\n" - "fmla z15.s, z30.s, z1.s[0]\n" - "fmla z24.s, z30.s, z1.s[1]\n" - "mov z1.d, z11.d\n" - "fmla z14.s, z30.s, z0.s[2]\n" - "fmla z26.s, z30.s, z0.s[3]\n" - "mov z0.d, z8.d\n" - "fmla z17.s, z30.s, z1.s[0]\n" - "fmla z21.s, z30.s, z1.s[1]\n" - "ld1w { z19.s }, p2/Z, [%x[params], #1, MUL VL]\n" - "mov z1.d, z9.d\n" - "fmla z16.s, z28.s, z0.s[3]\n" - "mov z0.d, z10.d\n" - "fmla z25.s, z28.s, z1.s[0]\n" - "fmla z15.s, z28.s, z1.s[1]\n" - "fmla z24.s, z28.s, z1.s[2]\n" - "mov z1.d, z11.d\n" - "fmla z14.s, z28.s, z0.s[3]\n" - "fmla z26.s, z28.s, z1.s[0]\n" - "fmla z17.s, z28.s, z1.s[1]\n" - "fmla z21.s, z28.s, z1.s[2]\n" + "fmla z14.s, z31.s, z0.s[3]\n" + "fmla z26.s, z31.s, z1.s[0]\n" + "fmla z17.s, z31.s, z1.s[1]\n" + "fmla z23.s, z31.s, z1.s[2]\n" "mov z1.d, z9.d\n" - "ld1w { z18.s }, p2/Z, [%x[params], #2, MUL VL]\n" + "ld1w { z28.s }, p1/Z, [%x[params], #2, MUL VL]\n" "fmla z16.s, z27.s, z1.s[0]\n" "fmla z25.s, z27.s, z1.s[1]\n" "fmla z15.s, z27.s, z1.s[2]\n" @@ -283,102 +283,102 @@ void sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst_imp "fmla z14.s, z27.s, z1.s[0]\n" "fmla z26.s, z27.s, z1.s[1]\n" "fmla z17.s, z27.s, z1.s[2]\n" - "fmla z21.s, z27.s, z1.s[3]\n" - "fmla z16.s, z31.s, z0.s[0]\n" - "ld1w { z28.s }, p2/Z, [%x[params], #3, MUL VL]\n" - "fmla z25.s, z31.s, z0.s[1]\n" - "fmla z15.s, z31.s, z0.s[2]\n" - "fmla z24.s, z31.s, z0.s[3]\n" - "mov z0.d, z12.d\n" - "fmla z14.s, z31.s, z0.s[0]\n" - "fmla z26.s, z31.s, z0.s[1]\n" - "fmla z17.s, z31.s, z0.s[2]\n" - "fmla z21.s, z31.s, z0.s[3]\n" - "mov z0.d, z10.d\n" - "ld1w { z31.s }, p1/Z, [%x[params], #5, MUL VL]\n" - "fmla z16.s, z20.s, z0.s[1]\n" - "fmla z25.s, z20.s, z0.s[2]\n" - "fmla z15.s, z20.s, z0.s[3]\n" - "mov z0.d, z12.d\n" - "fmla z24.s, z20.s, z1.s[0]\n" - "mov z1.d, z13.d\n" - "fmla z14.s, z20.s, z0.s[1]\n" - "fmla z26.s, z20.s, z0.s[2]\n" - "fmla z17.s, z20.s, z0.s[3]\n" - "mov z0.d, z10.d\n" - "fmla z21.s, z20.s, z1.s[0]\n" - "mov z1.d, z11.d\n" - "ld1w { z30.s }, p1/Z, [%x[params], #6, MUL VL]\n" - "fmla z16.s, z19.s, z0.s[2]\n" - "fmla z25.s, z19.s, z0.s[3]\n" + "fmla z23.s, z27.s, z1.s[3]\n" + "ld1w { z27.s }, p1/Z, [%x[params], #3, MUL VL]\n" + "fmla z16.s, z20.s, z0.s[0]\n" + "fmla z25.s, z20.s, z0.s[1]\n" + "fmla z15.s, z20.s, z0.s[2]\n" + "fmla z24.s, z20.s, z0.s[3]\n" "mov z0.d, z12.d\n" - "fmla z15.s, z19.s, z1.s[0]\n" - "fmla z24.s, z19.s, z1.s[1]\n" - "mov