aboutsummaryrefslogtreecommitdiff
path: root/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_bf16_fp32.hpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_bf16_fp32.hpp')
-rw-r--r--src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_bf16_fp32.hpp156
1 files changed, 78 insertions, 78 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_bf16_fp32.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_bf16_fp32.hpp
index 34d25f27b8..42574295f1 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_bf16_fp32.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block1_bf16_fp32.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -31,7 +31,7 @@ void interleave_block<8, 1, VLType::None, false>(
)
{
__asm__ __volatile__(
- "movi v29.8h, #0x0\n"
+ "movi v30.8h, #0x0\n"
"ldr x27, [%x[in], #0x0]\n"
"cmp %x[height], #0x8\n"
"ldr x26, [%x[in], #0x8]\n"
@@ -80,53 +80,53 @@ void interleave_block<8, 1, VLType::None, false>(
"prfm pldl1keep, [x20, #0x40]\n"
"blt 3f\n"
"2:" // Main loop head
- "ldr d28, [x27], #0x8\n"
- "zip1 v28.8h, v29.8h, v28.8h\n"
+ "ldr d29, [x27], #0x8\n"
+ "zip1 v29.8h, v30.8h, v29.8h\n"
+ "ldr d28, [x26], #0x8\n"
+ "subs %x[width], %x[width], #0x4\n"
+ "zip1 v28.8h, v30.8h, v28.8h\n"
+ "ldr d24, [x25], #0x8\n"
+ "cmp %x[width], #0x4\n"
+ "zip1 v24.8h, v30.8h, v24.8h\n"
+ "ldr d27, [x24], #0x8\n"
+ "ldr d26, [x23], #0x8\n"
+ "zip1 v25.4s, v29.4s, v24.4s\n"
+ "zip2 v24.4s, v29.4s, v24.4s\n"
+ "ldr d23, [x22], #0x8\n"
+ "ldr d22, [x21], #0x8\n"
+ "zip1 v27.8h, v30.8h, v27.8h\n"
+ "ldr d21, [x20], #0x8\n"
+ "zip1 v26.8h, v30.8h, v26.8h\n"
"prfm pldl1keep, [x27, #0x70]\n"
- "ldr d27, [x26], #0x8\n"
- "zip1 v27.8h, v29.8h, v27.8h\n"
+ "zip1 v20.4s, v28.4s, v27.4s\n"
"prfm pldl1keep, [x26, #0x70]\n"
- "ldr d26, [x25], #0x8\n"
- "zip1 v26.8h, v29.8h, v26.8h\n"
+ "zip1 v23.8h, v30.8h, v23.8h\n"
"prfm pldl1keep, [x25, #0x70]\n"
- "ldr d25, [x24], #0x8\n"
- "zip1 v20.4s, v28.4s, v26.4s\n"
+ "zip1 v22.8h, v30.8h, v22.8h\n"
"prfm pldl1keep, [x24, #0x70]\n"
- "zip1 v25.8h, v29.8h, v25.8h\n"
- "ldr d24, [x23], #0x8\n"
- "zip1 v19.4s, v27.4s, v25.4s\n"
+ "zip1 v21.8h, v30.8h, v21.8h\n"
"prfm pldl1keep, [x23, #0x70]\n"
- "zip1 v24.8h, v29.8h, v24.8h\n"
- "ldr d23, [x22], #0x8\n"
- "zip1 v16.4s, v20.4s, v19.4s\n"
+ "zip1 v17.4s, v25.4s, v20.4s\n"
"prfm pldl1keep, [x22, #0x70]\n"
- "zip1 v23.8h, v29.8h, v23.8h\n"
- "ldr d22, [x21], #0x8\n"
- "zip2 v19.4s, v20.4s, v19.4s\n"
+ "zip1 v19.4s, v26.4s, v22.4s\n"
