From 6269f87f5caca3a83a4822e0556fbf284158615c Mon Sep 17 00:00:00 2001 From: Freddie Liardet Date: Wed, 4 Aug 2021 11:57:37 +0100 Subject: Fix depthwise convolution assembly kernels Resolves: COMPMID-4710 Change-Id: I35b964731aeed0e6f4f873f59341bee48e4a41fd Signed-off-by: Freddie Liardet Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6039 Tested-by: Arm Jenkins Reviewed-by: Gian Marco Iodice Reviewed-by: Michele Di Giorgio Comments-Addressed: Arm Jenkins --- .../generic_indirect.cpp | 2053 ++++++++++++-------- 1 file changed, 1230 insertions(+), 823 deletions(-) (limited to 'src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp16_nhwc_5x5_s1_output2x2_mla_depthfirst/generic_indirect.cpp') diff --git a/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp16_nhwc_5x5_s1_output2x2_mla_depthfirst/generic_indirect.cpp b/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp16_nhwc_5x5_s1_output2x2_mla_depthfirst/generic_indirect.cpp index 96e1ae496e..43eac0e6bc 100644 --- a/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp16_nhwc_5x5_s1_output2x2_mla_depthfirst/generic_indirect.cpp +++ b/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp16_nhwc_5x5_s1_output2x2_mla_depthfirst/generic_indirect.cpp @@ -99,916 +99,1323 @@ void a64_fp16_nhwc_5x5_s1_output2x2_mla_depthfirst_indirect_impl( __asm__ __volatile__( "ldr x21, [%x[params_struct], %[offsetof_args_outptrs]]\n" - "add x16, %x[params_struct], %[offsetof_Args_inptrs]\n" - "ldr x15, [%x[params_struct], %[offsetof_args_params]]\n" + "mov x28, #0x10\n" // cntb _, ALL, #1 + "lsr x27, %x[n_channels], #0x3\n" + "ldr x16, [%x[params_struct], %[offsetof_args_params]]\n" "add x20, %x[params_struct], %[offsetof_args_min]\n" "add x19, %x[params_struct], %[offsetof_args_max]\n" + "ldp x15, x14, [x21, #0x0]\n" + "ldp x13, x12, [x21, #0x10]\n" + "add x11, %x[params_struct], %[offsetof_Args_inptrs]\n" "ld1r { v18.8h }, [x20]\n" "ld1r { v17.8h }, [x19]\n" - "mov x14, #0x0\n" - "ldp x13, x12, [x21, #0x0]\n" - "mov x11, #0x10\n" // cntb _, ALL, #1 - "ldp x10, x9, [x21, #0x10]\n" - "sub x28, XZR, x11\n" - "lsr x27, %x[n_channels], #0x3\n" + "mov x10, #0x0\n" + "sub x9, XZR, x28\n" "cbz x27, 3f\n" - "ldr q16, [x15, #0x0]\n" - "ldr q0, [x15, #0x10]\n" - "cmp x11, x27, LSL #4\n" - "ldr q1, [x15, #0x20]\n" - "ldr q2, [x15, #0x30]\n" - "ldr q3, [x15, #0x40]\n" - "ldr q4, [x15, #0x50]\n" - "add x15, x15, #0x60\n" - "ldp x26, x25, [x16, #0x0]\n" - "ldp x24, x23, [x16, #0x10]\n" - "ldp x22, x21, [x16, #0x20]\n" - "ldr q5, [x26, x14]\n" - "ldr q6, [x25, x14]\n" - "ldr q7, [x24, x14]\n" - "ldr q8, [x23, x14]\n" - "ldr q9, [x22, x14]\n" - "ldr q13, [x21, x14]\n" - "ldp x20, x19, [x16, #0x30]\n" - "ldp x26, x25, [x16, #0x40]\n" - "ldr q11, [x20, x14]\n" - "ldr q12, [x19, x14]\n" - "ldr q10, [x26, x14]\n" - "ldr q14, [x25, x14]\n" + "ldp x26, x25, [x11, #0x0]\n" + "ldr q5, [x26, x10]\n" + "ldr q6, [x25, x10]\n" + "ldp x24, x23, [x11, #0x10]\n" + "cmp x28, x27, LSL #4\n" + "ldp x22, x21, [x11, #0x20]\n" + "ldp x20, x19, [x11, #0x30]\n" + "ldp x26, x25, [x11, #0x40]\n" + "ldr q16, [x16, #0x0]\n" + "ldr q0, [x16, #0x10]\n" + "ldr q1, [x16, #0x20]\n" + "ldr q2, [x16, #0x30]\n" + "ldr q3, [x16, #0x40]\n" + "ldr q4, [x16, #0x50]\n" + "ldr q7, [x24, x10]\n" + "add x16, x16, #0x60\n" + "ldr q8, [x23, x10]\n" + "ldr q9, [x22, x10]\n" + "ldr q13, [x21, x10]\n" + "ldr q11, [x20, x10]\n" + "ldr q12, [x19, x10]\n" + "ldr q10, [x26, x10]\n" + "ldr q14, [x25, x10]\n" "bge 2f\n" "1:" // Channel loop - "mov v31.16b, v16.16b\n fmla v31.8h, v0.8h, v5.8h\n" - "ldr x24, [x16, #0x50]\n" - "add x28, x28, #0x10\n" - "mov v30.16b, v16.16b\n fmla v30.8h, v0.8h, v6.8h\n" - "ldr x23, [x16, #0x58]\n" - "mov v29.16b, v16.16b\n fmla v29.8h, v0.8h, v7.8h\n" - "ldr x22, [x16, #0x60]\n" - "mov v28.16b, v16.16b\n fmla v28.8h, v0.8h, v8.8h\n" - "ldr q5, [x24, x14]\n" - "ldr q0, [x15, #0x0]\n" - "fmla v31.8h, v1.8h, v6.8h\n" - "ldr q6, [x23, x14]\n" - "fmla v30.8h, v1.8h, v9.8h\n" - "fmla v29.8h, v1.8h, v8.8h\n" - "ldr x21, [x16, #0x68]\n" - "fmla v28.8h, v1.8h, v13.8h\n" - "ldr q1, [x15, #0x10]\n" - "ldr x20, [x16, #0x70]\n" - "fmla v31.8h, v2.8h, v9.8h\n" - "ldr q9, [x22, x14]\n" - "fmla v30.8h, v2.8h, v11.8h\n" - "fmla v29.8h, v2.8h, v13.8h\n" - "ldr x19, [x16, #0x78]\n" - "fmla v28.8h, v2.8h, v5.8h\n" - "ldr q2, [x15, #0x20]\n" - "ldr x26, [x16, #0x80]\n" - "fmla v31.8h, v3.8h, v11.8h\n" - "ldr q11, [x21, x14]\n" - "fmla v30.8h, v3.8h, v12.8h\n" - "fmla v29.8h, v3.8h, v5.8h\n" - "ldr x25, [x16, #0x88]\n" - "fmla v28.8h, v3.8h, v6.8h\n" - "ldr q3, [x15, #0x30]\n" - "ldr x24, [x16, #0x90]\n" - "fmla v31.8h, v4.8h, v12.8h\n" - "ldr q12, [x20, x14]\n" - "fmla v30.8h, v4.8h, v9.8h\n" - "fmla v29.8h, v4.8h, v6.8h\n" - "ldr q9, [x19, x14]\n" - "fmla v28.8h, v4.8h, v10.8h\n" - "ldr q4, [x15, #0x40]\n" - "ldr x23, [x16, #0x98]\n" - "fmla v31.8h, v0.8h, v7.8h\n" - "ldr x22, [x16, #0xa0]\n" - "fmla v30.8h, v0.8h, v8.8h\n" - "fmla v29.8h, v0.8h, v14.8h\n" - "ldr x21, [x16, #0xa8]\n" - "fmla v28.8h, v0.8h, v11.8h\n" - "ldr q0, [x15, #0x50]\n" - "ldr x20, [x16, #0xb0]\n" - "fmla v31.8h, v1.8h, v8.8h\n" - "ldr q8, [x25, x14]\n" - "fmla v30.8h, v1.8h, v13.8h\n" - "fmla v29.8h, v1.8h, v11.8h\n" - "ldr x19, [x16, #0xb8]\n" - "fmla v28.8h, v1.8h, v12.8h\n" - "ldr q1, [x15, #0x60]\n" - "ldr x25, [x16, #0xc8]\n" - "fmla v31.8h, v2.8h, v13.8h\n" - "ldr q13, [x26, x14]\n" - "fmla v30.8h, v2.8h, v5.8h\n" - "fmla v29.8h, v2.8h, v12.8h\n" - "ldr x26, [x16, #0xc0]\n" + "mov v28.16b, v16.16b\n fmla v28.8h, v0.8h, v5.8h\n" + "mov v29.16b, v16.16b\n fmla v29.8h, v0.8h, v6.8h\n" + "ldr x24, [x11, #0x50]\n" + "ldr q5, [x24, x10]\n" + "mov v30.16b, v16.16b\n fmla v30.8h, v0.8h, v7.8h\n" + "mov v31.16b, v16.16b\n fmla v31.8h, v0.8h, v8.8h\n" + "ldr x23, [x11, #0x58]\n" + "ldr x22, [x11, #0x60]\n" + "fmla v28.8h, v1.8h, v6.8h\n" + "fmla v29.8h, v1.8h, v9.8h\n" + "ldr q6, [x23, x10]\n" + "ldr x21, [x11, #0x68]\n" + "fmla v30.8h, v1.8h, v8.8h\n" + "fmla v31.8h, v1.8h, v13.8h\n" + "ldr q0, [x16, #0x0]\n" + "ldr x20, [x11, #0x70]\n" "fmla v28.8h, v2.8h, v9.8h\n" - "ldr q2, [x15, #0x70]\n" - "ldr q16, [x15, #0x140]\n" - "fmla v31.8h, v3.8h, v5.8h\n" - "ldr q5, [x24, x14]\n" - "fmla v30.8h, v3.8h, v6.8h\n" - "fmla v29.8h, v3.8h, v9.8h\n" - "ldr x24, [x16, #0xd0]\n" - "fmla v28.8h, v3.8h, v13.8h\n" - "ldr q3, [x15, #0x80]\n" - "fmla v31.8h, v4.8h, v6.8h\n" - "ldr q6, [x23, x14]\n" - "ldr x23, [x16, #0xd8]\n" - "fmla v30.8h, v4.8h, v10.8h\n" - "fmla v29.8h, v4.8h, v13.8h\n" - "ldr q10, [x22, x14]\n" - "fmla v28.8h, v4.8h, v8.8h\n" - "ldr q4, [x15, #0x90]\n" - "ldr x22, [x16, #0xe0]\n" - "fmla v31.8h, v0.8h, v14.8h\n" - "ldr q14, [x19, x14]\n" - "fmla v30.8h, v0.8h, v11.8h\n" - "fmla v29.8h, v0.8h, v5.8h\n" - "ldr x19, [x16, #0xf8]\n" - "fmla v28.8h, v0.8h, v6.8h\n" - "ldr q0, [x15, #0xa0]\n" - "fmla v31.8h, v1.8h, v11.8h\n" - "ldr q11, [x21, x14]\n" - "ldr x21, [x16, #0xe8]\n" - "fmla v30.8h, v1.8h, v12.8h\n" - "fmla v29.8h, v1.8h, v6.8h\n" - "fmla v28.8h, v1.8h, v10.8h\n" - "ldr q1, [x15, #0xb0]\n" - "fmla v31.8h, v2.8h, v12.8h\n" - "ldr q12, [x20, x14]\n" - "ldr x20, [x16, #0xf0]\n" - "fmla v30.8h, v2.8h, v9.8h\n" - "fmla v29.8h, v2.8h, v10.8h\n" - "fmla v28.8h, v2.8h, v11.8h\n" - "ldr q2, [x15, #0xc0]\n" - "fmla v31.8h, v3.8h, v9.8h\n" - "ldr q9, [x26, x14]\n" - "ldr x26, [x16, #0x100]\n" - "fmla v30.8h, v3.8h, v13.8h\n" - "fmla v29.8h, v3.8h, v11.8h\n" - "fmla v28.8h, v3.8h, v12.8h\n" - "ldr q3, [x15, #0xd0]\n" - "fmla v31.8h, v4.8h, v13.8h\n" - "ldr q13, [x25, x14]\n" - "ldr x25, [x16, #0x108]\n" - "fmla v30.8h, v4.8h, v8.8h\n" - "fmla v29.8h, v4.8h, v12.8h\n" - "ldr q8, [x22, x14]\n" - "fmla v28.8h, v4.8h, v14.8h\n" - "ldr q4, [x15, #0xe0]\n" - "fmla v31.8h, v0.8h, v5.8h\n" - "ldr q5, [x24, x14]\n" - "ldr x24, [x16, #0x110]\n" - "fmla v30.8h, v0.8h, v6.8h\n" - "fmla v29.8h, v0.8h, v9.8h\n" - "fmla v28.8h, v0.8h, v13.8h\n" - "ldr q0, [x15, #0xf0]\n" - "fmla v31.8h, v1.8h, v6.8h\n" - "ldr q6, [x23, x14]\n" - "ldr x23, [x16, #0x118]\n" - "fmla v30.8h, v1.8h, v10.8h\n" + "fmla v29.8h, v2.8h, v11.8h\n" + "ldr q9, [x22, x10]\n" + "ldr q1, [x16, #0x10]\n" + "fmla v30.8h, v2.8h, v13.8h\n" + "fmla v31.8h, v2.8h, v5.8h\n" + "ldr x19, [x11, #0x78]\n" + "ldr q2, [x16, #0x20]\n" + "fmla v28.8h, v3.8h, v11.8h\n" + "fmla v29.8h, v3.8h, v12.8h\n" + "ldr q11, [x21, x10]\n" + "ldr x26, [x11, #0x80]\n" + "fmla v30.8h, v3.8h, v5.8h\n" + "fmla v31.8h, v3.8h, v6.8h\n" + "ldr q3, [x16, #0x30]\n" + "ldr x25, [x11, #0x88]\n" + "fmla v28.8h, v4.8h, v12.8h\n" + "fmla v29.8h, v4.8h, v9.8h\n" + "ldr q12, [x20, x10]\n" + "ldr q9, [x19, x10]\n" + "fmla v30.8h, v4.8h, v6.8h\n" + "fmla v31.8h, v4.8h, v10.8h\n" + "ldr q4, [x16, #0x40]\n" + "ldr x24, [x11, #0x90]\n" + "fmla v28.8h, v0.8h, v7.8h\n" + "fmla v29.8h, v0.8h, v8.8h\n" + "ldr x23, [x11, #0x98]\n" + "ldr x22, [x11, #0xa0]\n" + "fmla v30.8h, v0.8h, v14.8h\n" + "fmla v31.8h, v0.8h, v11.8h\n" + "ldr q0, [x16, #0x50]\n" + "ldr x21, [x11, #0xa8]\n" + "fmla v28.8h, v1.8h, v8.8h\n" "fmla v29.8h, v1.8h, v13.8h\n" - "fmla v28.8h, v1.8h, v5.8h\n" - "ldr