apply and fix

- apply rv64 patch - fix arm tests - fix the comments
openvinotoolkit · Dec 24, 2024 · 6720e7f · 6720e7f
1 parent c576825
commit 6720e7f
Show file tree

Hide file tree

Showing 11 changed files with 42 additions and 16 deletions.
diff --git a/src/plugins/intel_cpu/src/emitters/plugin/x64/jit_dnnl_emitters.cpp b/src/plugins/intel_cpu/src/emitters/plugin/x64/jit_dnnl_emitters.cpp
@@ -45,12 +45,19 @@ jit_dnnl_emitter::jit_dnnl_emitter(jit_generator* host,
 
 void jit_dnnl_emitter::set_injector() {
     if (host_isa_ == cpu::x64::sse41) {
-        eltwise_injector_sse42 = std::make_shared<jit_uni_eltwise_injector<cpu::x64::sse41>>(h, kind, alpha, beta, 1.f);
+        eltwise_injector_sse42 =
+            std::make_shared<jit_uni_eltwise_injector<cpu::x64::sse41>>(h, kind, alpha, beta, 1.f, data_type::f32);
     } else if (host_isa_ == cpu::x64::avx2) {
-        eltwise_injector_avx2 = std::make_shared<jit_uni_eltwise_injector<cpu::x64::avx2>>(h, kind, alpha, beta, 1.f);
+        eltwise_injector_avx2 =
+            std::make_shared<jit_uni_eltwise_injector<cpu::x64::avx2>>(h, kind, alpha, beta, 1.f, data_type::f32);
     } else if (host_isa_ == cpu::x64::avx512_core) {
         eltwise_injector_avx512_core =
-            std::make_shared<jit_uni_eltwise_injector<cpu::x64::avx512_core>>(h, kind, alpha, beta, 1.f);
+            std::make_shared<jit_uni_eltwise_injector<cpu::x64::avx512_core>>(h,
+                                                                              kind,
+                                                                              alpha,
+                                                                              beta,
+                                                                              1.f,
+                                                                              data_type::f32);
     } else {
         OV_CPU_JIT_EMITTER_THROW("Unsupported ISA ", host_isa_);
     }

diff --git a/src/plugins/intel_cpu/src/nodes/common/softmax.cpp b/src/plugins/intel_cpu/src/nodes/common/softmax.cpp
@@ -16,6 +16,7 @@
 #include "utils/bfloat16.hpp"
 
 using namespace dnnl;
+using namespace dnnl::impl;
 using namespace dnnl::impl::cpu;
 using namespace dnnl::impl::cpu::x64;
 using namespace dnnl::impl::utils;
@@ -67,7 +68,8 @@ struct jit_uni_softmax_kernel_f32 : public jit_uni_softmax_kernel, public jit_ge
     }
 
     void generate() override {
-        exp_injector.reset(new jit_uni_eltwise_injector<isa>(this, dnnl::impl::alg_kind::eltwise_exp, 0.f, 0.f, 1.0f));
+        exp_injector.reset(
+            new jit_uni_eltwise_injector<isa>(this, dnnl::impl::alg_kind::eltwise_exp, 0.f, 0.f, 1.0f, data_type::f32));
 
         if (mayiuse(avx512_core))
             uni_vcvtneps2bf16.reset(new jit_uni_vcvtneps2bf16(this, isa));

diff --git a/src/plugins/intel_cpu/src/nodes/interpolate.cpp b/src/plugins/intel_cpu/src/nodes/interpolate.cpp
@@ -77,7 +77,8 @@ struct jit_uni_interpolate_kernel_f32 : public jit_uni_interpolate_kernel, publi
                                                                                             post_op.eltwise.alg,
                                                                                             post_op.eltwise.alpha,
                                                                                             post_op.eltwise.beta,
-                                                                                            1.f));
+                                                                                            1.f,
+                                                                                            data_type::f32));
             } else if (post_op.is_depthwise()) {
                 depthwise_injectors.push_back(std::make_shared<jit_uni_depthwise_injector_f32<isa>>(this, post_op));
             } else if (post_op.is_quantization()) {

diff --git a/src/plugins/intel_cpu/src/nodes/kernels/x64/non_max_suppression.cpp b/src/plugins/intel_cpu/src/nodes/kernels/x64/non_max_suppression.cpp
@@ -6,6 +6,7 @@
 
 #include "utils/general_utils.h"
 
+using namespace dnnl::impl;
 using namespace dnnl::impl::cpu;
 
 #define GET_OFF(field) offsetof(NmsCallArgs, field)
@@ -19,7 +20,8 @@ void NonMaxSuppression<isa>::generate() {
     load_vector_emitter.reset(new jit_load_emitter(this, isa, ov::element::f32, ov::element::f32, vector_step));
     load_scalar_emitter.reset(new jit_load_emitter(this, isa, ov::element::f32, ov::element::f32, scalar_step));
 
