aboutsummaryrefslogtreecommitdiff
path: root/src/core/NEON/kernels/arm_gemm/gemm_hybrid_quantized.hpp
diff options
context:
space:
mode:
authorGeorgios Pinitas <georgios.pinitas@arm.com>2020-07-02 20:02:20 +0100
committerGeorgios Pinitas <georgios.pinitas@arm.com>2020-07-06 16:51:32 +0000
commit5aa1a0b7ca5eed010e4b297a95b1c4851f741328 (patch)
treeba882de9e86589dfdd33937d538a89bbdf01c40e /src/core/NEON/kernels/arm_gemm/gemm_hybrid_quantized.hpp
parent42550c039105597ff6acd4e5efc0ee3c7c20b08e (diff)
downloadComputeLibrary-5aa1a0b7ca5eed010e4b297a95b1c4851f741328.tar.gz
COMPID-3324: Clean GEMM kernels
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com> Change-Id: I170de1671e061a78740caee31fb4a1b8642c1369 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3505 Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/gemm_hybrid_quantized.hpp')
-rw-r--r--src/core/NEON/kernels/arm_gemm/gemm_hybrid_quantized.hpp23
1 files changed, 6 insertions, 17 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/gemm_hybrid_quantized.hpp b/src/core/NEON/kernels/arm_gemm/gemm_hybrid_quantized.hpp
index 2b936d0b8f..36545c16ba 100644
--- a/src/core/NEON/kernels/arm_gemm/gemm_hybrid_quantized.hpp
+++ b/src/core/NEON/kernels/arm_gemm/gemm_hybrid_quantized.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017-2020 ARM Limited.
+ * Copyright (c) 2017-2020 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -28,9 +28,8 @@
#include <algorithm>
#include "arm_gemm.hpp"
-#include "utils.hpp"
-
#include "ndrange.hpp"
+#include "utils.hpp"
#include "mergeresults.hpp"
#include "transform.hpp"
@@ -151,7 +150,7 @@ public:
// Interface implementation - Compulsory functions
ndrange_t get_window_size() const override {
- return { _window_range.total_size(), 1u, 1u, 1u, 1u, 1u };
+ return { _window_range.total_size() };
}
// This kernel can always be dynamically scheduled.
@@ -159,7 +158,8 @@ public:
return true;
}
- void execute_1d(unsigned int start, unsigned int end, int threadid) {
+ // Execute
+ void execute(const ndcoord_t &work_range, const ndcoord_t &, int threadid) override {
#ifdef CYCLE_PROFILING
profiler prof;
#endif
@@ -180,7 +180,7 @@ public:
unsigned int kmax = std::min(k0 + _k_block, _Ksize);
unsigned int kern_k = roundup(kmax-k0, strategy::k_unroll());
- auto p = _window_range.iterator(start, end);
+ auto p = _window_range.iterator(work_range.get_position(0), work_range.get_position_end(0));
if (p.done()) {
return;
@@ -234,17 +234,6 @@ public:
}
}
- // Execute
- void execute(const ndcoord_t& work_range, const ndcoord_t& thread_locator, int threadid) override {
- UNUSED(thread_locator);
-
- const auto start = work_range.get_position(0);
- const auto size = work_range.get_size(0);
- const auto stop = start + size;
-
- execute_1d(start, stop, threadid);
- }
-
// Working space needed for intermediate result buffers.
size_t get_working_size() const override {
return (_nthreads * strategy::out_height() * _Nsize * sizeof(Tri));