q1, [x15, #0x100]\n" - "fmla v31.8h, v2.8h, v10.8h\n" - "ldr q10, [x21, x14]\n" - "fmla v30.8h, v2.8h, v11.8h\n" + "ldr q8, [x25, x10]\n" + "ldr x20, [x11, #0xb0]\n" + "fmla v30.8h, v1.8h, v11.8h\n" + "fmla v31.8h, v1.8h, v12.8h\n" + "ldr q1, [x16, #0x60]\n" + "ldr x19, [x11, #0xb8]\n" + "fmla v28.8h, v2.8h, v13.8h\n" "fmla v29.8h, v2.8h, v5.8h\n" - "fmla v28.8h, v2.8h, v6.8h\n" - "ldr q2, [x15, #0x110]\n" - "fmla v31.8h, v3.8h, v11.8h\n" - "ldr q11, [x20, x14]\n" - "fmla v30.8h, v3.8h, v12.8h\n" + "ldr q13, [x26, x10]\n" + "ldr x26, [x11, #0xc0]\n" + "fmla v30.8h, v2.8h, v12.8h\n" + "fmla v31.8h, v2.8h, v9.8h\n" + "ldr q2, [x16, #0x70]\n" + "ldr x25, [x11, #0xc8]\n" + "fmla v28.8h, v3.8h, v5.8h\n" "fmla v29.8h, v3.8h, v6.8h\n" - "fmla v28.8h, v3.8h, v8.8h\n" - "ldr q3, [x15, #0x120]\n" - "fmla v31.8h, v4.8h, v12.8h\n" - "ldr q12, [x19, x14]\n" - "fmla v30.8h, v4.8h, v14.8h\n" - "fmla v29.8h, v4.8h, v8.8h\n" - "fmla v28.8h, v4.8h, v10.8h\n" - "ldr q4, [x15, #0x130]\n" - "fmla v31.8h, v0.8h, v9.8h\n" - "ldr q9, [x26, x14]\n" - "fmla v30.8h, v0.8h, v13.8h\n" + "ldr q5, [x24, x10]\n" + "ldr x24, [x11, #0xd0]\n" + "fmla v30.8h, v3.8h, v9.8h\n" + "fmla v31.8h, v3.8h, v13.8h\n" + "ldr q3, [x16, #0x80]\n" + "add x9, x9, #0x10\n" + "fmla v28.8h, v4.8h, v6.8h\n" + "fmla v29.8h, v4.8h, v10.8h\n" + "ldr q6, [x23, x10]\n" + "ldr q10, [x22, x10]\n" + "fmla v30.8h, v4.8h, v13.8h\n" + "fmla v31.8h, v4.8h, v8.8h\n" + "ldr q4, [x16, #0x90]\n" + "ldr x23, [x11, #0xd8]\n" + "fmla v28.8h, v0.8h, v14.8h\n" "fmla v29.8h, v0.8h, v11.8h\n" - "ldr q11, [x25, x14]\n" - "fmla v28.8h, v0.8h, v12.8h\n" - "ldp x26, x25, [x16, #0x0]\n" - "ldr q0, [x15, #0x150]\n" - "fmla v31.8h, v1.8h, v13.8h\n" - "fmla v30.8h, v1.8h, v5.8h\n" + "ldr q14, [x19, x10]\n" + "ldr x22, [x11, #0xe0]\n" + "fmla v30.8h, v0.8h, v5.8h\n" + "fmla v31.8h, v0.8h, v6.8h\n" + "ldr q0, [x16, #0xa0]\n" + "ldr x19, [x11, #0xf8]\n" + "fmla v28.8h, v1.8h, v11.8h\n" "fmla v29.8h, v1.8h, v12.8h\n" - "ldr q12, [x24, x14]\n" - "fmla v28.8h, v1.8h, v9.8h\n" - "ldr q1, [x15, #0x160]\n" - "fmla v31.8h, v2.8h, v5.8h\n" - "ldr q5, [x26, x11]\n" - "fmla v30.8h, v2.8h, v6.8h\n" + "ldr q11, [x21, x10]\n" + "ldr x21, [x11, #0xe8]\n" + "fmla v30.8h, v1.8h, v6.8h\n" + "fmla v31.8h, v1.8h, v10.8h\n" + "ldr q1, [x16, #0xb0]\n" + "ldr q16, [x16, #0x140]\n" + "fmla v28.8h, v2.8h, v12.8h\n" "fmla v29.8h, v2.8h, v9.8h\n" - "ldr q9, [x23, x14]\n" - "add x14, x14, #0x10\n" - "fmla v28.8h, v2.8h, v11.8h\n" - "ldp x24, x23, [x16, #0x10]\n" - "ldp x22, x21, [x16, #0x20]\n" - "fmla v31.8h, v3.8h, v6.8h\n" - "ldr q6, [x25, x11]\n" - "fmla v30.8h, v3.8h, v8.8h\n" - "fmla v29.8h, v3.8h, v11.8h\n" - "ldr q7, [x24, x11]\n" - "ldr q13, [x21, x11]\n" - "fmla v28.8h, v3.8h, v12.8h\n" - "ldp x20, x19, [x16, #0x30]\n" - "fmla v31.8h, v4.8h, v8.8h\n" - "ldr q8, [x23, x11]\n" - "fmla v30.8h, v4.8h, v10.8h\n" - "fmla v29.8h, v4.8h, v12.8h\n" - "ldr q11, [x20, x11]\n" - "ldr q12, [x19, x11]\n" - "fmla v28.8h, v4.8h, v9.8h\n" - "ldr q9, [x22, x11]\n" - "fmax v31.8h, v31.8h, v18.8h\n" - "ldp x26, x25, [x16, #0x40]\n" - "fmax v30.8h, v30.8h, v18.8h\n" - "fmax v29.8h, v29.8h, v18.8h\n" - "ldr q2, [x15, #0x170]\n" - "ldr q3, [x15, #0x180]\n" - "fmin v31.8h, v31.8h, v17.8h\n" - "fmax v28.8h, v28.8h, v18.8h\n" - "ldr q10, [x26, x11]\n" - "fmin v30.8h, v30.8h, v17.8h\n" - "ldr q14, [x25, x11]\n" - "add x11, x11, #0x10\n" - "fmin v29.8h, v29.8h, v17.8h\n" - "str q31, [x13, x28]\n" - "cmp x11, x27, LSL #4\n" - "fmin v28.8h, v28.8h, v17.8h\n" - "str q30, [x12, x28]\n" - "ldr q4, [x15, #0x190]\n" - "add x15, x15, #0x1a0\n" - "str q29, [x10, x28]\n" - "str q28, [x9, x28]\n" - "blt 1b\n" - "2:" // Channel tail - "mov v31.16b, v16.16b\n fmla v31.8h, v0.8h, v5.8h\n" - "ldr x24, [x16, #0x50]\n" - "add x28, x28, #0x10\n" - "mov v30.16b, v16.16b\n fmla v30.8h, v0.8h, v6.8h\n" - "ldr x23, [x16, #0x58]\n" - "mov v29.16b, v16.16b\n fmla v29.8h, v0.8h, v7.8h\n" - "ldr x22, [x16, #0x60]\n" - "mov v28.16b, v16.16b\n fmla v28.8h, v0.8h, v8.8h\n" - "ldr q5, [x24, x14]\n" - "ldr q0, [x15, #0x0]\n" - "fmla v31.8h, v1.8h, v6.8h\n" - "ldr q6, [x23, x14]\n" - "fmla v30.8h, v1.8h, v9.8h\n" - "fmla v29.8h, v1.8h, v8.8h\n" - "ldr x21, [x16, #0x68]\n" - "fmla v28.8h, v1.8h, v13.8h\n" - "ldr q1, [x15, #0x10]\n" - "ldr x20, [x16, #0x70]\n" - "fmla v31.8h, v2.8h, v9.8h\n" - "ldr q9, [x22, x14]\n" - "fmla v30.8h, v2.8h, v11.8h\n" - "fmla v29.8h, v2.8h, v13.8h\n" - "ldr x19, [x16, #0x78]\n" - "fmla v28.8h, v2.8h, v5.8h\n" - "ldr q2, [x15, #0x20]\n" - "ldr x26, [x16, #0x80]\n" - "fmla v31.8h, v3.8h, v11.8h\n" - "ldr q11, [x21, x14]\n" - "fmla v30.8h, v3.8h, v12.8h\n" - "fmla v29.8h, v3.8h, v5.8h\n" - "ldr x25, [x16, #0x88]\n" - "fmla v28.8h, v3.8h, v6.8h\n" - "ldr q3, [x15, #0x30]\n" - "ldr x24, [x16, #0x90]\n" - "fmla v31.8h, v4.8h, v12.8h\n" - "ldr q12, [x20, x14]\n" - "fmla v30.8h, v4.8h, v9.8h\n" - "fmla v29.8h, v4.8h, v6.8h\n" - "ldr q9, [x19, x14]\n" - "fmla v28.8h, v4.8h, v10.8h\n" - "ldr q4, [x15, #0x40]\n" - "ldr x23, [x16, #0x98]\n" - "fmla v31.8h, v0.8h, v7.8h\n" - "ldr x22, [x16, #0xa0]\n" - "fmla v30.8h, v0.8h, v8.8h\n" - "fmla v29.8h, v0.8h, v14.8h\n" - "ldr x21, [x16, #0xa8]\n" - "fmla v28.8h, v0.8h, v11.8h\n" - "ldr q0, [x15, #0x50]\n" - "ldr x20, [x16, #0xb0]\n" - "fmla v31.8h, v1.8h, v8.8h\n" - "ldr q8, [x25, x14]\n" + "ldr q12, [x20, x10]\n" + "ldr x20, [x11, #0xf0]\n" + "fmla v30.8h, v2.8h, v10.8h\n" + "fmla v31.8h, v2.8h, v11.8h\n" + "ldr q2, [x16, #0xc0]\n" + "fmla v28.8h, v3.8h, v9.8h\n" + "fmla v29.8h, v3.8h, v13.8h\n" + "ldr q9, [x26, x10]\n" + "ldr x26, [x11, #0x100]\n" + "fmla v30.8h, v3.8h, v11.8h\n" + "fmla v31.8h, v3.8h, v12.8h\n" + "ldr q3, [x16, #0xd0]\n" + "fmla v28.8h, v4.8h, v13.8h\n" + "fmla v29.8h, v4.8h, v8.8h\n" + "ldr q13, [x25, x10]\n" + "ldr q8, [x22, x10]\n" + "fmla v30.8h, v4.8h, v12.8h\n" + "fmla v31.8h, v4.8h, v14.8h\n" + "ldr q4, [x16, #0xe0]\n" + "ldr x25, [x11, #0x108]\n" + "fmla v28.8h, v0.8h, v5.8h\n" + "fmla v29.8h, v0.8h, v6.8h\n" + "ldr q5, [x24, x10]\n" + "ldr x24, [x11, #0x110]\n" + "fmla v30.8h, v0.8h, v9.8h\n" + "fmla v31.8h, v0.8h, v13.8h\n" + "ldr q0, [x16, #0xf0]\n" + "fmla v28.8h, v1.8h, v6.8h\n" + "fmla v29.8h, v1.8h, v10.8h\n" + "ldr q6, [x23, x10]\n" + "ldr x23, [x11, #0x118]\n" "fmla v30.8h, v1.8h, v13.8h\n" - "fmla v29.8h, v1.8h, v11.8h\n" - "ldr x19, [x16, #0xb8]\n" - "fmla v28.8h, v1.8h, v12.8h\n" - "ldr q1, [x15, #0x60]\n" - "ldr x25, [x16, #0xc8]\n" - "fmla v31.8h, v2.8h, v13.8h\n" - "ldr q13, [x26, x14]\n" + "fmla v31.8h, v1.8h, v5.8h\n" + "ldr q1, [x16, #0x100]\n" + "fmla v28.8h, v2.8h, v10.8h\n" + "fmla v29.8h, v2.8h, v11.8h\n" + "ldr q10, [x21, x10]\n" "fmla v30.8h, v2.8h, v5.8h\n" - "fmla v29.8h, v2.8h, v12.8h\n" - "ldr x26, [x16, #0xc0]\n" - "fmla v28.8h, v2.8h, v9.8h\n" - "ldr q2, [x15, #0x70]\n" - "fmla v31.8h, v3.8h, v5.8h\n" - "ldr q5, [x24, x14]\n" - "ldr x24, [x16, #0xd0]\n" + "fmla v31.8h, v2.8h, v6.8h\n" + "ldr q2, [x16, #0x110]\n" + "fmla v28.8h, v3.8h, v11.8h\n" + "fmla v29.8h, v3.8h, v12.8h\n" + "ldr q11, [x20, x10]\n" "fmla v30.8h, v3.8h, v6.8h\n" - "fmla v29.8h, v3.8h, v9.8h\n" - "fmla v28.8h, v3.8h, v13.8h\n" - "ldr q3, [x15, #0x80]\n" - "fmla v31.8h, v4.8h, v6.8h\n" - "ldr q6, [x23, x14]\n" - "ldr x23, [x16, #0xd8]\n" - "fmla v30.8h, v4.8h, v10.8h\n" - "fmla v29.8h, v4.8h, v13.8h\n" - "ldr q10, [x22, x14]\n" - "fmla v28.8h, v4.8h, v8.8h\n" - "ldr q4, [x15, #0x90]\n" - "ldr x22, [x16, #0xe0]\n" - "fmla v31.8h, v0.8h, v14.8h\n" - "ldr q14, [x19, x14]\n" + "fmla v31.8h, v3.8h, v8.8h\n" + "ldr q3, [x16, #0x120]\n" + "fmla v28.8h, v4.8h, v12.8h\n" + "fmla v29.8h, v4.8h, v14.8h\n" + "ldr q12, [x19, x10]\n" + "fmla v30.8h, v4.8h, v8.8h\n" + "fmla v31.8h, v4.8h, v10.8h\n" + "ldr q4, [x16, #0x130]\n" + "fmla v28.8h, v0.8h, v9.8h\n" + "fmla v29.8h, v0.8h, v13.8h\n" + "ldr q9, [x26, x10]\n" "fmla v30.8h, v0.8h, v11.8h\n" - "fmla v29.8h, v0.8h, v5.8h\n" - "ldr x19, [x16, #0xf8]\n" - "fmla v28.8h, v0.8h, v6.8h\n" - "ldr q0, [x15, #0xa0]\n" - "fmla v31.8h, v1.8h, v11.8h\n" - "ldr q11, [x21, x14]\n" - "ldr x21, [x16, #0xe8]\n" + "fmla v31.8h, v0.8h, v12.8h\n" + "ldr q11, [x25, x10]\n" + "ldp x26, x25, [x11, #0x0]\n" + "fmla v28.8h, v1.8h, v13.8h\n" + "fmla v29.8h, v1.8h, v5.8h\n" + "ldr q0, [x16, #0x150]\n" "fmla v30.8h, v1.8h, v12.8h\n" - "fmla v29.8h, v1.8h, v6.8h\n" - "fmla v28.8h, v1.8h, v10.8h\n" - "ldr q1, [x15, #0xb0]\n" - "fmla v31.8h, v2.8h, v12.8h\n" - "ldr q12, [x20, x14]\n" - "ldr x20, [x16, #0xf0]\n" + "fmla v31.8h, v1.8h, v9.8h\n" + "ldr q12, [x24, x10]\n" + "ldr q1, [x16, #0x160]\n" + "fmla v28.8h, v2.8h, v5.8h\n" + "fmla v29.8h, v2.8h, v6.8h\n" + "ldr q5, [x26, x28]\n" "fmla v30.8h, v2.8h, v9.8h\n" - "fmla v29.8h, v2.8h, v10.8h\n" - "fmla v28.8h, v2.8h, v11.8h\n" - "ldr q2, [x15, #0xc0]\n" - "fmla v31.8h, v3.8h, v9.8h\n" - "ldr q9, [x26, x14]\n" - "ldr x26, [x16, #0x100]\n" - "fmla v30.8h, v3.8h, v13.8h\n" - "fmla v29.8h, v3.8h, v11.8h\n" - "fmla v28.8h, v3.8h, v12.8h\n" - "ldr q3, [x15, #0xd0]\n" - "fmla v31.8h, v4.8h, v13.8h\n" - "ldr q13, [x25, x14]\n" - "ldr x25, [x16, #0x108]\n" - "fmla