From 4c30de056afe8680b42723b26a2241811715b989 Mon Sep 17 00:00:00 2001
From: Michael Tyler <michael.tyler@arm.com>
Date: Fri, 7 Jul 2023 12:01:32 +0100
Subject: Enable premultiplication for depthwise convolution with fp16 and
 quantized types

Resolves: COMPMID-6337
Change-Id: I81542e51c9c0329f202ac8452f173b138e51a0f6
Signed-off-by: Michael Tyler <michael.tyler@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9883
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
---
 .../depthwise/kernels/a64_s8qs_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp | 2 +-
 .../depthwise/kernels/a64_u8q_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp  | 2 +-
 .../depthwise/kernels/sve_s8q_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp  | 2 +-
 .../depthwise/kernels/sve_s8qs_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp | 2 +-
 .../depthwise/kernels/sve_u8q_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp  | 2 +-
 5 files changed, 5 insertions(+), 5 deletions(-)

(limited to 'src/core/NEON/kernels/arm_conv/depthwise/kernels')

diff --git a/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_s8qs_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp b/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_s8qs_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp
index fc83aaf5d2..bd2941b3d6 100644
--- a/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_s8qs_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp
+++ b/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_s8qs_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp
@@ -64,7 +64,7 @@ class a64_s8qs_nhwc_3x3_s1_output2x2_dot_depthfirst : public DepthwiseDepthfirst
   ) const override
   {
     interleave_a64_s8q_3x3_dot::pack_parameters(
-      args.input_channels, buffer, reinterpret_cast<const int32_t *>(biases),
+      args.input_channels * args.channel_multiplier, buffer, reinterpret_cast<const int32_t *>(biases),
       reinterpret_cast<const int8_t *>(weights), qp, ld_weight_col, ld_weight_row
     );
   }
diff --git a/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_u8q_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp b/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_u8q_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp
index bea97a54b6..ed24f8fa3c 100644
--- a/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_u8q_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp
+++ b/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_u8q_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp
@@ -64,7 +64,7 @@ class a64_u8q_nhwc_3x3_s1_output2x2_dot_depthfirst : public DepthwiseDepthfirstS
   ) const override
   {
     interleave_a64_u8q_3x3_dot::pack_parameters(
-      args.input_channels, buffer, reinterpret_cast<const int32_t *>(biases),
+      args.input_channels * args.channel_multiplier, buffer, reinterpret_cast<const int32_t *>(biases),
       reinterpret_cast<const uint8_t *>(weights), qp, ld_weight_col, ld_weight_row
     );
   }
diff --git a/src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_s8q_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp b/src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_s8q_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp
index 32ea009e8a..04cf0d4036 100644
--- a/src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_s8q_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp
+++ b/src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_s8q_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp
@@ -64,7 +64,7 @@ class sve_s8q_nhwc_3x3_s1_output2x2_dot_depthfirst : public DepthwiseDepthfirstS
   ) const override
   {
     interleave_sve_s8q_3x3_dot::pack_parameters(
-      args.input_channels, buffer, reinterpret_cast<const int32_t *>(biases),
+      args.input_channels * args.channel_multiplier, buffer, reinterpret_cast<const int32_t *>(biases),
       reinterpret_cast<const int8_t *>(weights), qp, ld_weight_col, ld_weight_row
     );
   }
diff --git a/src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_s8qs_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp b/src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_s8qs_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp
index 1730574933..6799b10ed9 100644
--- a/src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_s8qs_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp
+++ b/src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_s8qs_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp
@@ -64,7 +64,7 @@ class sve_s8qs_nhwc_3x3_s1_output2x2_dot_depthfirst : public DepthwiseDepthfirst
   ) const override
   {
     interleave_sve_s8q_3x3_dot::pack_parameters(
-      args.input_channels, buffer, reinterpret_cast<const int32_t *>(biases),
+      args.input_channels * args.channel_multiplier, buffer, reinterpret_cast<const int32_t *>(biases),
       reinterpret_cast<const int8_t *>(weights), qp, ld_weight_col, ld_weight_row
     );
   }
diff --git a/src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_u8q_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp b/src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_u8q_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp
index 9432cd7550..6b006e8d51 100644
--- a/src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_u8q_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp
+++ b/src/core/NEON/kernels/arm_conv/depthwise/kernels/sve_u8q_nhwc_3x3_s1_output2x2_dot_depthfirst.hpp
@@ -64,7 +64,7 @@ class sve_u8q_nhwc_3x3_s1_output2x2_dot_depthfirst : public DepthwiseDepthfirstS
   ) const override
   {
     interleave_sve_u8q_3x3_dot::pack_parameters(
-      args.input_channels, buffer, reinterpret_cast<const int32_t *>(biases),
+      args.input_channels * args.channel_multiplier, buffer, reinterpret_cast<const int32_t *>(biases),
       reinterpret_cast<const uint8_t *>(weights), qp, ld_weight_col, ld_weight_row
     );
   }
-- 
cgit v1.2.1