From 905786ea0c1abb2b8df36c56eae93a97823cace1 Mon Sep 17 00:00:00 2001
From: Milos Puzovic <milos.puzovic@arm.com>
Date: Tue, 26 Mar 2024 14:34:30 +0000
Subject: Added new NEON fixed format fast math mode hybrid kernel with maximum
 height of 6 for accumulation and updated heuristics

Change-Id: Ib52ea6825e164f4a8b8422eab7991b50af0b0d7c
Signed-off-by: Milos Puzovic <milos.puzovic@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/11354
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Jakub Sujak <jakub.sujak@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
---
 src/core/NEON/kernels/arm_gemm/gemm_fp32.cpp | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)

(limited to 'src/core/NEON/kernels/arm_gemm/gemm_fp32.cpp')

diff --git a/src/core/NEON/kernels/arm_gemm/gemm_fp32.cpp b/src/core/NEON/kernels/arm_gemm/gemm_fp32.cpp
index 44a7bb894a..e85dd59425 100644
--- a/src/core/NEON/kernels/arm_gemm/gemm_fp32.cpp
+++ b/src/core/NEON/kernels/arm_gemm/gemm_fp32.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2023 Arm Limited.
+ * Copyright (c) 2017-2024 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -34,6 +34,7 @@
 #ifdef ARM_COMPUTE_ENABLE_FIXED_FORMAT_KERNELS
 #include "kernels/a64_ffhybrid_fp32_mla_6x16.hpp"
 #include "kernels/a64_ffhybrid_fp32bf16fp32_mmla_4x24.hpp"
+#include "kernels/a64_ffhybrid_fp32bf16fp32_mmla_6x16.hpp"
 #include "kernels/a64_ffinterleaved_bf16fp32_mmla_8x12.hpp"
 #include "kernels/a64_ffinterleaved_fp32_mla_8x12.hpp"
 #endif // ARM_COMPUTE_ENABLE_FIXED_FORMAT_KERNELS
@@ -350,6 +351,14 @@ GemmImplementation<float, float>::with_estimate(
     [](const GemmArgs &args) { return GemmHybridIndirectFixedFormat<cls_a64_ffhybrid_fp32bf16fp32_mmla_4x24, float, float>::estimate_cycles<float>(args); },
     [](const GemmArgs &args) { return new GemmHybridIndirectFixedFormat<cls_a64_ffhybrid_fp32bf16fp32_mmla_4x24, float, float>(args); }
 ),
+GemmImplementation<float, float>::with_estimate(
+    GemmMethod::GEMM_HYBRID,
+    "a64_ffhybrid_fp32bf16fp32_mmla_6x16",
+    KernelWeightFormat::VL256_BL64_BF16,
+    [](const GemmArgs &args) { return args._fast_mode && args._ci->has_bf16(); },
+    [](const GemmArgs &args) { return GemmHybridIndirectFixedFormat<cls_a64_ffhybrid_fp32bf16fp32_mmla_6x16, float, float>::estimate_cycles<float>(args); },
+    [](const GemmArgs &args) { return new GemmHybridIndirectFixedFormat<cls_a64_ffhybrid_fp32bf16fp32_mmla_6x16, float, float>(args); }
+),
 #endif // BF16
 GemmImplementation<float, float>::with_estimate(
     GemmMethod::GEMM_INTERLEAVED,
-- 
cgit v1.2.1