v30.8h, v4.8h, v8.8h\n" - "fmla v29.8h, v4.8h, v12.8h\n" - "ldr q8, [x22, x14]\n" - "fmla v28.8h, v4.8h, v14.8h\n" - "ldr q4, [x15, #0xe0]\n" - "fmla v31.8h, v0.8h, v5.8h\n" - "ldr q5, [x24, x14]\n" - "ldr x24, [x16, #0x110]\n" - "fmla v30.8h, v0.8h, v6.8h\n" - "fmla v29.8h, v0.8h, v9.8h\n" - "fmla v28.8h, v0.8h, v13.8h\n" - "ldr q0, [x15, #0xf0]\n" - "fmla v31.8h, v1.8h, v6.8h\n" - "ldr q6, [x23, x14]\n" - "ldr x23, [x16, #0x118]\n" - "fmla v30.8h, v1.8h, v10.8h\n" + "fmla v31.8h, v2.8h, v11.8h\n" + "ldr q9, [x23, x10]\n" + "ldp x24, x23, [x11, #0x10]\n" + "fmla v28.8h, v3.8h, v6.8h\n" + "fmla v29.8h, v3.8h, v8.8h\n" + "ldr q6, [x25, x28]\n" + "ldp x22, x21, [x11, #0x20]\n" + "fmla v30.8h, v3.8h, v11.8h\n" + "fmla v31.8h, v3.8h, v12.8h\n" + "ldp x20, x19, [x11, #0x30]\n" + "ldp x26, x25, [x11, #0x40]\n" + "fmla v28.8h, v4.8h, v8.8h\n" + "fmla v29.8h, v4.8h, v10.8h\n" + "fmax v28.8h, v28.8h, v18.8h\n" + "ldr q7, [x24, x28]\n" + "fmla v30.8h, v4.8h, v12.8h\n" + "fmla v31.8h, v4.8h, v9.8h\n" + "fmax v29.8h, v29.8h, v18.8h\n" + "ldr q8, [x23, x28]\n" + "fmax v30.8h, v30.8h, v18.8h\n" + "fmax v31.8h, v31.8h, v18.8h\n" + "ldr q9, [x22, x28]\n" + "ldr q13, [x21, x28]\n" + "ldr q11, [x20, x28]\n" + "ldr q12, [x19, x28]\n" + "fmin v28.8h, v28.8h, v17.8h\n" + "fmin v29.8h, v29.8h, v17.8h\n" + "ldr q10, [x26, x28]\n" + "ldr q14, [x25, x28]\n" + "add x28, x28, #0x10\n" + "cmp x28, x27, LSL #4\n" + "fmin v30.8h, v30.8h, v17.8h\n" + "fmin v31.8h, v31.8h, v17.8h\n" + "add x10, x10, #0x10\n" + "str q28, [x15, x9]\n" + "str q29, [x14, x9]\n" + "ldr q2, [x16, #0x170]\n" + "ldr q3, [x16, #0x180]\n" + "str q30, [x13, x9]\n" + "ldr q4, [x16, #0x190]\n" + "add x16, x16, #0x1a0\n" + "str q31, [x12, x9]\n" + "blt 1b\n" + "2:" // Channel tail + "mov v28.16b, v16.16b\n fmla v28.8h, v0.8h, v5.8h\n" + "mov v29.16b, v16.16b\n fmla v29.8h, v0.8h, v6.8h\n" + "ldr x24, [x11, #0x50]\n" + "ldr q5, [x24, x10]\n" + "mov v30.16b, v16.16b\n fmla v30.8h, v0.8h, v7.8h\n" + "mov v31.16b, v16.16b\n fmla v31.8h, v0.8h, v8.8h\n" + "ldr x23, [x11, #0x58]\n" + "ldr x22, [x11, #0x60]\n" + "fmla v28.8h, v1.8h, v6.8h\n" + "fmla v29.8h, v1.8h, v9.8h\n" + "ldr q6, [x23, x10]\n" + "ldr x21, [x11, #0x68]\n" + "fmla v30.8h, v1.8h, v8.8h\n" + "fmla v31.8h, v1.8h, v13.8h\n" + "ldr q0, [x16, #0x0]\n" + "ldr x20, [x11, #0x70]\n" + "fmla v28.8h, v2.8h, v9.8h\n" + "fmla v29.8h, v2.8h, v11.8h\n" + "ldr q9, [x22, x10]\n" + "ldr q1, [x16, #0x10]\n" + "fmla v30.8h, v2.8h, v13.8h\n" + "fmla v31.8h, v2.8h, v5.8h\n" + "ldr x19, [x11, #0x78]\n" + "ldr q2, [x16, #0x20]\n" + "fmla v28.8h, v3.8h, v11.8h\n" + "fmla v29.8h, v3.8h, v12.8h\n" + "ldr q11, [x21, x10]\n" + "ldr x26, [x11, #0x80]\n" + "fmla v30.8h, v3.8h, v5.8h\n" + "fmla v31.8h, v3.8h, v6.8h\n" + "ldr q3, [x16, #0x30]\n" + "ldr x25, [x11, #0x88]\n" + "fmla v28.8h, v4.8h, v12.8h\n" + "fmla v29.8h, v4.8h, v9.8h\n" + "ldr q12, [x20, x10]\n" + "ldr q9, [x19, x10]\n" + "fmla v30.8h, v4.8h, v6.8h\n" + "fmla v31.8h, v4.8h, v10.8h\n" + "ldr q4, [x16, #0x40]\n" + "ldr x24, [x11, #0x90]\n" + "fmla v28.8h, v0.8h, v7.8h\n" + "fmla v29.8h, v0.8h, v8.8h\n" + "ldr x23, [x11, #0x98]\n" + "ldr x22, [x11, #0xa0]\n" + "fmla v30.8h, v0.8h, v14.8h\n" + "fmla v31.8h, v0.8h, v11.8h\n" + "ldr q0, [x16, #0x50]\n" + "ldr x21, [x11, #0xa8]\n" + "fmla v28.8h, v1.8h, v8.8h\n" "fmla v29.8h, v1.8h, v13.8h\n" - "fmla v28.8h, v1.8h, v5.8h\n" - "ldr q1, [x15, #0x100]\n" - "fmla v31.8h, v2.8h, v10.8h\n" - "ldr q10, [x21, x14]\n" - "fmla v30.8h, v2.8h, v11.8h\n" + "ldr q8, [x25, x10]\n" + "ldr x20, [x11, #0xb0]\n" + "fmla v30.8h, v1.8h, v11.8h\n" + "fmla v31.8h, v1.8h, v12.8h\n" + "ldr q1, [x16, #0x60]\n" + "ldr x19, [x11, #0xb8]\n" + "fmla v28.8h, v2.8h, v13.8h\n" "fmla v29.8h, v2.8h, v5.8h\n" - "fmla v28.8h, v2.8h, v6.8h\n" - "ldr q2, [x15, #0x110]\n" - "fmla v31.8h, v3.8h, v11.8h\n" - "ldr q11, [x20, x14]\n" - "fmla v30.8h, v3.8h, v12.8h\n" + "ldr q13, [x26, x10]\n" + "ldr x26, [x11, #0xc0]\n" + "fmla v30.8h, v2.8h, v12.8h\n" + "fmla v31.8h, v2.8h, v9.8h\n" + "ldr q2, [x16, #0x70]\n" + "ldr x25, [x11, #0xc8]\n" + "fmla v28.8h, v3.8h, v5.8h\n" "fmla v29.8h, v3.8h, v6.8h\n" - "fmla v28.8h, v3.8h, v8.8h\n" - "ldr q3, [x15, #0x120]\n" - "fmla v31.8h, v4.8h, v12.8h\n" - "ldr q12, [x19, x14]\n" - "fmla v30.8h, v4.8h, v14.8h\n" - "fmla v29.8h, v4.8h, v8.8h\n" - "fmla v28.8h, v4.8h, v10.8h\n" - "ldr q4, [x15, #0x130]\n" - "add x15, x15, #0x140\n" - "fmla v31.8h, v0.8h, v9.8h\n" - "ldr q9, [x26, x14]\n" - "fmla v30.8h, v0.8h, v13.8h\n" + "ldr q5, [x24, x10]\n" + "ldr x24, [x11, #0xd0]\n" + "fmla v30.8h, v3.8h, v9.8h\n" + "fmla v31.8h, v3.8h, v13.8h\n" + "ldr q3, [x16, #0x80]\n" + "add x9, x9, #0x10\n" + "fmla v28.8h, v4.8h, v6.8h\n" + "fmla v29.8h, v4.8h, v10.8h\n" + "ldr q6, [x23, x10]\n" + "ldr q10, [x22, x10]\n" + "fmla v30.8h, v4.8h, v13.8h\n" + "fmla v31.8h, v4.8h, v8.8h\n" + "ldr q4, [x16, #0x90]\n" + "ldr x23, [x11, #0xd8]\n" + "fmla v28.8h, v0.8h, v14.8h\n" "fmla v29.8h, v0.8h, v11.8h\n" - "ldr q11, [x25, x14]\n" - "fmla v28.8h, v0.8h, v12.8h\n" - "fmla v31.8h, v1.8h, v13.8h\n" - "fmla v30.8h, v1.8h, v5.8h\n" + "ldr q14, [x19, x10]\n" + "ldr x22, [x11, #0xe0]\n" + "fmla v30.8h, v0.8h, v5.8h\n" + "fmla v31.8h, v0.8h, v6.8h\n" + "ldr q0, [x16, #0xa0]\n" + "ldr x19, [x11, #0xf8]\n" + "fmla v28.8h, v1.8h, v11.8h\n" "fmla v29.8h, v1.8h, v12.8h\n" - "ldr q12, [x24, x14]\n" - "fmla v28.8h, v1.8h, v9.8h\n" - "fmla v31.8h, v2.8h, v5.8h\n" - "fmla v30.8h, v2.8h, v6.8h\n" + "ldr q11, [x21, x10]\n" + "ldr x21, [x11, #0xe8]\n" + "fmla v30.8h, v1.8h, v6.8h\n" + "fmla v31.8h, v1.8h, v10.8h\n" + "ldr q1, [x16, #0xb0]\n" + "fmla v28.8h, v2.8h, v12.8h\n" "fmla v29.8h, v2.8h, v9.8h\n" - "ldr q9, [x23, x14]\n" - "add x14, x14, #0x10\n" - "fmla v28.8h, v2.8h, v11.8h\n" - "fmla v31.8h, v3.8h, v6.8h\n" - "fmla v30.8h, v3.8h, v8.8h\n" - "fmla v29.8h, v3.8h, v11.8h\n" - "fmla v28.8h, v3.8h, v12.8h\n" - "fmla v31.8h, v4.8h, v8.8h\n" - "fmla v30.8h, v4.8h, v10.8h\n" - "fmla v29.8h, v4.8h, v12.8h\n" - "fmla v28.8h, v4.8h, v9.8h\n" - "fmax v31.8h, v31.8h, v18.8h\n" - "fmax v30.8h, v30.8h, v18.8h\n" - "fmax v29.8h, v29.8h, v18.8h\n" - "fmin v31.8h, v31.8h, v17.8h\n" - "str q31, [x13, x28]\n" - "fmin v30.8h, v30.8h, v17.8h\n" - "fmin v29.8h, v29.8h, v17.8h\n" - "str q30, [x12, x28]\n" + "ldr q12, [x20, x10]\n" + "ldr x20, [x11, #0xf0]\n" + "fmla v30.8h, v2.8h, v10.8h\n" + "fmla v31.8h, v2.8h, v11.8h\n" + "ldr q2, [x16, #0xc0]\n" + "fmla v28.8h, v3.8h, v9.8h\n" + "fmla v29.8h, v3.8h, v13.8h\n" + "ldr q9, [x26, x10]\n" + "ldr x26, [x11, #0x100]\n" + "fmla v30.8h, v3.8h, v11.8h\n" + "fmla v31.8h, v3.8h, v12.8h\n" + "ldr q3, [x16, #0xd0]\n" + "fmla v28.8h, v4.8h, v13.8h\n" + "fmla v29.8h, v4.8h, v8.8h\n" + "ldr q13, [x25, x10]\n" + "ldr q8, [x22, x10]\n" + "fmla v30.8h, v4.8h, v12.8h\n" + "fmla v31.8h, v4.8h, v14.8h\n" + "ldr q4, [x16, #0xe0]\n" + "ldr x25, [x11, #0x108]\n" + "fmla v28.8h, v0.8h, v5.8h\n" + "fmla v29.8h, v0.8h, v6.8h\n" + "ldr q5, [x24, x10]\n" + "ldr x24, [x11, #0x110]\n" + "fmla v30.8h, v0.8h, v9.8h\n" + "fmla v31.8h, v0.8h, v13.8h\n" + "ldr q0, [x16, #0xf0]\n" + "fmla v28.8h, v1.8h, v6.8h\n" + "fmla v29.8h, v1.8h, v10.8h\n" + "ldr q6, [x23, x10]\n" + "ldr x23, [x11, #0x118]\n" + "fmla v30.8h, v1.8h, v13.8h\n" + "fmla v31.8h, v1.8h, v5.8h\n" + "ldr q1, [x16, #0x100]\n" + "fmla v28.8h, v2.8h, v10.8h\n" + "fmla v29.8h, v2.8h, v11.8h\n" + "ldr q10, [x21, x10]\n" + "fmla v30.8h, v2.8h, v5.8h\n" + "fmla v31.8h, v2.8h, v6.8h\n" + "ldr q2, [x16, #0x110]\n" + "fmla v28.8h, v3.8h, v11.8h\n" + "fmla v29.8h, v3.8h, v12.8h\n" + "ldr q11, [x20, x10]\n" + "fmla v30.8h, v3.8h, v6.8h\n" + "fmla v31.8h, v3.8h, v8.8h\n" + "ldr q3, [x16, #0x120]\n" + "fmla v28.8h, v4.8h, v12.8h\n" + "fmla v29.8h, v4.8h, v14.8h\n" + "ldr q12, [x19, x10]\n" + "fmla v30.8h, v4.8h, v8.8h\n" + "fmla v31.8h, v4.8h, v10.8h\n" + "ldr q4, [x16, #0x130]\n" + "add x16, x16, #0x140\n" + "fmla v28.8h, v0.8h, v9.8h\n" + "fmla v29.8h, v0.8h, v13.8h\n" + "ldr q9, [x26, x10]\n" + "fmla v30.8h, v0.8h, v11.8h\n" + "fmla v31.8h, v0.8h, v12.8h\n" + "ldr q11, [x25, x10]\n" + "fmla v28.8h, v1.8h, v13.8h\n" + "fmla v29.8h, v1.8h, v5.8h\n" + "fmla v30.8h, v1.8h, v12.8h\n" + "fmla v31.8h, v1.8h, v9.8h\n" + "ldr q12, [x24, x10]\n" + "fmla v28.8h, v2.8h, v5.8h\n" + "fmla v29.8h, v2.8h, v6.8h\n" + "fmla v30.8h, v2.8h, v9.8h\n" + "fmla v31.8h, v2.8h, v11.8h\n" + "ldr q9, [x23, x10]\n" + "add x10, x10, #0x10\n" + "fmla v28.8h, v3.8h, v6.8h\n" + "fmla v29.8h, v3.8h, v8.8h\n" + "fmla v30.8h, v3.8h, v11.8h\n" + "fmla v31.8h, v3.8h, v12.8h\n" + "fmla v28.8h, v4.8h, v8.8h\n" + "fmla v29.8h, v4.8h, v10.8h\n" "fmax v28.8h, v28.8h, v18.8h\n" - "str q29, [x10, x28]\n" + "fmla v30.8h, v4.8h, v12.8h\n" + "fmla v31.8h, v4.8h, v9.8h\n" + "fmax v29.8h, v29.8h, v18.8h\n" + "fmax v30.8h, v30.8h, v18.8h\n" + "fmax v31.8h, v31.8h, v18.8h\n" "fmin v28.8h, v28.8h, v17.8h\n" - "str q28, [x9, x28]\n" + "fmin v29.8h, v29.8h, v17.8h\n" + "str q28, [x15, x9]\n" + "fmin v30.8h, v30.8h, v17.8h\n" + "fmin v31.8h, v31.8h, v17.8h\n" + "str q29, [x14, x9]\n" + "str q30, [x13, x9]\n" + "str