From a50f19346c5b79e2743f882ce0c691c07076f207 Mon Sep 17 00:00:00 2001
From: Pablo Marquez Tello <pablo.tello@arm.com>
Date: Mon, 8 Mar 2021 17:27:05 +0000
Subject: Updated cpu detection

* Added the case in the cpu detection code for Klein cores
* Added has_sve() and set_sve() methods in CpuInfo
* Detection code checks for presence of SVE via HWCAP_SVE
* Updated the heuristic in sve kernels to check for the absence of Klein
* Resolves: COMPMID-4085

Change-Id: I0b8c72ff19dc5a3a81628d121a1afa836e724b4f
Signed-off-by: Pablo Marquez Tello <pablo.tello@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5257
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
---
 src/core/NEON/kernels/arm_gemm/gemm_quint8.cpp | 22 +++++++++++-----------
 1 file changed, 11 insertions(+), 11 deletions(-)

(limited to 'src/core/NEON/kernels/arm_gemm/gemm_quint8.cpp')

diff --git a/src/core/NEON/kernels/arm_gemm/gemm_quint8.cpp b/src/core/NEON/kernels/arm_gemm/gemm_quint8.cpp
index 9720c7d06e..be27b3a117 100644
--- a/src/core/NEON/kernels/arm_gemm/gemm_quint8.cpp
+++ b/src/core/NEON/kernels/arm_gemm/gemm_quint8.cpp
@@ -55,39 +55,39 @@ static const GemmImplementation<uint8_t, uint8_t, Requantize32> gemm_quint8_meth
 {
     GemmMethod::GEMM_INTERLEAVED,
     "sve_interleaved_u8u32_mmla_8x3VL",
-    [](const GemmArgs &args, const Requantize32 &) { return (args._Ksize>8); },
-    nullptr,
+    [](const GemmArgs &args, const Requantize32 &) { return args._ci->has_sve() && (args._Ksize>8); },
+    [](const GemmArgs &args, const Requantize32 &) { return args._ci->get_cpu_model() != CPUModel::KLEIN; },
     [](const GemmArgs &args, const Requantize32 &qp) { return new GemmInterleavedQuantized<cls_sve_interleaved_u8u32_mmla_8x3VL, uint8_t, uint8_t>(args, qp); }
 },
 #endif
 {
     GemmMethod::GEMM_HYBRID_QUANTIZED,
     "sve_smallK_hybrid_u8u32_dot_8x1VL",
-    [](const GemmArgs &args, const Requantize32 &) { return args._Ksize<=64 && !args._indirect_input; },
-    nullptr,
+    [](const GemmArgs &args, const Requantize32 &) { return args._ci->has_sve() && args._Ksize<=64 && !args._indirect_input; },
+    [](const GemmArgs &args, const Requantize32 &) { return args._ci->get_cpu_model() != CPUModel::KLEIN; },
     [](const GemmArgs &args, const Requantize32 &qp) { return new GemmHybridQuantized<cls_sve_smallK_hybrid_u8u32_dot_8x1VL, uint8_t, uint8_t>(args, qp); }
 },
 #ifdef SVE2 // Requantizing kernels include some SVE2 only instructions (SQRDMULH, SRSHL)
 {
     GemmMethod::GEMM_HYBRID,
     "sve_hybrid_u8qa_dot_4x4VL",
-    [](const GemmArgs &, const Requantize32 &qp) { return quant_hybrid_asymmetric(qp); },
-    nullptr,
+    [](const GemmArgs &args, const Requantize32 &qp) { return  args._ci->has_sve() && quant_hybrid_asymmetric(qp); },
+    [](const GemmArgs &args, const Requantize32 &) { return args._ci->get_cpu_model() != CPUModel::KLEIN; },
     [](const GemmArgs &args, const Requantize32 &qp) { return new GemmHybridIndirect<cls_sve_hybrid_u8qa_dot_4x4VL, uint8_t, uint8_t, Requantize32>(args, qp); }
 },
 #endif
 {
     GemmMethod::GEMM_HYBRID,
     "sve_hybrid_u8u32_dot_6x4VL",
-    nullptr,
-    nullptr,
+    [](const GemmArgs &args, const Requantize32 &) { return  args._ci->has_sve(); },
+    [](const GemmArgs &args, const Requantize32 &) { return args._ci->get_cpu_model() != CPUModel::KLEIN; },
     [](const GemmArgs &args, const Requantize32 &qp) { return new GemmHybridIndirect<cls_sve_hybrid_u8u32_dot_6x4VL, uint8_t, uint8_t, Requantize32, true>(args, qp); }
 },
 {
     GemmMethod::GEMM_INTERLEAVED,
     "sve_interleaved_u8u32_dot_8x3VL",
-    [](const GemmArgs &args, const Requantize32 &) { return (args._Ksize>4); },
-    nullptr,
+    [](const GemmArgs &args, const Requantize32 &) { return  args._ci->has_sve() && (args._Ksize>4); },
+    [](const GemmArgs &args, const Requantize32 &) { return  args._ci->get_cpu_model() != CPUModel::KLEIN; },
     [](const GemmArgs &args, const Requantize32 &qp) { return new GemmInterleavedQuantized<cls_sve_interleaved_u8u32_dot_8x3VL, uint8_t, uint8_t>(args, qp); }
 },
 #endif
@@ -96,7 +96,7 @@ static const GemmImplementation<uint8_t, uint8_t, Requantize32> gemm_quint8_meth
     GemmMethod::GEMM_INTERLEAVED,
     "a64_interleaved_u8u32_mmla_8x12",
     [](const GemmArgs &args, const Requantize32 &) { return (args._Ksize>8); },
-    nullptr,
+    [](const GemmArgs &args, const Requantize32 &) { return args._ci->get_cpu_model() != CPUModel::KLEIN; },
     [](const GemmArgs &args, const Requantize32 &qp) { return new GemmInterleavedQuantized<cls_a64_interleaved_u8u32_mmla_8x12, uint8_t, uint8_t>(args, qp); }
 },
 #endif
-- 
cgit v1.2.1