aboutsummaryrefslogtreecommitdiff
path: root/src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_32_2x2.hpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_32_2x2.hpp')
-rw-r--r--src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_32_2x2.hpp7
1 files changed, 3 insertions, 4 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_32_2x2.hpp b/src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_32_2x2.hpp
index 706d7cd359..237536697c 100644
--- a/src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_32_2x2.hpp
+++ b/src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_32_2x2.hpp
@@ -24,7 +24,7 @@
#pragma once
-#ifdef __aarch64__
+#if defined(__aarch64__)
namespace {
@@ -280,7 +280,6 @@ void a64_transpose_interleave_32_2x2(uint16_t *out, const uint16_t *in, size_t w
"bge 1b\n"
"cbz %x[height], 24f\n"
"12:" // Main loop skip
-
"13:" // Tail row loop: Head
"mov x25, %x[in]\n"
"mov x20, %x[width]\n"
@@ -427,7 +426,6 @@ void a64_transpose_interleave_32_2x2(uint16_t *out, const uint16_t *in, size_t w
"add %x[out], %x[out], #0x80\n"
"bge 13b\n"
"24:" // Done
-
: [height] "+&r" (height), [in] "+&r" (in), [out] "+&r" (out)
: [in_stride] "r" (in_stride), [out_stride] "r" (out_stride), [pad_row] "r" (pad_row), [width] "r" (width)
: "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x20", "x21", "x22", "x23", "x24", "x25"
@@ -449,4 +447,5 @@ void Transform<32, 2, true, VLType::None>(
);
}
-#endif
+
+#endif // defined(__aarch64__)