q31, [x12, x9]\n" "3:" // Oddments - "tst %x[n_channels], #0x1\n" - "beq 60f\n" - "ldr q16, [x15, #0x0]\n" - "ldr q0, [x15, #0x10]\n" - "mov x28, x14\n" - "ldr q1, [x15, #0x20]\n" - "add x13, x13, x28\n" - "ldr q2, [x15, #0x30]\n" - "add x12, x12, x28\n" - "ldr q3, [x15, #0x40]\n" - "add x10, x10, x28\n" - "ldr q4, [x15, #0x50]\n" - "add x9, x9, x28\n" - "ldr x24, [x16, #0x10]\n" - "ldr x23, [x16, #0x18]\n" - "ldr x22, [x16, #0x20]\n" - "add x24, x24, x14\n" - "ldr x21, [x16, #0x28]\n" - "add x23, x23, x14\n" - "ldr x20, [x16, #0x30]\n" - "add x22, x22, x14\n" - "ldr x19, [x16, #0x38]\n" - "add x21, x21, x14\n" - "ldr x26, [x16, #0x40]\n" - "add x20, x20, x14\n" - "ldr x25, [x16, #0x48]\n" - "add x19, x19, x14\n" - "add x26, x26, x14\n" - "add x25, x25, x14\n" - "add x15, x15, #0x60\n" + "tst %x[n_channels], #0x7\n" + "beq 116f\n" + "mov x9, x10\n" + "ldr x28, [x11, #0x0]\n" + "ldr x27, [x11, #0x8]\n" + "ldr x26, [x11, #0x10]\n" + "add x15, x15, x9\n" + "add x14, x14, x9\n" + "ldr x25, [x11, #0x18]\n" + "ldr x24, [x11, #0x20]\n" + "add x13, x13, x9\n" + "add x12, x12, x9\n" + "ldr x23, [x11, #0x28]\n" + "ldr x22, [x11, #0x30]\n" + "add x28, x28, x10\n" + "add x27, x27, x10\n" + "ldr x21, [x11, #0x38]\n" + "ldr x20, [x11, #0x40]\n" + "add x26, x26, x10\n" + "add x25, x25, x10\n" + "ldr x19, [x11, #0x48]\n" + "ldr q16, [x16, #0x0]\n" + "add x24, x24, x10\n" + "add x23, x23, x10\n" + "ldr q0, [x16, #0x10]\n" + "ldr q1, [x16, #0x20]\n" + "add x22, x22, x10\n" + "add x21, x21, x10\n" + "ldr q2, [x16, #0x30]\n" + "ldr q3, [x16, #0x40]\n" + "add x20, x20, x10\n" + "add x19, x19, x10\n" + "ldr q4, [x16, #0x50]\n" + "add x16, x16, #0x60\n" + "tbz %x[n_channels], #2, 5f\n" + "ld1 { v5.d }[0], [x28], #0x8\n" + "ld1 { v6.d }[0], [x27], #0x8\n" + "ld1 { v7.d }[0], [x26], #0x8\n" + "ld1 { v8.d }[0], [x25], #0x8\n" + "ld1 { v9.d }[0], [x24], #0x8\n" + "ld1 { v13.d }[0], [x23], #0x8\n" + "ld1 { v11.d }[0], [x22], #0x8\n" + "ld1 { v12.d }[0], [x21], #0x8\n" + "ld1 { v10.d }[0], [x20], #0x8\n" + "ld1 { v14.d }[0], [x19], #0x8\n" "tbz %x[n_channels], #1, 4f\n" - "ld1 { v5.s }[0], [x26], #0x4\n" - "ld1 { v6.s }[0], [x25], #0x4\n" - "ld1 { v7.s }[0], [x24], #0x4\n" - "ld1 { v8.s }[0], [x23], #0x4\n" - "ld1 { v9.s }[0], [x22], #0x4\n" - "ld1 { v13.s }[0], [x21], #0x4\n" - "ld1 { v11.s }[0], [x20], #0x4\n" - "ld1 { v12.s }[0], [x19], #0x4\n" - "ld1 { v10.s }[0], [x26], #0x4\n" - "ld1 { v14.s }[0], [x25], #0x4\n" - "tbz %x[n_channels], #0, 5f\n" - "ld1 { v7.h }[2], [x24], #0x2\n" - "ld1 { v8.h }[2], [x23], #0x2\n" - "ld1 { v5.h }[2], [x26], #0x2\n" - "ld1 { v6.h }[2], [x25], #0x2\n" - "ld1 { v9.h }[2], [x22], #0x2\n" - "ld1 { v13.h }[2], [x21], #0x2\n" - "ld1 { v11.h }[2], [x20], #0x2\n" - "ld1 { v12.h }[2], [x19], #0x2\n" - "ld1 { v10.h }[2], [x26], #0x2\n" - "ld1 { v14.h }[2], [x25], #0x2\n" - "b 5f\n" - "4:" // Oddments: Load inputs (0, 0), (0, 1), (1, 0), (1, 1), (0, 2), (1, 2), (0, 3), (0, 4), (1, 5), (2, 0): Bit 1: Unset - "ld1 { v5.h }[0], [x26], #0x2\n" - "ld1 { v6.h }[0], [x25], #0x2\n" - "ld1 { v7.h }[0], [x24], #0x2\n" - "ld1 { v8.h }[0], [x23], #0x2\n" - "ld1 { v9.h }[0], [x22], #0x2\n" - "ld1 { v13.h }[0], [x21], #0x2\n" - "ld1 { v11.h }[0], [x20], #0x2\n" - "ld1 { v12.h }[0], [x19], #0x2\n" - "ld1 { v10.h }[0], [x26], #0x2\n" - "ld1 { v14.h }[0], [x25], #0x2\n" - "5:" // Oddments: Load inputs (0, 0), (0, 1), (1, 0), (1, 1), (0, 2), (1, 2), (0, 3), (0, 4), (1, 5), (2, 0): Bit 1: End - "mov v31.16b, v16.16b\n fmla v31.8h, v0.8h, v5.8h\n" - "ldr x24, [x16, #0x50]\n" - "mov v30.16b, v16.16b\n fmla v30.8h, v0.8h, v6.8h\n" - "add x24, x24, x14\n" - "mov v29.16b, v16.16b\n fmla v29.8h, v0.8h, v7.8h\n" - "mov v28.16b, v16.16b\n fmla v28.8h, v0.8h, v8.8h\n" - "fmla v31.8h, v1.8h, v6.8h\n" - "fmla v30.8h, v1.8h, v9.8h\n" - "fmla v29.8h, v1.8h, v8.8h\n" - "fmla v28.8h, v1.8h, v13.8h\n" - "fmla v31.8h, v2.8h, v9.8h\n" - "fmla v30.8h, v2.8h, v11.8h\n" - "fmla v29.8h, v2.8h, v13.8h\n" + "ld1 { v5.s }[2], [x28], #0x4\n" + "ld1 { v6.s }[2], [x27], #0x4\n" + "ld1 { v7.s }[2], [x26], #0x4\n" + "ld1 { v8.s }[2], [x25], #0x4\n" + "ld1 { v9.s }[2], [x24], #0x4\n" + "ld1 { v13.s }[2], [x23], #0x4\n" + "ld1 { v11.s }[2], [x22], #0x4\n" + "ld1 { v12.s }[2], [x21], #0x4\n" + "ld1 { v10.s }[2], [x20], #0x4\n" + "ld1 { v14.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 7f\n" + "ld1 { v5.h }[6], [x28], #0x2\n" + "ld1 { v6.h }[6], [x27], #0x2\n" + "ld1 { v7.h }[6], [x26], #0x2\n" + "ld1 { v8.h }[6], [x25], #0x2\n" + "ld1 { v9.h }[6], [x24], #0x2\n" + "ld1 { v13.h }[6], [x23], #0x2\n" + "ld1 { v11.h }[6], [x22], #0x2\n" + "ld1 { v12.h }[6], [x21], #0x2\n" + "ld1 { v10.h }[6], [x20], #0x2\n" + "ld1 { v14.h }[6], [x19], #0x2\n" + "b 7f\n" + "4:" // Oddments: Load inputs (0, 0), (0, 1), (1, 0), (1, 1), (0, 2), (1, 2), (0, 3), (0, 4), (1, 5), (2, 0): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 7f\n" + "ld1 { v5.h }[4], [x28], #0x2\n" + "ld1 { v6.h }[4], [x27], #0x2\n" + "ld1 { v7.h }[4], [x26], #0x2\n" + "ld1 { v8.h }[4], [x25], #0x2\n" + "ld1 { v9.h }[4], [x24], #0x2\n" + "ld1 { v13.h }[4], [x23], #0x2\n" + "ld1 { v11.h }[4], [x22], #0x2\n" + "ld1 { v12.h }[4], [x21], #0x2\n" + "ld1 { v10.h }[4], [x20], #0x2\n" + "ld1 { v14.h }[4], [x19], #0x2\n" + "b 7f\n" + "5:" // Oddments: Load inputs (0, 0), (0, 1), (1, 0), (1, 1), (0, 2), (1, 2), (0, 3), (0, 4), (1, 5), (2, 0): Bit 2: Unset "tbz %x[n_channels], #1, 6f\n" - "ld1 { v5.s }[0], [x24], #0x4\n" + "ld1 { v5.s }[0], [x28], #0x4\n" + "ld1 { v6.s }[0], [x27], #0x4\n" + "ld1 { v7.s }[0], [x26], #0x4\n" + "ld1 { v8.s }[0], [x25], #0x4\n" + "ld1 { v9.s }[0], [x24], #0x4\n" + "ld1 { v13.s }[0], [x23], #0x4\n" + "ld1 { v11.s }[0], [x22], #0x4\n" + "ld1 { v12.s }[0], [x21], #0x4\n" + "ld1 { v10.s }[0], [x20], #0x4\n" + "ld1 { v14.s }[0], [x19], #0x4\n" "tbz %x[n_channels], #0, 7f\n" - "ld1 { v5.h }[2], [x24], #0x2\n" + "ld1 { v5.h }[2], [x28], #0x2\n" + "ld1 { v6.h }[2], [x27], #0x2\n" + "ld1 { v7.h }[2], [x26], #0x2\n" + "ld1 { v8.h }[2], [x25], #0x2\n" + "ld1 { v9.h }[2], [x24], #0x2\n" + "ld1 { v13.h }[2], [x23], #0x2\n" + "ld1 { v11.h }[2], [x22], #0x2\n" + "ld1 { v12.h }[2], [x21], #0x2\n" + "ld1 { v10.h }[2], [x20], #0x2\n" + "ld1 { v14.h }[2], [x19], #0x2\n" "b 7f\n" - "6:" // Oddments: Load input (1, 3): Bit 1: Unset - "ld1 { v5.h }[0], [x24], #0x2\n" - "7:" // Oddments: Load input (1, 3): Bit 1: End - "fmla v28.8h, v2.8h, v5.8h\n" - "ldr x23, [x16, #0x58]\n" - "fmla v31.8h, v3.8h, v11.8h\n" - "add x23, x23, x14\n" - "fmla v30.8h, v3.8h, v12.8h\n" - "fmla v29.8h, v3.8h, v5.8h\n" + "6:" // Oddments: Load inputs (0, 0), (0, 1), (1, 0), (1, 1), (0, 2), (1, 2), (0, 3), (0, 4), (1, 5), (2, 0): Bit 2: Unset: Bit 1: Unset + "ld1 { v5.h }[0], [x28], #0x2\n" + "ld1 { v6.h }[0], [x27], #0x2\n" + "ld1 { v7.h }[0], [x26], #0x2\n" + "ld1 { v8.h }[0], [x25], #0x2\n" + "ld1 { v9.h }[0], [x24], #0x2\n" + "ld1 { v13.h }[0], [x23], #0x2\n" + "ld1 { v11.h }[0], [x22], #0x2\n" + "ld1 { v12.h }[0], [x21], #0x2\n" + "ld1 { v10.h }[0], [x20], #0x2\n" + "ld1 { v14.h }[0], [x19], #0x2\n" + "7:" // Oddments: Load inputs (0, 0), (0, 1), (1, 0), (1, 1), (0, 2), (1, 2), (0, 3), (0, 4), (1, 5), (2, 0): Bit 2: End + "mov v28.16b, v16.16b\n fmla v28.8h, v0.8h, v5.8h\n" + "mov v29.16b, v16.16b\n fmla v29.8h, v0.8h, v6.8h\n" + "ldr x19, [x11, #0x50]\n" + "add x19, x19, x10\n" + "mov v30.16b, v16.16b\n fmla v30.8h, v0.8h, v7.8h\n" + "mov v31.16b, v16.16b\n fmla v31.8h, v0.8h, v8.8h\n" + "fmla v28.8h, v1.8h, v6.8h\n" + "fmla v29.8h, v1.8h, v9.8h\n" + "fmla v30.8h, v1.8h, v8.8h\n" + "fmla v31.8h, v1.8h, v13.8h\n" + "fmla v28.8h, v2.8h, v9.8h\n" + "fmla v29.8h, v2.8h, v11.8h\n" + "fmla v30.8h, v2.8h, v13.8h\n" + "tbz %x[n_channels], #2, 9f\n" + "ld1 { v5.d }[0], [x19], #0x8\n" "tbz %x[n_channels], #1, 8f\n" - "ld1 { v6.s }[0], [x23], #0x4\n" - "tbz %x[n_channels], #0, 9f\n" - "ld1 { v6.h }[2], [x23], #0x2\n" - "b 9f\n" - "8:" // Oddments: Load input (1, 4): Bit 1: Unset - "ld1 { v6.h }[0], [x23], #0x2\n" - "9:" // Oddments: Load input (1, 4): Bit 1: End - "fmla v28.8h, v3.8h, v6.8h\n" - "ldr x22, [x16, #0x60]\n" - "fmla v31.8h, v4.8h, v12.8h\n" - "add x22, x22, x14\n" + "ld1 { v5.