z1.d, z13.d\n" - "fmla z14.s, z19.s, z0.s[2]\n" - "fmla z26.s, z19.s, z0.s[3]\n" + "fmla z14.s, z20.s, z0.s[0]\n" + "fmla z26.s, z20.s, z0.s[1]\n" + "fmla z17.s, z20.s, z0.s[2]\n" + "fmla z23.s, z20.s, z0.s[3]\n" "mov z0.d, z10.d\n" - "fmla z17.s, z19.s, z1.s[0]\n" - "fmla z21.s, z19.s, z1.s[1]\n" - "mov z1.d, z11.d\n" - "ld1w { z29.s }, p1/Z, [%x[params], #7, MUL VL]\n" - "fmla z16.s, z18.s, z0.s[3]\n" - "mov z0.d, z12.d\n" - "fmla z25.s, z18.s, z1.s[0]\n" - "fmla z15.s, z18.s, z1.s[1]\n" - "fmla z24.s, z18.s, z1.s[2]\n" - "mov z1.d, z13.d\n" - "fmla z14.s, z18.s, z0.s[3]\n" - "fmla z26.s, z18.s, z1.s[0]\n" - "fmla z17.s, z18.s, z1.s[1]\n" - "fmla z21.s, z18.s, z1.s[2]\n" + "ld1w { z31.s }, p2/Z, [%x[params], #5, MUL VL]\n" + "fmla z16.s, z19.s, z0.s[1]\n" + "fmla z25.s, z19.s, z0.s[2]\n" + "fmla z15.s, z19.s, z0.s[3]\n" + "fmla z24.s, z19.s, z1.s[0]\n" + "mov z1.d, z12.d\n" + "mov z0.d, z13.d\n" + "fmla z14.s, z19.s, z1.s[1]\n" + "fmla z26.s, z19.s, z1.s[2]\n" + "fmla z17.s, z19.s, z1.s[3]\n" + "fmla z23.s, z19.s, z0.s[0]\n" + "mov z1.d, z10.d\n" + "ld1w { z30.s }, p2/Z, [%x[params], #6, MUL VL]\n" + "mov z0.d, z11.d\n" + "fmla z16.s, z18.s, z1.s[2]\n" + "fmla z25.s, z18.s, z1.s[3]\n" + "fmla z15.s, z18.s, z0.s[0]\n" + "fmla z24.s, z18.s, z0.s[1]\n" + "mov z1.d, z12.d\n" + "mov z0.d, z13.d\n" + "fmla z14.s, z18.s, z1.s[2]\n" + "fmla z26.s, z18.s, z1.s[3]\n" + "fmla z17.s, z18.s, z0.s[0]\n" + "fmla z23.s, z18.s, z0.s[1]\n" + "mov z1.d, z10.d\n" + "ld1w { z29.s }, p2/Z, [%x[params], #7, MUL VL]\n" "mov z0.d, z11.d\n" - "fmla z16.s, z28.s, z0.s[0]\n" - "fmla z25.s, z28.s, z0.s[1]\n" - "fmla z15.s, z28.s, z0.s[2]\n" - "fmla z24.s, z28.s, z0.s[3]\n" + "fmla z16.s, z28.s, z1.s[3]\n" + "fmla z25.s, z28.s, z0.s[0]\n" + "fmla z15.s, z28.s, z0.s[1]\n" + "fmla z24.s, z28.s, z0.s[2]\n" "mov z0.d, z13.d\n" - "fmla z14.s, z28.s, z0.s[0]\n" - "fmla z26.s, z28.s, z0.s[1]\n" - "fmla z17.s, z28.s, z0.s[2]\n" - "fmla z21.s, z28.s, z0.s[3]\n" - "fmin z16.s, p2/M, z16.s, z22.s\n" - "fmin z25.s, p2/M, z25.s, z22.s\n" - "fmin z15.s, p2/M, z15.s, z22.s\n" - "fmin z24.s, p2/M, z24.s, z22.s\n" - "fmin z14.s, p2/M, z14.s, z22.s\n" - "fmax z16.s, p2/M, z16.s, z23.s\n" - "fmin z26.s, p2/M, z26.s, z22.s\n" - "fmin z17.s, p2/M, z17.s, z22.s\n" - "fmin z21.s, p2/M, z21.s, z22.s\n" - "fmax z25.s, p2/M, z25.s, z23.s\n" + "mov z1.d, z12.d\n" + "fmla z26.s, z28.s, z0.s[0]\n" + "fmla