-    exp_injector.reset(new x64::jit_uni_eltwise_injector<isa>(this, dnnl::impl::alg_kind::eltwise_exp, 0.f, 0.f, 1.f));
+    exp_injector.reset(
+        new x64::jit_uni_eltwise_injector<isa>(this, dnnl::impl::alg_kind::eltwise_exp, 0.f, 0.f, 1.f, data_type::f32));
 
     this->preamble();
 

diff --git a/src/plugins/intel_cpu/src/nodes/mvn.cpp b/src/plugins/intel_cpu/src/nodes/mvn.cpp
@@ -953,7 +953,8 @@ struct jit_uni_mvn_kernel_f32 : public jit_uni_mvn_kernel, public jit_generator
                                                                                             post_op.eltwise.alg,
                                                                                             post_op.eltwise.alpha,
                                                                                             post_op.eltwise.beta,
-                                                                                            post_op.eltwise.scale));
+                                                                                            post_op.eltwise.scale,
+                                                                                            data_type::f32));
             } else if (post_op.is_depthwise()) {
                 depthwise_injectors.push_back(std::make_shared<jit_uni_depthwise_injector_f32<isa>>(this, post_op));
             } else if (post_op.is_quantization()) {

diff --git a/src/plugins/intel_cpu/src/nodes/normalize.cpp b/src/plugins/intel_cpu/src/nodes/normalize.cpp
@@ -228,7 +228,8 @@ struct jit_uni_normalize_kernel_f32 : public jit_uni_normalize_kernel, public ji
                                                                                             post_op.eltwise.alg,
                                                                                             post_op.eltwise.alpha,
                                                                                             post_op.eltwise.beta,
-                                                                                            post_op.eltwise.scale));
+                                                                                            post_op.eltwise.scale,
+                                                                                            data_type::f32));
             } else if (post_op.is_depthwise()) {
                 depthwise_injectors.push_back(std::make_shared<jit_uni_depthwise_injector_f32<isa>>(this, post_op));
             } else if (post_op.is_quantization()) {

diff --git a/src/plugins/intel_cpu/src/nodes/reduce.cpp b/src/plugins/intel_cpu/src/nodes/reduce.cpp
@@ -142,7 +142,12 @@ struct jit_uni_reduce_kernel_f32 : public jit_uni_reduce_kernel, public jit_gene
 
     void generate() override {
         if (jcp_.reduce_mode == Algorithm::ReduceLogSumExp) {
-            exp_injector = std::make_shared<jit_uni_eltwise_injector<isa>>(this, alg_kind::eltwise_exp, 0.f, 0.f, 1.f);
+            exp_injector = std::make_shared<jit_uni_eltwise_injector<isa>>(this,
+                                                                           alg_kind::eltwise_exp,
+                                                                           0.f,
+                                                                           0.f,
+                                                                           1.f,
+                                                                           data_type::f32);
         }
 
         if (mayiuse(avx512_core))
@@ -1209,7 +1214,8 @@ struct jit_uni_reduce_post_kernel_f32 : public jit_uni_reduce_post_kernel, publi
                                                                                             post_op.eltwise.alg,
                                                                                             post_op.eltwise.alpha,
                                                                                             post_op.eltwise.beta,
-                                                                                            post_op.eltwise.scale));
+                                                                                            post_op.eltwise.scale,
+                                                                                            data_type::f32));
             } else if (post_op.is_depthwise()) {
                 depthwise_injectors.push_back(std::make_shared<jit_uni_depthwise_injector_f32<isa>>(this, post_op));
             } else if (post_op.is_quantization()) {
@@ -1223,7 +1229,12 @@ struct jit_uni_reduce_post_kernel_f32 : public jit_uni_reduce_post_kernel, publi
         }
 
         if (jcp_.reduce_mode == Algorithm::ReduceLogSum || jcp_.reduce_mode == Algorithm::ReduceLogSumExp) {
-            log_injector = std::make_shared<jit_uni_eltwise_injector<isa>>(this, alg_kind::eltwise_log, 0.f, 0.f, 1.f);
+            log_injector = std::make_shared<jit_uni_eltwise_injector<isa>>(this,
+                                                                           alg_kind::eltwise_log,
+                                                                           0.f,
+                                                                           0.f,
+                                                                           1.f,
+                                                                           data_type::f32);
         }
 
         if (mayiuse(avx512_core))

diff --git a/src/plugins/intel_cpu/src/nodes/region_yolo.cpp b/src/plugins/intel_cpu/src/nodes/region_yolo.cpp
@@ -18,6 +18,7 @@
 #include "openvino/opsets/opset1.hpp"
 #include "utils/bfloat16.hpp"
 