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 11f\n" + "ld1 { v5.h }[6], [x19], #0x2\n" + "b 11f\n" + "8:" // Oddments: Load input (1, 3): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 11f\n" + "ld1 { v5.h }[4], [x19], #0x2\n" + "b 11f\n" + "9:" // Oddments: Load input (1, 3): Bit 2: Unset "tbz %x[n_channels], #1, 10f\n" - "ld1 { v9.s }[0], [x22], #0x4\n" + "ld1 { v5.s }[0], [x19], #0x4\n" "tbz %x[n_channels], #0, 11f\n" - "ld1 { v9.h }[2], [x22], #0x2\n" + "ld1 { v5.h }[2], [x19], #0x2\n" "b 11f\n" - "10:" // Oddments: Load input (0, 5): Bit 1: Unset - "ld1 { v9.h }[0], [x22], #0x2\n" - "11:" // Oddments: Load input (0, 5): Bit 1: End - "fmla v30.8h, v4.8h, v9.8h\n" - "ldr h0, [x15, #0xc]\n" - "fmla v29.8h, v4.8h, v6.8h\n" - "ldr x21, [x16, #0x68]\n" - "add x21, x21, x14\n" - "fmla v28.8h, v4.8h, v10.8h\n" - "fmla v31.8h, v0.8h, v7.8h\n" - "fmla v30.8h, v0.8h, v8.8h\n" - "fmla v29.8h, v0.8h, v14.8h\n" + "10:" // Oddments: Load input (1, 3): Bit 2: Unset: Bit 1: Unset + "ld1 { v5.h }[0], [x19], #0x2\n" + "11:" // Oddments: Load input (1, 3): Bit 2: End + "ldr x19, [x11, #0x58]\n" + "fmla v31.8h, v2.8h, v5.8h\n" + "fmla v28.8h, v3.8h, v11.8h\n" + "add x19, x19, x10\n" + "fmla v29.8h, v3.8h, v12.8h\n" + "fmla v30.8h, v3.8h, v5.8h\n" + "tbz %x[n_channels], #2, 13f\n" + "ld1 { v6.d }[0], [x19], #0x8\n" "tbz %x[n_channels], #1, 12f\n" - "ld1 { v11.s }[0], [x21], #0x4\n" - "tbz %x[n_channels], #0, 13f\n" - "ld1 { v11.h }[2], [x21], #0x2\n" - "b 13f\n" - "12:" // Oddments: Load input (2, 1): Bit 1: Unset - "ld1 { v11.h }[0], [x21], #0x2\n" - "13:" // Oddments: Load input (2, 1): Bit 1: End - "fmla v28.8h, v0.8h, v11.8h\n" - "ldr h1, [x15, #0xe]\n" - "fmla v31.8h, v1.8h, v8.8h\n" - "ldr x20, [x16, #0x70]\n" - "add x20, x20, x14\n" - "fmla v30.8h, v1.8h, v13.8h\n" - "fmla v29.8h, v1.8h, v11.8h\n" + "ld1 { v6.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 15f\n" + "ld1 { v6.h }[6], [x19], #0x2\n" + "b 15f\n" + "12:" // Oddments: Load input (1, 4): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 15f\n" + "ld1 { v6.h }[4], [x19], #0x2\n" + "b 15f\n" + "13:" // Oddments: Load input (1, 4): Bit 2: Unset "tbz %x[n_channels], #1, 14f\n" - "ld1 { v12.s }[0], [x20], #0x4\n" + "ld1 { v6.s }[0], [x19], #0x4\n" "tbz %x[n_channels], #0, 15f\n" - "ld1 { v12.h }[2], [x20], #0x2\n" + "ld1 { v6.h }[2], [x19], #0x2\n" "b 15f\n" - "14:" // Oddments: Load input (2, 2): Bit 1: Unset - "ld1 { v12.h }[0], [x20], #0x2\n" - "15:" // Oddments: Load input (2, 2): Bit 1: End - "fmla v28.8h, v1.8h, v12.8h\n" - "ldr h2, [x15, #0x10]\n" - "fmla v31.8h, v2.8h, v13.8h\n" - "ldr x19, [x16, #0x78]\n" - "add x19, x19, x14\n" - "fmla v30.8h, v2.8h, v5.8h\n" - "fmla v29.8h, v2.8h, v12.8h\n" + "14:" // Oddments: Load input (1, 4): Bit 2: Unset: Bit 1: Unset + "ld1 { v6.h }[0], [x19], #0x2\n" + "15:" // Oddments: Load input (1, 4): Bit 2: End + "ldr x19, [x11, #0x60]\n" + "fmla v31.8h, v3.8h, v6.8h\n" + "fmla v28.8h, v4.8h, v12.8h\n" + "add x19, x19, x10\n" + "tbz %x[n_channels], #2, 17f\n" + "ld1 { v9.d }[0], [x19], #0x8\n" "tbz %x[n_channels], #1, 16f\n" - "ld1 { v9.s }[0], [x19], #0x4\n" - "tbz %x[n_channels], #0, 17f\n" - "ld1 { v9.h }[2], [x19], #0x2\n" - "b 17f\n" - "16:" // Oddments: Load input (2, 3): Bit 1: Unset - "ld1 { v9.h }[0], [x19], #0x2\n" - "17:" // Oddments: Load input (2, 3): Bit 1: End - "fmla v28.8h, v2.8h, v9.8h\n" - "ldr h3, [x15, #0x12]\n" - "fmla v31.8h, v3.8h, v5.8h\n" - "ldr x26, [x16, #0x80]\n" - "add x26, x26, x14\n" - "fmla v30.8h, v3.8h, v6.8h\n" - "fmla v29.8h, v3.8h, v9.8h\n" + "ld1 { v9.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 19f\n" + "ld1 { v9.h }[6], [x19], #0x2\n" + "b 19f\n" + "16:" // Oddments: Load input (0, 5): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 19f\n" + "ld1 { v9.h }[4], [x19], #0x2\n" + "b 19f\n" + "17:" // Oddments: Load input (0, 5): Bit 2: Unset "tbz %x[n_channels], #1, 18f\n" - "ld1 { v13.s }[0], [x26], #0x4\n" + "ld1 { v9.s }[0], [x19], #0x4\n" "tbz %x[n_channels], #0, 19f\n" - "ld1 { v13.h }[2], [x26], #0x2\n" + "ld1 { v9.h }[2], [x19], #0x2\n" "b 19f\n" - "18:" // Oddments: Load input (2, 4): Bit 1: Unset - "ld1 { v13.h }[0], [x26], #0x2\n" - "19:" // Oddments: Load input (2, 4): Bit 1: End - "fmla v28.8h, v3.8h, v13.8h\n" - "ldr h4, [x15, #0x14]\n" - "fmla v31.8h, v4.8h, v6.8h\n" - "ldr x25, [x16, #0x88]\n" - "add x25, x25, x14\n" - "fmla v30.8h, v4.8h, v10.8h\n" - "fmla v29.8h, v4.8h, v13.8h\n" + "18:" // Oddments: Load input (0, 5): Bit 2: Unset: Bit 1: Unset + "ld1 { v9.h }[0], [x19], #0x2\n" + "19:" // Oddments: Load input (0, 5): Bit 2: End + "fmla v29.8h, v4.8h, v9.8h\n" + "fmla v30.8h, v4.8h, v6.8h\n" + "ldr q0, [x16, #0x0]\n" + "ldr x19, [x11, #0x68]\n" + "fmla v31.8h, v4.8h, v10.8h\n" + "fmla v28.8h, v0.8h, v7.8h\n" + "add x19, x19, x10\n" + "fmla v29.8h, v0.8h, v8.8h\n" + "fmla v30.8h, v0.8h, v14.8h\n" + "add x16, x16, #0x10\n" + "tbz %x[n_channels], #2, 21f\n" + "ld1 { v11.d }[0], [x19], #0x8\n" "tbz %x[n_channels], #1, 20f\n" - "ld1 { v8.s }[0], [x25], #0x4\n" - "tbz %x[n_channels], #0, 21f\n" - "ld1 { v8.h }[2], [x25], #0x2\n" - "b 21f\n" - "20:" // Oddments: Load input (2, 5): Bit 1: Unset - "ld1 { v8.h }[0], [x25], #0x2\n" - "21:" // Oddments: Load input (2, 5): Bit 1: End - "fmla v28.8h, v4.8h, v8.8h\n" - "ldr h0, [x15, #0x16]\n" - "fmla v31.8h, v0.8h, v14.8h\n" - "ldr x24, [x16, #0x90]\n" - "add x24, x24, x14\n" - "fmla v30.8h, v0.8h, v11.8h\n" + "ld1 { v11.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 23f\n" + "ld1 { v11.h }[6], [x19], #0x2\n" + "b 23f\n" + "20:" // Oddments: Load input (2, 1): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 23f\n" + "ld1 { v11.h }[4], [x19], #0x2\n" + "b 23f\n" + "21:" // Oddments: Load input (2, 1): Bit 2: Unset "tbz %x[n_channels], #1, 22f\n" - "ld1 { v5.s }[0], [x24], #0x4\n" + "ld1 { v11.s }[0], [x19], #0x4\n" "tbz %x[n_channels], #0, 23f\n" - "ld1 { v5.h }[2], [x24], #0x2\n" + "ld1 { v11.h }[2], [x19], #0x2\n" "b 23f\n" - "22:" // Oddments: Load input (3, 0): Bit 1: Unset - "ld1 { v5.h }[0], [x24], #0x2\n" - "23:" // Oddments: Load input (3, 0): Bit 1: End - "fmla v29.8h, v0.8h, v5.8h\n" - "ldr x23, [x16, #0x98]\n" - "add x23, x23, x14\n" + "22:" // Oddments: Load input (2, 1): Bit 2: Unset: Bit 1: Unset + "ld1 { v11.h }[0], [x19], #0x2\n" + "23:" // Oddments: Load input (2, 1): Bit 2: End + "ldr q1, [x16, #0x0]\n" + "ldr x19, [x11, #0x70]\n" + "fmla v31.8h, v0.8h, v11.8h\n" + "fmla v28.8h, v1.8h, v8.8h\n" + "fmla v29.8h, v1.8h, v13.8h\n" + "fmla v30.8h, v1.8h, v11.8h\n" + "add x19, x19, x10\n" + "add x16, x16, #0x10\n" + "tbz %x[n_channels], #2, 25f\n" + "ld1 { v12.d }[0], [x19], #0x8\n" "tbz %x[n_channels], #1, 24f\n" - "ld1 { v6.s }[0], [x23], #0x4\n" - "tbz %x[n_channels], #0, 25f\n" - "ld1 { v6.h }[2], [x23], #0x2\n" - "b 25f\n" - "24:" // Oddments: Load input (3, 1): Bit 1: Unset - "ld1 { v6.h }[0], [x23], #0x2\n" - "25:" // Oddments: Load input (3, 1): Bit 1: End - "fmla v28.8h, v0.8h, v6.8h\n" - "ldr h1, [x15, #0x18]\n" - "fmla v31.8h, v1.8h, v11.8h\n" - "ldr x22, [x16, #0xa0]\n" - "add x22, x22, x14\n" - "fmla v30.8h, v1.8h, v12.8h\n" - "fmla v29.8h, v1.8h, v6.8h\n" + "ld1 { v12.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 27f\n" + "ld1 { v12.h }[6], [x19], #0x2\n" + "b 27f\n" + "24:" // Oddments: Load input (2, 2): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 27f\n" + "ld1 { v12.h }[4], [x19], #0x2\n" + "b 27f\n" + "25:" // Oddments: Load input (2, 2): Bit 2: Unset "tbz %x[n_channels], #1, 26f\n" - "ld1 { v10.s }[0], [x22], #0x4\n" + "ld1 { v12.s }[0], [x19], #0x4\n" "tbz %x[n_channels], #0, 27f\n" - "ld1 { v10.h }[2], [x22], #0x2\n" + "ld1 { v12.h }[2], [x19], #0x2\n" "b 27f\n" - "26:" // Oddments: Load input (3, 2): Bit 1: Unset - "ld1 { v10.h }[0], [x22], #0x2\n" - "27:" // Oddments: Load input (3, 2): Bit 1: End - "fmla v28.8h, v1.8h, v10.8h\n" - "ldr h2, [x15, #0x1a]\n" - "fmla v31.8h, v2.8h, v12.8h\n" - "ldr x21, [x16, #0xa8]\n" - "add x21, x21, x14\n" - "fmla v30.8h, v2.8h, v9.8h\n" - "fmla v29.8h, v2.8h, v10.8h\n" + "26:" // Oddments: Load input (2, 2): Bit 2: Unset: Bit 1: Unset + "ld1 { v12.h }[0], [x19], #0x2\n" + "27:" // Oddments: Load input (2, 2): Bit 2: End + "ldr q2, [x16, #0x0]\n" + "ldr x19, [x11, #0x78]\n" + "fmla v31.8h, v1.8h, v12.8h\n" + "fmla v28.8h, v2.8h, v13.8h\n" + "fmla v29.8h, v2.8h, v5.8h\n" + "fmla v30.8h, v2.8h, v12.8h\n" + "add x19, x19, x10\n" + "add x16, x16, #0x10\n" + "tbz %x[n_channels], #2, 29f\n" + "ld1 { v9.d }[0], [x19], #0x8\n" "tbz %x[n_channels], #1, 28f\n" - "ld1 { v11.s }[0], [x21], #0x4\n" - "tbz %x[n_channels], #0, 29f\n" - "ld1 { v11.h }[2], [x21], #0x2\n" - "b 29f\n" - "28:" // Oddments: Load input (3, 3): Bit 1: Unset - "ld1 { v11.h }[0], [x21], #0x2\n" - "29:" // Oddments: Load input (3, 3): Bit 1: End - "fmla v28.8h, v2.8h, v11.8h\n" - "ldr h3, [x15, #0x1c]\n" - "fmla v31.8h, v3.8h, v9.8h\n" - "ldr x20, [x16, #0xb0]\n" - "add x20, x20, x14\n" - "fmla v30.8h, v3.8h, v13.8h\n" - "fmla v29.8h, v3.8h, v11.8h\n" + "ld1 { v9.