z17.s, z28.s, z0.s[1]\n" + "fmla z23.s, z28.s, z0.s[2]\n" + "mov z0.d, z11.d\n" + "fmla z14.s, z28.s, z1.s[3]\n" + "fmla z16.s, z27.s, z0.s[0]\n" + "fmla z25.s, z27.s, z0.s[1]\n" + "fmin z16.s, p1/M, z16.s, z22.s\n" + "fmax z16.s, p1/M, z16.s, z21.s\n" + "fmla z15.s, z27.s, z0.s[2]\n" + "fmla z24.s, z27.s, z0.s[3]\n" + "mov z0.d, z13.d\n" + "fmin z25.s, p1/M, z25.s, z22.s\n" + "fmla z14.s, z27.s, z0.s[0]\n" + "fmla z26.s, z27.s, z0.s[1]\n" + "fmin z15.s, p1/M, z15.s, z22.s\n" + "fmin z24.s, p1/M, z24.s, z22.s\n" + "fmla z17.s, z27.s, z0.s[2]\n" + "fmla z23.s, z27.s, z0.s[3]\n" + "fmin z14.s, p1/M, z14.s, z22.s\n" + "fmin z26.s, p1/M, z26.s, z22.s\n" + "fmin z17.s, p1/M, z17.s, z22.s\n" + "fmin z23.s, p1/M, z23.s, z22.s\n" "st1w { z16.s }, p0, [x27, x10, LSL #2]\n" - "ld1w { z16.s }, p1/Z, [%x[params], #4, MUL VL]\n" + "ld1w { z16.s }, p2/Z, [%x[params], #4, MUL VL]\n" "addvl %x[params], %x[params], #16\n" - "fmax z15.s, p2/M, z15.s, z23.s\n" - "fmax z24.s, p2/M, z24.s, z23.s\n" - "fmax z14.s, p2/M, z14.s, z23.s\n" - "fmax z26.s, p2/M, z26.s, z23.s\n" - "fmax z17.s, p2/M, z17.s, z23.s\n" + "fmax z25.s, p1/M, z25.s, z21.s\n" "st1w { z25.s }, p0, [x26, x10, LSL #2]\n" "mov z25.d, z16.d\n" - "fmax z21.s, p2/M, z21.s, z23.s\n" + "fmax z15.s, p1/M, z15.s, z21.s\n" + "fmax z24.s, p1/M, z24.s, z21.s\n" "st1w { z15.s }, p0, [x25, x10, LSL #2]\n" "mov z15.d, z16.d\n" - "ld1w { z28.s }, p1/Z, [%x[params], #-8, MUL VL]\n" + "fmax z14.s, p1/M, z14.s, z21.s\n" + "fmax z26.s, p1/M, z26.s, z21.s\n" "st1w { z24.s }, p0, [x24, x10, LSL #2]\n" "mov z24.d, z16.d\n" - "ld1w { z27.s }, p1/Z, [%x[params], #-7, MUL VL]\n" - "addvl %x[params], %x[params], #-6\n" + "fmax z17.s, p1/M, z17.s, z21.s\n" + "fmax z23.s, p1/M, z23.s, z21.s\n" "st1w { z14.s }, p0, [x23, x10, LSL #2]\n" "mov z14.d, z16.d\n" "st1w { z26.s }, p0, [x22, x10, LSL #2]\n" "mov z26.d, z16.d\n" + "ld1w { z28.s }, p2/Z, [%x[params], #-8, MUL VL]\n" + "ld1w { z27.s }, p2/Z, [%x[params], #-7, MUL VL]\n" "st1w { z17.s }, p0, [x21, x10, LSL #2]\n" "mov z17.d, z16.d\n" - "st1w { z21.s }, p0, [x20, x10, LSL #2]\n" + "addvl %x[params], %x[params], #-6\n" + "st1w { z23.s }, p0, [x20, x10, LSL #2]\n" "incw x10\n" - "mov z21.d, z16.d\n" + "mov z23.d, z16.d\n" "b.any 1b\n" : [params] "+&r" (params) : [channel_multiplier] "r" (n_output_channels), [clamps] "r" (minmax_vals), [inptrs] "r" (inptrs), [outptrs] "r" (outptrs) |