"prfm pldl1keep, [x21, #0x70]\n"
- "zip1 v22.8h, v29.8h, v22.8h\n"
- "ldr d21, [x20], #0x8\n"
- "zip1 v18.4s, v24.4s, v22.4s\n"
+ "zip1 v18.4s, v23.4s, v21.4s\n"
"prfm pldl1keep, [x20, #0x70]\n"
- "zip1 v21.8h, v29.8h, v21.8h\n"
- "str q16, [%x[out_ptr], #0x0]\n"
- "zip1 v17.4s, v23.4s, v21.4s\n"
- "subs %x[width], %x[width], #0x4\n"
- "zip2 v20.4s, v28.4s, v26.4s\n"
- "cmp %x[width], #0x4\n"
- "zip1 v16.4s, v18.4s, v17.4s\n"
+ "zip1 v16.4s, v19.4s, v18.4s\n"
+ "str q17, [%x[out_ptr], #0x0]\n"
+ "zip2 v17.4s, v25.4s, v20.4s\n"
"str q16, [%x[out_ptr], #0x10]\n"
- "zip2 v16.4s, v18.4s, v17.4s\n"
- "str q19, [%x[out_ptr], #0x20]\n"
- "zip2 v19.4s, v27.4s, v25.4s\n"
+ "zip2 v16.4s, v19.4s, v18.4s\n"
+ "str q17, [%x[out_ptr], #0x20]\n"
+ "zip2 v19.4s, v28.4s, v27.4s\n"
"str q16, [%x[out_ptr], #0x30]\n"
- "zip1 v16.4s, v20.4s, v19.4s\n"
+ "zip1 v16.4s, v24.4s, v19.4s\n"
"str q16, [%x[out_ptr], #0x40]\n"
- "zip2 v18.4s, v24.4s, v22.4s\n"
+ "zip2 v18.4s, v26.4s, v22.4s\n"
"zip2 v17.4s, v23.4s, v21.4s\n"
"zip1 v16.4s, v18.4s, v17.4s\n"
"str q16, [%x[out_ptr], #0x50]\n"
- "zip2 v16.4s, v20.4s, v19.4s\n"
+ "zip2 v16.4s, v24.4s, v19.4s\n"
"str q16, [%x[out_ptr], #0x60]\n"
"zip2 v16.4s, v18.4s, v17.4s\n"
"str q16, [%x[out_ptr], #0x70]\n"
@@ -135,77 +135,77 @@ void interleave_block<8, 1, VLType::None, false>(
"3:" // Main loop skip
"cbz %x[width], 6f\n"
"tbz %x[width], #1, 4f\n"
- "ldr s28, [x27], #0x4\n"
- "ldr s27, [x26], #0x4\n"
- "ldr s26, [x25], #0x4\n"
- "ldr s25, [x24], #0x4\n"
- "ldr s24, [x23], #0x4\n"
+ "ldr s29, [x27], #0x4\n"
+ "ldr s28, [x26], #0x4\n"
+ "mov x19, #0x2\n"
+ "ldr s24, [x25], #0x4\n"
+ "ldr s27, [x24], #0x4\n"
+ "ldr s26, [x23], #0x4\n"
"ldr s23, [x22], #0x4\n"
"ldr s22, [x21], #0x4\n"
"ldr s21, [x20], #0x4\n"
- "mov x19, #0x2\n"
"tbz %x[width], #0, 5f\n"
- "ld1 { v28.h }[2], [x27]\n"
- "ld1 { v27.h }[2], [x26]\n"
- "ld1 { v26.h }[2], [x25]\n"
- "ld1 { v25.h }[2], [x24]\n"
- "ld1 { v24.h }[2], [x23]\n"
+ "ld1 { v29.h }[2], [x27]\n"
+ "mov x19, #0x3\n"
+ "ld1 { v28.h }[2], [x26]\n"
+ "ld1 { v24.h }[2], [x25]\n"
+ "ld1 { v27.h }[2], [x24]\n"
+ "ld1 { v26.h }[2], [x23]\n"
"ld1 { v23.h }[2], [x22]\n"
"ld1 { v22.h }[2], [x21]\n"
"ld1 { v21.h }[2], [x20]\n"
- "mov x19, #0x3\n"
"b 5f\n"
"4:" // odd_loads_1_0
- "ldr h28, [x27, #0x0]\n"
- "ldr h27, [x26, #0x0]\n"
- "ldr h26, [x25, #0x0]\n"
- "ldr h25, [x24, #0x0]\n"
- "ldr h24, [x23, #0x0]\n"
+ "ldr h29, [x27, #0x0]\n"
+ "mov x19, #0x1\n"
+ "ldr h28, [x26, #0x0]\n"
+ "ldr h24, [x25, #0x0]\n"