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 31f\n" + "ld1 { v9.h }[6], [x19], #0x2\n" + "b 31f\n" + "28:" // Oddments: Load input (2, 3): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 31f\n" + "ld1 { v9.h }[4], [x19], #0x2\n" + "b 31f\n" + "29:" // Oddments: Load input (2, 3): Bit 2: Unset "tbz %x[n_channels], #1, 30f\n" - "ld1 { v12.s }[0], [x20], #0x4\n" + "ld1 { v9.s }[0], [x19], #0x4\n" "tbz %x[n_channels], #0, 31f\n" - "ld1 { v12.h }[2], [x20], #0x2\n" + "ld1 { v9.h }[2], [x19], #0x2\n" "b 31f\n" - "30:" // Oddments: Load input (3, 4): Bit 1: Unset - "ld1 { v12.h }[0], [x20], #0x2\n" - "31:" // Oddments: Load input (3, 4): Bit 1: End - "fmla v28.8h, v3.8h, v12.8h\n" - "ldr h4, [x15, #0x1e]\n" - "fmla v31.8h, v4.8h, v13.8h\n" - "ldr x19, [x16, #0xb8]\n" - "add x19, x19, x14\n" - "fmla v30.8h, v4.8h, v8.8h\n" - "fmla v29.8h, v4.8h, v12.8h\n" + "30:" // Oddments: Load input (2, 3): Bit 2: Unset: Bit 1: Unset + "ld1 { v9.h }[0], [x19], #0x2\n" + "31:" // Oddments: Load input (2, 3): Bit 2: End + "ldr q3, [x16, #0x0]\n" + "ldr x19, [x11, #0x80]\n" + "fmla v31.8h, v2.8h, v9.8h\n" + "fmla v28.8h, v3.8h, v5.8h\n" + "fmla v29.8h, v3.8h, v6.8h\n" + "fmla v30.8h, v3.8h, v9.8h\n" + "add x19, x19, x10\n" + "add x16, x16, #0x10\n" + "tbz %x[n_channels], #2, 33f\n" + "ld1 { v13.d }[0], [x19], #0x8\n" "tbz %x[n_channels], #1, 32f\n" - "ld1 { v14.s }[0], [x19], #0x4\n" - "tbz %x[n_channels], #0, 33f\n" - "ld1 { v14.h }[2], [x19], #0x2\n" - "b 33f\n" - "32:" // Oddments: Load input (3, 5): Bit 1: Unset - "ld1 { v14.h }[0], [x19], #0x2\n" - "33:" // Oddments: Load input (3, 5): Bit 1: End - "fmla v28.8h, v4.8h, v14.8h\n" - "ldr h0, [x15, #0x20]\n" - "fmla v31.8h, v0.8h, v5.8h\n" - "ldr x26, [x16, #0xc0]\n" - "add x26, x26, x14\n" - "fmla v30.8h, v0.8h, v6.8h\n" + "ld1 { v13.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 35f\n" + "ld1 { v13.h }[6], [x19], #0x2\n" + "b 35f\n" + "32:" // Oddments: Load input (2, 4): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 35f\n" + "ld1 { v13.h }[4], [x19], #0x2\n" + "b 35f\n" + "33:" // Oddments: Load input (2, 4): Bit 2: Unset "tbz %x[n_channels], #1, 34f\n" - "ld1 { v9.s }[0], [x26], #0x4\n" + "ld1 { v13.s }[0], [x19], #0x4\n" "tbz %x[n_channels], #0, 35f\n" - "ld1 { v9.h }[2], [x26], #0x2\n" + "ld1 { v13.h }[2], [x19], #0x2\n" "b 35f\n" - "34:" // Oddments: Load input (4, 0): Bit 1: Unset - "ld1 { v9.h }[0], [x26], #0x2\n" - "35:" // Oddments: Load input (4, 0): Bit 1: End - "fmla v29.8h, v0.8h, v9.8h\n" - "ldr x25, [x16, #0xc8]\n" - "add x25, x25, x14\n" + "34:" // Oddments: Load input (2, 4): Bit 2: Unset: Bit 1: Unset + "ld1 { v13.h }[0], [x19], #0x2\n" + "35:" // Oddments: Load input (2, 4): Bit 2: End + "ldr q4, [x16, #0x0]\n" + "ldr x19, [x11, #0x88]\n" + "fmla v31.8h, v3.8h, v13.8h\n" + "fmla v28.8h, v4.8h, v6.8h\n" + "fmla v29.8h, v4.8h, v10.8h\n" + "fmla v30.8h, v4.8h, v13.8h\n" + "add x19, x19, x10\n" + "add x16, x16, #0x10\n" + "tbz %x[n_channels], #2, 37f\n" + "ld1 { v8.d }[0], [x19], #0x8\n" "tbz %x[n_channels], #1, 36f\n" - "ld1 { v13.s }[0], [x25], #0x4\n" - "tbz %x[n_channels], #0, 37f\n" - "ld1 { v13.h }[2], [x25], #0x2\n" - "b 37f\n" - "36:" // Oddments: Load input (4, 1): Bit 1: Unset - "ld1 { v13.h }[0], [x25], #0x2\n" - "37:" // Oddments: Load input (4, 1): Bit 1: End - "fmla v28.8h, v0.8h, v13.8h\n" - "ldr h1, [x15, #0x22]\n" - "fmla v31.8h, v1.8h, v6.8h\n" - "ldr x24, [x16, #0xd0]\n" - "add x24, x24, x14\n" - "fmla v30.8h, v1.8h, v10.8h\n" - "fmla v29.8h, v1.8h, v13.8h\n" + "ld1 { v8.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 39f\n" + "ld1 { v8.h }[6], [x19], #0x2\n" + "b 39f\n" + "36:" // Oddments: Load input (2, 5): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 39f\n" + "ld1 { v8.h }[4], [x19], #0x2\n" + "b 39f\n" + "37:" // Oddments: Load input (2, 5): Bit 2: Unset "tbz %x[n_channels], #1, 38f\n" - "ld1 { v5.s }[0], [x24], #0x4\n" + "ld1 { v8.s }[0], [x19], #0x4\n" "tbz %x[n_channels], #0, 39f\n" - "ld1 { v5.h }[2], [x24], #0x2\n" + "ld1 { v8.h }[2], [x19], #0x2\n" "b 39f\n" - "38:" // Oddments: Load input (4, 2): Bit 1: Unset - "ld1 { v5.h }[0], [x24], #0x2\n" - "39:" // Oddments: Load input (4, 2): Bit 1: End - "fmla v28.8h, v1.8h, v5.8h\n" - "ldr h2, [x15, #0x24]\n" - "fmla v31.8h, v2.8h, v10.8h\n" - "ldr x23, [x16, #0xd8]\n" - "add x23, x23, x14\n" - "fmla v30.8h, v2.8h, v11.8h\n" - "fmla v29.8h, v2.8h, v5.8h\n" + "38:" // Oddments: Load input (2, 5): Bit 2: Unset: Bit 1: Unset + "ld1 { v8.h }[0], [x19], #0x2\n" + "39:" // Oddments: Load input (2, 5): Bit 2: End + "ldr q0, [x16, #0x0]\n" + "ldr x19, [x11, #0x90]\n" + "fmla v31.8h, v4.8h, v8.8h\n" + "fmla v28.8h, v0.8h, v14.8h\n" + "fmla v29.8h, v0.8h, v11.8h\n" + "add x19, x19, x10\n" + "add x16, x16, #0x10\n" + "tbz %x[n_channels], #2, 41f\n" + "ld1 { v5.d }[0], [x19], #0x8\n" "tbz %x[n_channels], #1, 40f\n" - "ld1 { v6.s }[0], [x23], #0x4\n" - "tbz %x[n_channels], #0, 41f\n" - "ld1 { v6.h }[2], [x23], #0x2\n" - "b 41f\n" - "40:" // Oddments: Load input (4, 3): Bit 1: Unset - "ld1 { v6.h }[0], [x23], #0x2\n" - "41:" // Oddments: Load input (4, 3): Bit 1: End - "fmla v28.8h, v2.8h, v6.8h\n" - "ldr h3, [x15, #0x26]\n" - "fmla v31.8h, v3.8h, v11.8h\n" - "ldr x22, [x16, #0xe0]\n" - "add x22, x22, x14\n" - "fmla v30.8h, v3.8h, v12.8h\n" - "fmla v29.8h, v3.8h, v6.8h\n" + "ld1 { v5.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 43f\n" + "ld1 { v5.h }[6], [x19], #0x2\n" + "b 43f\n" + "40:" // Oddments: Load input (3, 0): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 43f\n" + "ld1 { v5.h }[4], [x19], #0x2\n" + "b 43f\n" + "41:" // Oddments: Load input (3, 0): Bit 2: Unset "tbz %x[n_channels], #1, 42f\n" - "ld1 { v8.s }[0], [x22], #0x4\n" + "ld1 { v5.s }[0], [x19], #0x4\n" "tbz %x[n_channels], #0, 43f\n" - "ld1 { v8.h }[2], [x22], #0x2\n" + "ld1 { v5.h }[2], [x19], #0x2\n" "b 43f\n" - "42:" // Oddments: Load input (4, 4): Bit 1: Unset - "ld1 { v8.h }[0], [x22], #0x2\n" - "43:" // Oddments: Load input (4, 4): Bit 1: End - "fmla v28.8h, v3.8h, v8.8h\n" - "ldr h4, [x15, #0x28]\n" - "fmla v31.8h, v4.8h, v12.8h\n" - "ldr x21, [x16, #0xe8]\n" - "add x21, x21, x14\n" - "fmla v30.8h, v4.8h, v14.8h\n" - "fmla v29.8h, v4.8h, v8.8h\n" + "42:" // Oddments: Load input (3, 0): Bit 2: Unset: Bit 1: Unset + "ld1 { v5.h }[0], [x19], #0x2\n" + "43:" // Oddments: Load input (3, 0): Bit 2: End + "ldr x19, [x11, #0x98]\n" + "fmla v30.8h, v0.8h, v5.8h\n" + "add x19, x19, x10\n" + "tbz %x[n_channels], #2, 45f\n" + "ld1 { v6.d }[0], [x19], #0x8\n" "tbz %x[n_channels], #1, 44f\n" - "ld1 { v10.s }[0], [x21], #0x4\n" - "tbz %x[n_channels], #0, 45f\n" - "ld1 { v10.h }[2], [x21], #0x2\n" - "b 45f\n" - "44:" // Oddments: Load input (4, 5): Bit 1: Unset - "ld1 { v10.h }[0], [x21], #0x2\n" - "45:" // Oddments: Load input (4, 5): Bit 1: End - "fmla v28.8h, v4.8h, v10.8h\n" - "ldr h0, [x15, #0x2a]\n" - "fmla v31.8h, v0.8h, v9.8h\n" - "ldr x20, [x16, #0xf0]\n" - "add x20, x20, x14\n" - "fmla v30.8h, v0.8h, v13.8h\n" + "ld1 { v6.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 47f\n" + "ld1 { v6.h }[6], [x19], #0x2\n" + "b 47f\n" + "44:" // Oddments: Load input (3, 1): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 47f\n" + "ld1 { v6.h }[4], [x19], #0x2\n" + "b 47f\n" + "45:" // Oddments: Load input (3, 1): Bit 2: Unset "tbz %x[n_channels], #1, 46f\n" - "ld1 { v11.s }[0], [x20], #0x4\n" + "ld1 { v6.s }[0], [x19], #0x4\n" "tbz %x[n_channels], #0, 47f\n" - "ld1 { v11.h }[2], [x20], #0x2\n" + "ld1 { v6.h }[2], [x19], #0x2\n" "b 47f\n" - "46:" // Oddments: Load input (5, 0): Bit 1: Unset - "ld1 { v11.h }[0], [x20], #0x2\n" - "47:" // Oddments: Load input (5, 0): Bit 1: End - "fmla v29.8h, v0.8h, v11.8h\n" - "ldr x19, [x16, #0xf8]\n" - "add x19, x19, x14\n" - "tbz %x[n_channels], #1, 48f\n" - "ld1 { v12.s }[0], [x19], #0x4\n" - "tbz %x[n_channels], #0, 49f\n" - "ld1 { v12.h }[2], [x19], #0x2\n" - "b 49f\n" - "48:" // Oddments: Load input (5, 1): Bit 1: Unset - "ld1 { v12.h }[0], [x19], #0x2\n" - "49:" // Oddments: Load input (5, 1): Bit 1: End - "fmla v28.8h, v0.8h, v12.8h\n" - "ldr h1, [x15, #0x2c]\n" - "fmla v31.8h, v1.8h, v13.8h\n" - "ldr x26, [x16, #0x100]\n" - "add x26, x26, x14\n" - "fmla v30.8h, v1.8h, v5.8h\n" + "46:" // Oddments: Load input (3, 1): Bit 2: Unset: Bit 1: Unset + "ld1 { v6.h }[0], [x19], #0x2\n" + "47:" // Oddments: Load input (3, 1): Bit 2: End + "ldr q1, [x16, #0x0]\n" + "ldr x19, [x11, #0xa0]\n" + "fmla v31.8h, v0.8h, v6.8h\n" + "fmla v28.8h, v1.8h, v11.8h\n" "fmla v29.8h, v1.8h, v12.8h\n" + "fmla v30.8h, v1.8h, v6.8h\n" + "add x19, x19, x10\n" + "add x16, x16, #0x10\n" + "tbz %x[n_channels], #2, 49f\n" + "ld1 { v10.d }[0], [x19], #0x8\n" + "tbz %x[n_channels], #1, 48f\n" + "ld1 { v10.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 51f\n" + "ld1 { v10.h }[6], [x19], #0x2\n" + "b 51f\n" + "48:" // Oddments: Load input (3, 2): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 51f\n" + "ld1 { v10.h }[4], [x19], #0x2\n" + "b 51f\n" + "49:" // Oddments: Load input (3, 2): Bit 2: Unset "tbz %x[n_channels], #1, 50f\n" - "ld1 { v9.s }[0], [x26], #0x4\n" + "ld1 { v10.s }[0], [x19], #0x4\n" "tbz %x[n_channels], #0, 51f\n" - "ld1 { v9.h }[2], [x26], #0x2\n" + "ld1 { v10.h }[2], [x19], #0x2\n" "b 51f\n" - "50:" // Oddments: Load input (5, 2): Bit 1: Unset - "ld1 { v9.h }[0], [x26], #0x2\n" - "51:" // Oddments: Load input (5, 2): Bit 1: End - "fmla v28.8h, v1.8h, v9.8h\n" - "ldr h2, [x15, #0x2e]\n" - "fmla v31.8h, v2.8h, v5.8h\n" - "ldr x25, [x16, #0x108]\n" - "add x25, x25, x14\n" - "fmla v30.8h, v2.8h, v6.8h\n" + "50:" // Oddments: Load input (3, 2): Bit 2: Unset: Bit 1: Unset + "ld1 { v10.h }[0], [x19], #0x2\n" + "51:" // Oddments: Load input (3, 2): Bit 2: End + "ldr q2, [x16, #0x0]\n" + "ldr x19, [x11, #0xa8]\n" + "fmla v31.8h, v1.8h, v10.8h\n" + "fmla v28.8h, v2.8h, v12.8h\n" "fmla v29.8h, v2.8h, v9.8h\n" + "fmla v30.8h, v2.8h, v10.8h\n" + "add x19, x19, x10\n" + "add x16, x16, #0x10\n" + "tbz %x[n_channels], #2, 53f\n" + "ld1 { v11.d }[0], [x19], #0x8\n" "tbz %x[n_channels], #1, 52f\n" - "ld1 { v11.s }[0], [x25], #0x4\n" - "tbz %x[n_channels], #0, 53f\n" - "ld1 { v11.h }[2], [x25], #0x2\n" - "b 53f\n" - "52:" // Oddments: Load input (5, 3): Bit 1: Unset - "ld1 { v11.h }[0], [x25], #0x2\n" - "53:" // Oddments: Load input (5, 3): Bit 1: End - "fmla v28.8h, v2.8h, v11.8h\n" - "ldr h3, [x15, #0x30]\n" - "fmla v31.8h, v3.8h, v6.8h\n" - "ldr x24, [x16, #0x110]\n" - "add x24, x24, x14\n" - "fmla v30.8h, v3.8h, v8.8h\n" - "fmla v29.8h, v3.8h, v11.8h\n" + "ld1 { v11.