commit 146138378c1587b7297d245b7177641315f6180b
author Georgios Pinitas <georgios.pinitas@arm.com> Fri Mar 01 19:07:11 2019 +0000
committer Georgios Pinitas <georgios.pinitas@arm.com> Tue Mar 19 11:53:45 2019 +0000
tree 4689218b48a0884418099f373015429f2845ceaf
parent 3dd5b6884a65c06bcb9d15589ee2dc2978e3b336

COMPMID-1995: Update RSH GEMM assembly kernels.

-Updates u8/s8 hybrid dot product kernels to work for any N and any K >=16.
-Adds hybrid FP32 kernels with generic and A55 variants.
-Adds SVE native kernels for fp16/u8/s8.

Change-Id: Ifc0eaba9e3c8ea5bb19d334e870e1b39e4e7e728
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/c/863
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com>

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 8bc33cc..6869279 100644
--- a/src/core/NEON/kernels/arm_gemm/gemm_fp32.cpp
+++ b/src/core/NEON/kernels/arm_gemm/gemm_fp32.cpp
@@ -32,6 +32,7 @@
 #include "gemv_pretransposed.hpp"
 
 #include "kernels/a32_sgemm_8x6.hpp"
+#include "kernels/a64_hybrid_fp32_mla_16x4.hpp"
 #include "kernels/a64_sgemm_12x8.hpp"
 #include "kernels/a64_sgemm_native_16x4.hpp"
 #include "kernels/a64_sgemm_nativeA_pretransposeB_16x4.hpp"
@@ -112,6 +113,13 @@
     [](const GemmArgs<float> &args) { return new GemmHybrid<sgemm_nativeA_pretransposeB_16x4, float, float>(args); }
 },
 {
+    GemmMethod::GEMM_HYBRID,
+    "hybrid_fp32_mla_16x4",
+    [](const GemmArgs<float> &args) { return (args._Ksize >= 4) && (args._alpha == 1.0f) && !args._trA && args._pretransposed_hint; },
+    [](const GemmArgs<float> &args) { return ((args._Ksize <= 256) && (args._Nsize <= 256)) || ((args._nmulti > 1) && ((args._Msize / args._maxthreads) < 8)); },
+    [](const GemmArgs<float> &args) { return new GemmHybrid<hybrid_fp32_mla_16x4, float, float>(args); }
+},
+{
     GemmMethod::GEMM_NATIVE,
     "sgemm_native_16x4",
     [](const GemmArgs<float> &args) { return (args._Ksize>4 && (args._Nsize % 16)==0 && args._alpha==1.0f && !args._trA && !args._trB); },
@@ -165,6 +173,6 @@
 template UniqueGemmCommon<float, float> gemm<float, float>(const GemmArgs<float> &args);
 template KernelDescription get_gemm_method<float, float>(const GemmArgs<float> &args);
 template bool method_is_compatible<float, float>(GemmMethod method, const GemmArgs<float> &args);
-template std::vector<std::string> get_compatible_kernels<float, float> (const GemmArgs<float> &args);
+template std::vector<KernelDescription> get_compatible_kernels<float, float> (const GemmArgs<float> &args);
 
-} // namespace arm_gemm
\ No newline at end of file
+} // namespace arm_gemm