From fa269bb2f150b5f9f00cd8ad5c57ff02613cb628 Mon Sep 17 00:00:00 2001 From: morgolock Date: Tue, 8 Sep 2020 16:00:56 +0100 Subject: MLCE-229: Fixed requantization per channel in asm kernel Change-Id: Iaf1465f3144371e153ce123ac00da5cc092f77df Signed-off-by: morgolock Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3939 Reviewed-by: Michele Di Giorgio Reviewed-by: Georgios Pinitas Comments-Addressed: Arm Jenkins Tested-by: Arm Jenkins --- src/core/NEON/kernels/arm_gemm/quantized.cpp | 1 + src/runtime/NEON/functions/NEGEMMAssemblyDispatch.cpp | 2 +- 2 files changed, 2 insertions(+), 1 deletion(-) diff --git a/src/core/NEON/kernels/arm_gemm/quantized.cpp b/src/core/NEON/kernels/arm_gemm/quantized.cpp index 201bd9dc2c..cac02cf28e 100644 --- a/src/core/NEON/kernels/arm_gemm/quantized.cpp +++ b/src/core/NEON/kernels/arm_gemm/quantized.cpp @@ -137,6 +137,7 @@ void requantize_block_32_int(const Requantize32 &qp, unsigned int width, unsigne v_shf1l = vld1q_s32(perch_shiftl_ptr + 4); v_shf2l = vld1q_s32(perch_shiftl_ptr + 8); v_shf3l = vld1q_s32(perch_shiftl_ptr + 12); + perch_shiftl_ptr += 16; } } else { v_mul0=v_mul1=v_mul2=v_mul3=v_mul; diff --git a/src/runtime/NEON/functions/NEGEMMAssemblyDispatch.cpp b/src/runtime/NEON/functions/NEGEMMAssemblyDispatch.cpp index eeea3a45ee..30232b4435 100644 --- a/src/runtime/NEON/functions/NEGEMMAssemblyDispatch.cpp +++ b/src/runtime/NEON/functions/NEGEMMAssemblyDispatch.cpp @@ -252,7 +252,7 @@ std::tuple Fallback 0 && !need_left) + if(s < 0 && !need_left) { need_left = true; } -- cgit v1.2.1