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 55f\n" + "ld1 { v11.h }[6], [x19], #0x2\n" + "b 55f\n" + "52:" // Oddments: Load input (3, 3): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 55f\n" + "ld1 { v11.h }[4], [x19], #0x2\n" + "b 55f\n" + "53:" // Oddments: Load input (3, 3): Bit 2: Unset "tbz %x[n_channels], #1, 54f\n" - "ld1 { v12.s }[0], [x24], #0x4\n" + "ld1 { v11.s }[0], [x19], #0x4\n" "tbz %x[n_channels], #0, 55f\n" - "ld1 { v12.h }[2], [x24], #0x2\n" + "ld1 { v11.h }[2], [x19], #0x2\n" "b 55f\n" - "54:" // Oddments: Load input (5, 4): Bit 1: Unset - "ld1 { v12.h }[0], [x24], #0x2\n" - "55:" // Oddments: Load input (5, 4): Bit 1: End - "fmla v28.8h, v3.8h, v12.8h\n" - "ldr h4, [x15, #0x32]\n" - "fmla v31.8h, v4.8h, v8.8h\n" - "ldr x23, [x16, #0x118]\n" - "add x23, x23, x14\n" - "fmla v30.8h, v4.8h, v10.8h\n" - "fmla v29.8h, v4.8h, v12.8h\n" + "54:" // Oddments: Load input (3, 3): Bit 2: Unset: Bit 1: Unset + "ld1 { v11.h }[0], [x19], #0x2\n" + "55:" // Oddments: Load input (3, 3): Bit 2: End + "ldr q3, [x16, #0x0]\n" + "ldr x19, [x11, #0xb0]\n" + "fmla v31.8h, v2.8h, v11.8h\n" + "fmla v28.8h, v3.8h, v9.8h\n" + "fmla v29.8h, v3.8h, v13.8h\n" + "fmla v30.8h, v3.8h, v11.8h\n" + "add x19, x19, x10\n" + "add x16, x16, #0x10\n" + "tbz %x[n_channels], #2, 57f\n" + "ld1 { v12.d }[0], [x19], #0x8\n" "tbz %x[n_channels], #1, 56f\n" - "ld1 { v9.s }[0], [x23], #0x4\n" - "tbz %x[n_channels], #0, 57f\n" - "ld1 { v9.h }[2], [x23], #0x2\n" - "b 57f\n" - "56:" // Oddments: Load input (5, 5): Bit 1: Unset - "ld1 { v9.h }[0], [x23], #0x2\n" - "57:" // Oddments: Load input (5, 5): Bit 1: End - "fmla v28.8h, v4.8h, v9.8h\n" - "fmax v31.8h, v31.8h, v18.8h\n" - "fmax v30.8h, v30.8h, v18.8h\n" - "fmax v29.8h, v29.8h, v18.8h\n" - "fmin v31.8h, v31.8h, v17.8h\n" - "fmin v30.8h, v30.8h, v17.8h\n" - "fmin v29.8h, v29.8h, v17.8h\n" - "fmax v28.8h, v28.8h, v18.8h\n" - "fmin v28.8h, v28.8h, v17.8h\n" + "ld1 { v12.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 59f\n" + "ld1 { v12.h }[6], [x19], #0x2\n" + "b 59f\n" + "56:" // Oddments: Load input (3, 4): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 59f\n" + "ld1 { v12.h }[4], [x19], #0x2\n" + "b 59f\n" + "57:" // Oddments: Load input (3, 4): Bit 2: Unset "tbz %x[n_channels], #1, 58f\n" - "st1 { v31.s }[0], [x13], #0x4\n" - "st1 { v30.s }[0], [x12], #0x4\n" - "st1 { v29.s }[0], [x10], #0x4\n" - "st1 { v28.s }[0], [x9], #0x4\n" + "ld1 { v12.s }[0], [x19], #0x4\n" "tbz %x[n_channels], #0, 59f\n" - "st1 { v31.h }[2], [x13], #0x2\n" - "st1 { v30.h }[2], [x12], #0x2\n" - "st1 { v29.h }[2], [x10], #0x2\n" - "st1 { v28.h }[2], [x9], #0x2\n" + "ld1 { v12.h }[2], [x19], #0x2\n" "b 59f\n" - "58:" // Oddments: Store: Bit 1: Unset - "st1 { v31.h }[0], [x13], #0x2\n" - "st1 { v30.h }[0], [x12], #0x2\n" - "st1 { v29.h }[0], [x10], #0x2\n" - "st1 { v28.h }[0], [x9], #0x2\n" - "59:" // Oddments: Store: Bit 1: End + "58:" // Oddments: Load input (3, 4): Bit 2: Unset: Bit 1: Unset + "ld1 { v12.h }[0], [x19], #0x2\n" + "59:" // Oddments: Load input (3, 4): Bit 2: End + "ldr q4, [x16, #0x0]\n" + "ldr x19, [x11, #0xb8]\n" + "fmla v31.8h, v3.8h, v12.8h\n" + "fmla v28.8h, v4.8h, v13.8h\n" + "fmla v29.8h, v4.8h, v8.8h\n" + "fmla v30.8h, v4.8h, v12.8h\n" + "add x19, x19, x10\n" + "add x16, x16, #0x10\n" + "tbz %x[n_channels], #2, 61f\n" + "ld1 { v14.d }[0], [x19], #0x8\n" + "tbz %x[n_channels], #1, 60f\n" + "ld1 { v14.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 63f\n" + "ld1 { v14.h }[6], [x19], #0x2\n" + "b 63f\n" + "60:" // Oddments: Load input (3, 5): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 63f\n" + "ld1 { v14.h }[4], [x19], #0x2\n" + "b 63f\n" + "61:" // Oddments: Load input (3, 5): Bit 2: Unset + "tbz %x[n_channels], #1, 62f\n" + "ld1 { v14.s }[0], [x19], #0x4\n" + "tbz %x[n_channels], #0, 63f\n" + "ld1 { v14.h }[2], [x19], #0x2\n" + "b 63f\n" + "62:" // Oddments: Load input (3, 5): Bit 2: Unset: Bit 1: Unset + "ld1 { v14.h }[0], [x19], #0x2\n" + "63:" // Oddments: Load input (3, 5): Bit 2: End + "ldr q0, [x16, #0x0]\n" + "ldr x19, [x11, #0xc0]\n" + "fmla v31.8h, v4.8h, v14.8h\n" + "fmla v28.8h, v0.8h, v5.8h\n" + "fmla v29.8h, v0.8h, v6.8h\n" + "add x19, x19, x10\n" + "add x16, x16, #0x10\n" + "tbz %x[n_channels], #2, 65f\n" + "ld1 { v9.d }[0], [x19], #0x8\n" + "tbz %x[n_channels], #1, 64f\n" + "ld1 { v9.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 67f\n" + "ld1 { v9.h }[6], [x19], #0x2\n" + "b 67f\n" + "64:" // Oddments: Load input (4, 0): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 67f\n" + "ld1 { v9.h }[4], [x19], #0x2\n" + "b 67f\n" + "65:" // Oddments: Load input (4, 0): Bit 2: Unset + "tbz %x[n_channels], #1, 66f\n" + "ld1 { v9.s }[0], [x19], #0x4\n" + "tbz %x[n_channels], #0, 67f\n" + "ld1 { v9.h }[2], [x19], #0x2\n" + "b 67f\n" + "66:" // Oddments: Load input (4, 0): Bit 2: Unset: Bit 1: Unset + "ld1 { v9.h }[0], [x19], #0x2\n" + "67:" // Oddments: Load input (4, 0): Bit 2: End + "ldr x19, [x11, #0xc8]\n" + "fmla v30.8h, v0.8h, v9.8h\n" + "add x19, x19, x10\n" + "tbz %x[n_channels], #2, 69f\n" + "ld1 { v13.d }[0], [x19], #0x8\n" + "tbz %x[n_channels], #1, 68f\n" + "ld1 { v13.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 71f\n" + "ld1 { v13.h }[6], [x19], #0x2\n" + "b 71f\n" + "68:" // Oddments: Load input (4, 1): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 71f\n" + "ld1 { v13.h }[4], [x19], #0x2\n" + "b 71f\n" + "69:" // Oddments: Load input (4, 1): Bit 2: Unset + "tbz %x[n_channels], #1, 70f\n" + "ld1 { v13.s }[0], [x19], #0x4\n" + "tbz %x[n_channels], #0, 71f\n" + "ld1 { v13.h }[2], [x19], #0x2\n" + "b 71f\n" + "70:" // Oddments: Load input (4, 1): Bit 2: Unset: Bit 1: Unset + "ld1 { v13.h }[0], [x19], #0x2\n" + "71:" // Oddments: Load input (4, 1): Bit 2: End + "ldr q1, [x16, #0x0]\n" + "ldr x19, [x11, #0xd0]\n" + "fmla v31.8h, v0.8h, v13.8h\n" + "fmla v28.8h, v1.8h, v6.8h\n" + "fmla v29.8h, v1.8h, v10.8h\n" + "fmla v30.8h, v1.8h, v13.8h\n" + "add x19, x19, x10\n" + "add x16, x16, #0x10\n" + "tbz %x[n_channels], #2, 73f\n" + "ld1 { v5.d }[0], [x19], #0x8\n" + "tbz %x[n_channels], #1, 72f\n" + "ld1 { v5.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 75f\n" + "ld1 { v5.h }[6], [x19], #0x2\n" + "b 75f\n" + "72:" // Oddments: Load input (4, 2): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 75f\n" + "ld1 { v5.h }[4], [x19], #0x2\n" + "b 75f\n" + "73:" // Oddments: Load input (4, 2): Bit 2: Unset + "tbz %x[n_channels], #1, 74f\n" + "ld1 { v5.s }[0], [x19], #0x4\n" + "tbz %x[n_channels], #0, 75f\n" + "ld1 { v5.h }[2], [x19], #0x2\n" + "b 75f\n" + "74:" // Oddments: Load input (4, 2): Bit 2: Unset: Bit 1: Unset + "ld1 { v5.h }[0], [x19], #0x2\n" + "75:" // Oddments: Load input (4, 2): Bit 2: End + "ldr q2, [x16, #0x0]\n" + "ldr x19, [x11, #0xd8]\n" + "fmla v31.8h, v1.8h, v5.8h\n" + "fmla v28.8h, v2.8h, v10.8h\n" + "fmla v29.8h, v2.8h, v11.8h\n" + "fmla v30.8h, v2.8h, v5.8h\n" + "add x19, x19, x10\n" + "add x16, x16, #0x10\n" + "tbz %x[n_channels], #2, 77f\n" + "ld1 { v6.d }[0], [x19], #0x8\n" + "tbz %x[n_channels], #1, 76f\n" + "ld1 { v6.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 79f\n" + "ld1 { v6.h }[6], [x19], #0x2\n" + "b 79f\n" + "76:" // Oddments: Load input (4, 3): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 79f\n" + "ld1 { v6.h }[4], [x19], #0x2\n" + "b 79f\n" + "77:" // Oddments: Load input (4, 3): Bit 2: Unset + "tbz %x[n_channels], #1, 78f\n" + "ld1 { v6.s }[0], [x19], #0x4\n" + "tbz %x[n_channels], #0, 79f\n" + "ld1 { v6.h }[2], [x19], #0x2\n" + "b 79f\n" + "78:" // Oddments: Load input (4, 3): Bit 2: Unset: Bit 1: Unset + "ld1 { v6.h }[0], [x19], #0x2\n" + "79:" // Oddments: Load input (4, 3): Bit 2: End + "ldr q3, [x16, #0x0]\n" + "ldr x19, [x11, #0xe0]\n" + "fmla v31.8h, v2.8h, v6.8h\n" + "fmla v28.8h, v3.8h, v11.8h\n" + "fmla v29.8h, v3.8h, v12.8h\n" + "fmla v30.8h, v3.8h, v6.8h\n" + "add x19, x19, x10\n" + "add x16, x16, #0x10\n" + "tbz %x[n_channels], #2, 81f\n" + "ld1 { v8.d }[0], [x19], #0x8\n" + "tbz %x[n_channels], #1, 80f\n" + "ld1 { v8.