+ "ldr h27, [x24, #0x0]\n"
+ "ldr h26, [x23, #0x0]\n"
"ldr h23, [x22, #0x0]\n"
"ldr h22, [x21, #0x0]\n"
"ldr h21, [x20, #0x0]\n"
- "mov x19, #0x1\n"
"5:" // Odd load end
- "zip1 v28.8h, v29.8h, v28.8h\n"
+ "zip1 v29.8h, v30.8h, v29.8h\n"
"subs x19, x19, #0x1\n"
- "zip1 v27.8h, v29.8h, v27.8h\n"
- "zip1 v26.8h, v29.8h, v26.8h\n"
- "zip1 v25.8h, v29.8h, v25.8h\n"
- "zip1 v24.8h, v29.8h, v24.8h\n"
- "zip1 v23.8h, v29.8h, v23.8h\n"
- "zip1 v22.8h, v29.8h, v22.8h\n"
- "zip1 v21.8h, v29.8h, v21.8h\n"
- "zip1 v20.4s, v28.4s, v26.4s\n"
- "zip1 v19.4s, v27.4s, v25.4s\n"
- "zip1 v16.4s, v20.4s, v19.4s\n"
- "str q16, [%x[out_ptr], #0x0]\n"
- "zip1 v18.4s, v24.4s, v22.4s\n"
- "zip1 v17.4s, v23.4s, v21.4s\n"
- "zip1 v16.4s, v18.4s, v17.4s\n"
+ "zip1 v28.8h, v30.8h, v28.8h\n"
+ "zip1 v24.8h, v30.8h, v24.8h\n"
+ "zip1 v27.8h, v30.8h, v27.8h\n"
+ "zip1 v26.8h, v30.8h, v26.8h\n"
+ "zip1 v23.8h, v30.8h, v23.8h\n"
+ "zip1 v22.8h, v30.8h, v22.8h\n"
+ "zip1 v21.8h, v30.8h, v21.8h\n"
+ "zip1 v25.4s, v29.4s, v24.4s\n"
+ "zip1 v20.4s, v28.4s, v27.4s\n"
+ "zip1 v17.4s, v25.4s, v20.4s\n"
+ "str q17, [%x[out_ptr], #0x0]\n"
+ "zip1 v19.4s, v26.4s, v22.4s\n"
+ "zip1 v18.4s, v23.4s, v21.4s\n"
+ "zip1 v16.4s, v19.4s, v18.4s\n"
"str q16, [%x[out_ptr], #0x10]\n"
"add %x[out_ptr], %x[out_ptr], #0x20\n"
"beq 6f\n"
- "zip2 v19.4s, v20.4s, v19.4s\n"
- "zip2 v16.4s, v18.4s, v17.4s\n"
- "str q19, [%x[out_ptr], #0x0]\n"
- "str q16, [%x[out_ptr], #0x10]\n"
+ "zip2 v17.4s, v25.4s, v20.4s\n"
+ "str q17, [%x[out_ptr], #0x0]\n"
+ "zip2 v16.4s, v19.4s, v18.4s\n"
"subs x19, x19, #0x1\n"
+ "str q16, [%x[out_ptr], #0x10]\n"
"add %x[out_ptr], %x[out_ptr], #0x20\n"
"beq 6f\n"
- "zip2 v20.4s, v28.4s, v26.4s\n"
- "zip2 v19.4s, v27.4s, v25.4s\n"
- "zip1 v16.4s, v20.4s, v19.4s\n"
+ "zip2 v24.4s, v29.4s, v24.4s\n"
+ "zip2 v19.4s, v28.4s, v27.4s\n"
+ "zip1 v16.4s, v24.4s, v19.4s\n"
"str q16, [%x[out_ptr], #0x0]\n"
- "zip2 v18.4s, v24.4s, v22.4s\n"
+ "zip2 v18.4s, v26.4s, v22.4s\n"
"zip2 v17.4s, v23.4s, v21.4s\n"
"zip1 v16.4s, v18.4s, v17.4s\n"
"str q16, [%x[out_ptr], #0x10]\n"
"add %x[out_ptr], %x[out_ptr], #0x20\n"
"6:" // Odds skip
- : [out_ptr] "+r" (out_ptr), [width] "+r" (width)
+ : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
: [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
- : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
+ : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
);
}
generated by cgit v1.2.1 (git 2.23.0) at 2024-05-08 19:34:01 +0000