From 4ee8b1599dbaf7634d25607fa5ac96ba3dc6b0f2 Mon Sep 17 00:00:00 2001 From: Georgios Pinitas Date: Fri, 16 Jul 2021 16:16:43 +0100 Subject: Update GEMM assembly kernels - Introduce Fp32 kernels with internal calculations in Bfloat16 when fast_mode is enabled - Improve kernel selection heuristics Signed-off-by: Georgios Pinitas Change-Id: I68a9e7e862b6fd2721b46e0d7cc791091c4ab279 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5965 Tested-by: Arm Jenkins Comments-Addressed: Arm Jenkins --- .../arm_gemm/kernels/a64_hybrid_s8qa_dot_4x16/a55.cpp | 18 +++++++++--------- 1 file changed, 9 insertions(+), 9 deletions(-) (limited to 'src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_s8qa_dot_4x16/a55.cpp') diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_s8qa_dot_4x16/a55.cpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_s8qa_dot_4x16/a55.cpp index 11aa05a9b7..ee7e55f179 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_s8qa_dot_4x16/a55.cpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_s8qa_dot_4x16/a55.cpp @@ -406,10 +406,10 @@ void a64_hybrid_s8qa_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_s8qa_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_s8qa_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" -- cgit v1.2.1