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 83f\n" + "ld1 { v8.h }[6], [x19], #0x2\n" + "b 83f\n" + "80:" // Oddments: Load input (4, 4): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 83f\n" + "ld1 { v8.h }[4], [x19], #0x2\n" + "b 83f\n" + "81:" // Oddments: Load input (4, 4): Bit 2: Unset + "tbz %x[n_channels], #1, 82f\n" + "ld1 { v8.s }[0], [x19], #0x4\n" + "tbz %x[n_channels], #0, 83f\n" + "ld1 { v8.h }[2], [x19], #0x2\n" + "b 83f\n" + "82:" // Oddments: Load input (4, 4): Bit 2: Unset: Bit 1: Unset + "ld1 { v8.h }[0], [x19], #0x2\n" + "83:" // Oddments: Load input (4, 4): Bit 2: End + "ldr q4, [x16, #0x0]\n" + "ldr x19, [x11, #0xe8]\n" + "fmla v31.8h, v3.8h, v8.8h\n" + "fmla v28.8h, v4.8h, v12.8h\n" + "fmla v29.8h, v4.8h, v14.8h\n" + "fmla v30.8h, v4.8h, v8.8h\n" + "add x19, x19, x10\n" + "add x16, x16, #0x10\n" + "tbz %x[n_channels], #2, 85f\n" + "ld1 { v10.d }[0], [x19], #0x8\n" + "tbz %x[n_channels], #1, 84f\n" + "ld1 { v10.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 87f\n" + "ld1 { v10.h }[6], [x19], #0x2\n" + "b 87f\n" + "84:" // Oddments: Load input (4, 5): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 87f\n" + "ld1 { v10.h }[4], [x19], #0x2\n" + "b 87f\n" + "85:" // Oddments: Load input (4, 5): Bit 2: Unset + "tbz %x[n_channels], #1, 86f\n" + "ld1 { v10.s }[0], [x19], #0x4\n" + "tbz %x[n_channels], #0, 87f\n" + "ld1 { v10.h }[2], [x19], #0x2\n" + "b 87f\n" + "86:" // Oddments: Load input (4, 5): Bit 2: Unset: Bit 1: Unset + "ld1 { v10.h }[0], [x19], #0x2\n" + "87:" // Oddments: Load input (4, 5): Bit 2: End + "ldr q0, [x16, #0x0]\n" + "ldr x19, [x11, #0xf0]\n" + "fmla v31.8h, v4.8h, v10.8h\n" + "fmla v28.8h, v0.8h, v9.8h\n" + "fmla v29.8h, v0.8h, v13.8h\n" + "add x19, x19, x10\n" + "add x16, x16, #0x10\n" + "tbz %x[n_channels], #2, 89f\n" + "ld1 { v11.d }[0], [x19], #0x8\n" + "tbz %x[n_channels], #1, 88f\n" + "ld1 { v11.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 91f\n" + "ld1 { v11.h }[6], [x19], #0x2\n" + "b 91f\n" + "88:" // Oddments: Load input (5, 0): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 91f\n" + "ld1 { v11.h }[4], [x19], #0x2\n" + "b 91f\n" + "89:" // Oddments: Load input (5, 0): Bit 2: Unset + "tbz %x[n_channels], #1, 90f\n" + "ld1 { v11.s }[0], [x19], #0x4\n" + "tbz %x[n_channels], #0, 91f\n" + "ld1 { v11.h }[2], [x19], #0x2\n" + "b 91f\n" + "90:" // Oddments: Load input (5, 0): Bit 2: Unset: Bit 1: Unset + "ld1 { v11.h }[0], [x19], #0x2\n" + "91:" // Oddments: Load input (5, 0): Bit 2: End + "ldr x19, [x11, #0xf8]\n" + "fmla v30.8h, v0.8h, v11.8h\n" + "add x19, x19, x10\n" + "tbz %x[n_channels], #2, 93f\n" + "ld1 { v12.d }[0], [x19], #0x8\n" + "tbz %x[n_channels], #1, 92f\n" + "ld1 { v12.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 95f\n" + "ld1 { v12.h }[6], [x19], #0x2\n" + "b 95f\n" + "92:" // Oddments: Load input (5, 1): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 95f\n" + "ld1 { v12.h }[4], [x19], #0x2\n" + "b 95f\n" + "93:" // Oddments: Load input (5, 1): Bit 2: Unset + "tbz %x[n_channels], #1, 94f\n" + "ld1 { v12.s }[0], [x19], #0x4\n" + "tbz %x[n_channels], #0, 95f\n" + "ld1 { v12.h }[2], [x19], #0x2\n" + "b 95f\n" + "94:" // Oddments: Load input (5, 1): Bit 2: Unset: Bit 1: Unset + "ld1 { v12.h }[0], [x19], #0x2\n" + "95:" // Oddments: Load input (5, 1): Bit 2: End + "ldr q1, [x16, #0x0]\n" + "ldr x19, [x11, #0x100]\n" + "fmla v31.8h, v0.8h, v12.8h\n" + "fmla v28.8h, v1.8h, v13.8h\n" + "fmla v29.8h, v1.8h, v5.8h\n" + "fmla v30.8h, v1.8h, v12.8h\n" + "add x19, x19, x10\n" + "add x16, x16, #0x10\n" + "tbz %x[n_channels], #2, 97f\n" + "ld1 { v9.d }[0], [x19], #0x8\n" + "tbz %x[n_channels], #1, 96f\n" + "ld1 { v9.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 99f\n" + "ld1 { v9.h }[6], [x19], #0x2\n" + "b 99f\n" + "96:" // Oddments: Load input (5, 2): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 99f\n" + "ld1 { v9.h }[4], [x19], #0x2\n" + "b 99f\n" + "97:" // Oddments: Load input (5, 2): Bit 2: Unset + "tbz %x[n_channels], #1, 98f\n" + "ld1 { v9.s }[0], [x19], #0x4\n" + "tbz %x[n_channels], #0, 99f\n" + "ld1 { v9.h }[2], [x19], #0x2\n" + "b 99f\n" + "98:" // Oddments: Load input (5, 2): Bit 2: Unset: Bit 1: Unset + "ld1 { v9.h }[0], [x19], #0x2\n" + "99:" // Oddments: Load input (5, 2): Bit 2: End + "ldr q2, [x16, #0x0]\n" + "ldr x19, [x11, #0x108]\n" + "fmla v31.8h, v1.8h, v9.8h\n" + "fmla v28.8h, v2.8h, v5.8h\n" + "fmla v29.8h, v2.8h, v6.8h\n" + "fmla v30.8h, v2.8h, v9.8h\n" + "add x19, x19, x10\n" + "add x16, x16, #0x10\n" + "tbz %x[n_channels], #2, 101f\n" + "ld1 { v11.d }[0], [x19], #0x8\n" + "tbz %x[n_channels], #1, 100f\n" + "ld1 { v11.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 103f\n" + "ld1 { v11.h }[6], [x19], #0x2\n" + "b 103f\n" + "100:" // Oddments: Load input (5, 3): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 103f\n" + "ld1 { v11.h }[4], [x19], #0x2\n" + "b 103f\n" + "101:" // Oddments: Load input (5, 3): Bit 2: Unset + "tbz %x[n_channels], #1, 102f\n" + "ld1 { v11.s }[0], [x19], #0x4\n" + "tbz %x[n_channels], #0, 103f\n" + "ld1 { v11.h }[2], [x19], #0x2\n" + "b 103f\n" + "102:" // Oddments: Load input (5, 3): Bit 2: Unset: Bit 1: Unset + "ld1 { v11.h }[0], [x19], #0x2\n" + "103:" // Oddments: Load input (5, 3): Bit 2: End + "ldr q3, [x16, #0x0]\n" + "ldr x19, [x11, #0x110]\n" + "fmla v31.8h, v2.8h, v11.8h\n" + "fmla v28.8h, v3.8h, v6.8h\n" + "fmla v29.8h, v3.8h, v8.8h\n" + "fmla v30.8h, v3.8h, v11.8h\n" + "add x19, x19, x10\n" + "add x16, x16, #0x10\n" + "tbz %x[n_channels], #2, 105f\n" + "ld1 { v12.d }[0], [x19], #0x8\n" + "tbz %x[n_channels], #1, 104f\n" + "ld1 { v12.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 107f\n" + "ld1 { v12.h }[6], [x19], #0x2\n" + "b 107f\n" + "104:" // Oddments: Load input (5, 4): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 107f\n" + "ld1 { v12.h }[4], [x19], #0x2\n" + "b 107f\n" + "105:" // Oddments: Load input (5, 4): Bit 2: Unset + "tbz %x[n_channels], #1, 106f\n" + "ld1 { v12.s }[0], [x19], #0x4\n" + "tbz %x[n_channels], #0, 107f\n" + "ld1 { v12.h }[2], [x19], #0x2\n" + "b 107f\n" + "106:" // Oddments: Load input (5, 4): Bit 2: Unset: Bit 1: Unset + "ld1 { v12.h }[0], [x19], #0x2\n" + "107:" // Oddments: Load input (5, 4): Bit 2: End + "ldr q4, [x16, #0x0]\n" + "ldr x19, [x11, #0x118]\n" + "fmla v31.8h, v3.8h, v12.8h\n" + "fmla v28.8h, v4.8h, v8.8h\n" + "fmla v29.8h, v4.8h, v10.8h\n" + "fmla v30.8h, v4.8h, v12.8h\n" + "add x19, x19, x10\n" + "tbz %x[n_channels], #2, 109f\n" + "ld1 { v9.d }[0], [x19], #0x8\n" + "tbz %x[n_channels], #1, 108f\n" + "ld1 { v9.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 111f\n" + "ld1 { v9.h }[6], [x19], #0x2\n" + "b 111f\n" + "108:" // Oddments: Load input (5, 5): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 111f\n" + "ld1 { v9.h }[4], [x19], #0x2\n" + "b 111f\n" + "109:" // Oddments: Load input (5, 5): Bit 2: Unset + "tbz %x[n_channels], #1, 110f\n" + "ld1 { v9.s }[0], [x19], #0x4\n" + "tbz %x[n_channels], #0, 111f\n" + "ld1 { v9.h }[2], [x19], #0x2\n" + "b 111f\n" + "110:" // Oddments: Load input (5, 5): Bit 2: Unset: Bit 1: Unset + "ld1 { v9.h }[0], [x19], #0x2\n" + "111:" // Oddments: Load input (5, 5): Bit 2: End + "fmla v31.8h, v4.8h, v9.8h\n" + "fmax v28.8h, v28.8h, v18.8h\n" + "fmax v29.8h, v29.8h, v18.8h\n" + "fmax v30.8h, v30.8h, v18.8h\n" + "fmax v31.8h, v31.8h, v18.8h\n" + "fmin v28.8h, v28.8h, v17.8h\n" + "fmin v29.8h, v29.8h, v17.8h\n" + "fmin v30.8h, v30.8h, v17.8h\n" + "fmin v31.8h, v31.8h, v17.8h\n" + "tbz %x[n_channels], #2, 113f\n" + "st1 { v28.d }[0], [x15], #0x8\n" + "st1 { v29.d }[0], [x14], #0x8\n" + "st1 { v30.d }[0], [x13], #0x8\n" + "st1 { v31.d }[0], [x12], #0x8\n" + "tbz %x[n_channels], #1, 112f\n" + "st1 { v28.s }[2], [x15], #0x4\n" + "st1 { v29.s }[2], [x14], #0x4\n" + "st1 { v30.s }[2], [x13], #0x4\n" + "st1 { v31.s }[2], [x12], #0x4\n" + "tbz %x[n_channels], #0, 115f\n" + "st1 { v28.h }[6], [x15], #0x2\n" + "st1 { v29.h }[6], [x14], #0x2\n" + "st1 { v30.h }[6], [x13], #0x2\n" + "st1 { v31.h }[6], [x12], #0x2\n" + "b 115f\n" + "112:" // Oddments: Store: Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 115f\n" + "st1 { v28.h }[4], [x15], #0x2\n" + "st1 { v29.h }[4], [x14], #0x2\n" + "st1 { v30.h }[4], [x13], #0x2\n" + "st1 { v31.h }[4], [x12], #0x2\n" + "b 115f\n" + "113:" // Oddments: Store: Bit 2: Unset + "tbz %x[n_channels], #1, 114f\n" + "st1 { v28.s }[0], [x15], #0x4\n" + "st1 { v29.s }[0], [x14], #0x4\n" + "st1 { v30.s }[0], [x13], #0x4\n" + "st1 { v31.s }[0], [x12], #0x4\n" + "tbz %x[n_channels], #0, 115f\n" + "st1 { v28.h }[2], [x15], #0x2\n" + "st1 { v29.h }[2], [x14], #0x2\n" + "st1 { v30.h }[2], [x13], #0x2\n" + "st1 { v31.h }[2], [x12], #0x2\n" + "b 115f\n" + "114:" // Oddments: Store: Bit 2: Unset: Bit 1: Unset + "st1 { v28.h }[0], [x15], #0x2\n" + "st1 { v29.h }[0], [x14], #0x2\n" + "st1 { v30.h }[0], [x13], #0x2\n" + "st1 { v31.h }[0], [x12], #0x2\n" + "115:" // Oddments: Store: Bit 2: End - "60:" // End + "116:" // End : : [n_channels] "r" ((unsigned long) n_channels), [offsetof_Args_inptrs] "I" (offsetof(Args, inptrs)), [offsetof_args_max] "I" (offsetof(Args, max)), [offsetof_args_min] "I" (offsetof(Args, min)), [offsetof_args_outptrs] "I" (offsetof(Args, outptrs)), [offsetof_args_params] "I" (offsetof(Args, params)), [params_struct] "r" (¶ms_struct) -- cgit v1.2.1