+using namespace dnnl::impl;
 using namespace dnnl::impl::cpu;
 using namespace dnnl::impl::cpu::x64;
 using namespace dnnl::impl::utils;
@@ -45,7 +46,8 @@ struct jit_uni_logistic_kernel_f32 : public jit_uni_logistic_kernel, public jit_
     }
 
     void generate() override {
-        exp_injector.reset(new jit_uni_eltwise_injector<isa>(this, dnnl::impl::alg_kind::eltwise_exp, 0.f, 0.f, 1.f));
+        exp_injector.reset(
+            new jit_uni_eltwise_injector<isa>(this, dnnl::impl::alg_kind::eltwise_exp, 0.f, 0.f, 1.f, data_type::f32));
 
         if (mayiuse(avx512_core))
             uni_vcvtneps2bf16.reset(new jit_uni_vcvtneps2bf16(this, isa));

diff --git a/src/plugins/intel_cpu/src/utils/verbose.cpp b/src/plugins/intel_cpu/src/utils/verbose.cpp
@@ -107,8 +107,7 @@ void Verbose::printInfo() {
         shift(written);
         written = snprintf(portsInfo + written_total, CPU_VERBOSE_DAT_LEN - written_total, "%s", prefix.c_str());
         shift(written);
-        std::string fmt_str =
-            dnnl::impl::md2fmt_str(prefix.c_str(), desc, dnnl::impl::format_kind_t::dnnl_format_kind_undef);
+        std::string fmt_str = dnnl::impl::md2fmt_str("", desc, dnnl::impl::format_kind_t::dnnl_format_kind_undef);
         written = snprintf(portsInfo + written_total, CPU_VERBOSE_DAT_LEN - written_total, "%s", fmt_str.c_str());
         shift(written);
         written = snprintf(portsInfo + written_total, CPU_VERBOSE_DAT_LEN - written_total, ":");

diff --git a/src/plugins/intel_cpu/tests/functional/shared_tests_instances/skip_tests_config.cpp b/src/plugins/intel_cpu/tests/functional/shared_tests_instances/skip_tests_config.cpp
@@ -518,7 +518,7 @@ std::vector<std::string> disabledTestPatterns() {
         // tests are useless on such platforms
         retVector.emplace_back(R"(.*(BF|bf)16.*)");
         retVector.emplace_back(R"(.*bfloat16.*)");
-        // Issue: MFDNN-12876
+        // Issue: MFDNN-12818
         retVector.emplace_back(R"(.*smoke_LPT/RecurrentCellTransformation.CompareWithRefImpl/f32_\[1,1,3\]_CPU_f32FQ_X_level=256_.*_FQ_W_level=255.*)");
         retVector.emplace_back(R"(.*smoke_static/ConvertFqRnnToQuantizedRnn.CompareWithRefs/Type=GRUSequence.*2.5.10.*2.1.4.*2.1.4.*)");
     }

diff --git a/src/plugins/intel_cpu/thirdparty/onednn b/src/plugins/intel_cpu/thirdparty/onednn
+4 −1		cmake/platform.cmake
+0 −15		src/common/memory_tracking.hpp
+2 −0		src/cpu/acl/acl_batch_normalization.hpp
+15 −189		src/cpu/acl/acl_binary.cpp
+210 −13		src/cpu/acl/acl_binary.hpp
+131 −0		src/cpu/acl/acl_convolution_utils.cpp
+42 −89		src/cpu/acl/acl_convolution_utils.hpp
+4 −1		src/cpu/acl/acl_deconvolution.hpp
+10 −61		src/cpu/acl/acl_depthwise_convolution.cpp
+91 −9		src/cpu/acl/acl_depthwise_convolution.hpp
+16 −105		src/cpu/acl/acl_gemm_convolution.cpp
+100 −22		src/cpu/acl/acl_gemm_convolution.hpp
+13 −98		src/cpu/acl/acl_indirect_gemm_convolution.cpp
+97 −12		src/cpu/acl/acl_indirect_gemm_convolution.hpp
+2 −0		src/cpu/acl/acl_inner_product.hpp
+8 −0		src/cpu/acl/acl_post_ops.hpp
+14 −129		src/cpu/acl/acl_winograd_convolution.cpp
+99 −18		src/cpu/acl/acl_winograd_convolution.hpp
+7 −3		src/cpu/acl/matmul/acl_lowp_matmul.cpp
+58 −275		src/cpu/acl/matmul/acl_matmul.cpp
+157 −7		src/cpu/acl/matmul/acl_matmul.hpp
+7 −20		src/cpu/acl/matmul/acl_matmul_utils.cpp
+11 −23		src/cpu/acl/matmul/acl_matmul_utils.hpp
+2 −2		src/cpu/rv64/CMakeLists.txt
+1 −0		src/cpu/rv64/rvv_nchw_pooling.hpp
+7 −3		src/cpu/x64/brgemm/brgemm_utils.cpp
+1 −1		src/cpu/x64/matmul/brgemm_matmul.cpp
+2 −2		src/cpu/x64/matmul/brgemm_matmul_utils.cpp
+5 −5		tests/benchdnn/inputs/matmul/test_matmul_ci
+1 −1		tests/benchdnn/inputs/matmul/test_matmul_int8