aboutsummaryrefslogtreecommitdiff
path: root/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_u8qa_dot_4x16/a55.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_u8qa_dot_4x16/a55.cpp')
-rw-r--r--src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_u8qa_dot_4x16/a55.cpp18
1 files changed, 9 insertions, 9 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_u8qa_dot_4x16/a55.cpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_u8qa_dot_4x16/a55.cpp
index 954e2891fb..c410374357 100644
--- a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_u8qa_dot_4x16/a55.cpp
+++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_u8qa_dot_4x16/a55.cpp
@@ -406,10 +406,10 @@ void a64_hybrid_u8qa_dot_4x16_a55 (
"b 122f\n"
"31:" // Height 2
"movi v11.4s, #0x0\n"
- "movi v12.4s, #0x0\n"
- "movi v15.16b, #0x1\n"
"ldr x11, [%x[args_ptr], %[offsetof_N]]\n"
+ "movi v12.4s, #0x0\n"
"ldr x10, [%x[args_ptr], %[offsetof_B_ptr]]\n"
+ "movi v15.16b, #0x1\n"
"mov x9, %x[col_bias]\n"
"bic %x[flags], %x[flags], #0x80000000\n"
"mov x28, %x[output_ptr]\n"
@@ -853,12 +853,12 @@ void a64_hybrid_u8qa_dot_4x16_a55 (
"b 122f\n"
"61:" // Height 3
"movi v11.4s, #0x0\n"
+ "ldr x11, [%x[args_ptr], %[offsetof_N]]\n"
"movi v12.4s, #0x0\n"
+ "ldr x10, [%x[args_ptr], %[offsetof_B_ptr]]\n"
"movi v13.4s, #0x0\n"
- "movi v15.16b, #0x1\n"
- "ldr x11, [%x[args_ptr], %[offsetof_N]]\n"
"mov x9, %x[col_bias]\n"
- "ldr x10, [%x[args_ptr], %[offsetof_B_ptr]]\n"
+ "movi v15.16b, #0x1\n"
"bic %x[flags], %x[flags], #0x80000000\n"
"mov x28, %x[output_ptr]\n"
"62:" // Height 3: Column loop
@@ -1426,14 +1426,14 @@ void a64_hybrid_u8qa_dot_4x16_a55 (
"b 122f\n"
"91:" // Height 4
"movi v11.4s, #0x0\n"
+ "ldr x11, [%x[args_ptr], %[offsetof_N]]\n"
"movi v12.4s, #0x0\n"
+ "ldr x10, [%x[args_ptr], %[offsetof_B_ptr]]\n"
"movi v13.4s, #0x0\n"
+ "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n"
"movi v14.4s, #0x0\n"
- "movi v15.16b, #0x1\n"
- "ldr x11, [%x[args_ptr], %[offsetof_N]]\n"
- "ldr x10, [%x[args_ptr], %[offsetof_B_ptr]]\n"
"mov x9, %x[col_bias]\n"
- "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n"
+ "movi v15.16b, #0x1\n"
"bic %x[flags], %x[flags], #0x80000000\n"
"mov x28, %x[output_ptr]\n"
"mov x19, #0x4\n"