IntenseWebs/openvino
3
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[CPU] Remove unused emitter parameters and API (#16117)

Browse Source
This commit is contained in:
Chenhu Wang
2023-03-08 18:15:44 +08:00
committed by GitHub
parent 6514b7600a
commit 0786a963ab
12 changed files with 127 additions and 417 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/plugins/intel_cpu/src/emitters/jit_bf16_emitters.hpp
View File

@@ -20,9 +20,7 @@ public:
size_t get_inputs_num() const override { return 1; }
private:
void emit_impl(const std::vector<size_t>& in_vec_idxs, const std::vector<size_t>& out_vec_idxs,
const std::vector<size_t>& pool_vec_idxs, const std::vector<size_t>& pool_gpr_idxs,
const emitter_context *emit_context) const override {
void emit_impl(const std::vector<size_t>& in_vec_idxs, const std::vector<size_t>& out_vec_idxs) const override {
if (host_isa_ == dnnl::impl::cpu::x64::avx512_core) {
emit_isa<dnnl::impl::cpu::x64::avx512_core>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == dnnl::impl::cpu::x64::avx2) {

8
src/plugins/intel_cpu/src/emitters/jit_conversion_emitters.cpp
View File

@@ -68,9 +68,7 @@ bool jit_convert_truncation_emitter::is_i8_and_u8_case() const {
one_of(output_type, ov::element::i8, ov::element::u8);
}
void jit_convert_truncation_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_convert_truncation_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
validate_types();
if (host_isa_ == cpu::x64::sse41) {
emit_isa<cpu::x64::sse41>(in_vec_idxs, out_vec_idxs);
@@ -185,9 +183,7 @@ jit_convert_saturation_emitter::jit_convert_saturation_emitter(jit_generator *ho
: jit_convert_emitter(host, host_isa, node, exec_prc) {
}
void jit_convert_saturation_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_convert_saturation_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
validate_types();
if (host_isa_ == cpu::x64::sse41) {
emit_isa<cpu::x64::sse41>(in_vec_idxs, out_vec_idxs);

8
src/plugins/intel_cpu/src/emitters/jit_conversion_emitters.hpp
View File

@@ -49,9 +49,7 @@ public:
const std::shared_ptr<ngraph::Node>& n, InferenceEngine::Precision exec_prc = InferenceEngine::Precision::FP32);
private:
void emit_impl(const std::vector<size_t>& in, const std::vector<size_t>& out,
const std::vector<size_t>& pool, const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t>& in, const std::vector<size_t>& out) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -72,9 +70,7 @@ public:
const std::shared_ptr<ngraph::Node>& n, InferenceEngine::Precision exec_prc = InferenceEngine::Precision::FP32);
private:
void emit_impl(const std::vector<size_t>& in, const std::vector<size_t>& out,
const std::vector<size_t>& pool, const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t>& in, const std::vector<size_t>& out) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;

4
src/plugins/intel_cpu/src/emitters/jit_dnnl_emitters.hpp
View File

@@ -18,9 +18,7 @@ public:
void emit_data() const override;
void emit_impl(const std::vector<size_t> &in_idxs, const std::vector<size_t> &out_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context = nullptr) const override {};
void emit_impl(const std::vector<size_t> &in_idxs, const std::vector<size_t> &out_idxs) const override {};
protected:
jit_dnnl_emitter(dnnl::impl::cpu::x64::jit_generator *host, dnnl::impl::cpu::x64::cpu_isa_t host_isa,

133
src/plugins/intel_cpu/src/emitters/jit_eltwise_emitters.cpp
View File

@@ -24,9 +24,7 @@ jit_add_emitter::jit_add_emitter(x64::jit_generator *host, x64::cpu_isa_t host_i
size_t jit_add_emitter::get_inputs_num() const { return 2; }
void jit_add_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_add_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -73,9 +71,7 @@ jit_mul_add_emitter::jit_mul_add_emitter(x64::jit_generator *host, x64::cpu_isa_
size_t jit_mul_add_emitter::get_inputs_num() const { return 3; }
void jit_mul_add_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_mul_add_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -166,9 +162,7 @@ jit_subtract_emitter::jit_subtract_emitter(x64::jit_generator *host, x64::cpu_is
size_t jit_subtract_emitter::get_inputs_num() const { return 2; }
void jit_subtract_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_subtract_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -215,9 +209,7 @@ jit_multiply_emitter::jit_multiply_emitter(x64::jit_generator *host, x64::cpu_is
size_t jit_multiply_emitter::get_inputs_num() const { return 2; }
void jit_multiply_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_multiply_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -264,9 +256,7 @@ jit_divide_emitter::jit_divide_emitter(x64::jit_generator *host, x64::cpu_isa_t
size_t jit_divide_emitter::get_inputs_num() const { return 2; }
void jit_divide_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_divide_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -331,9 +321,7 @@ jit_floor_emitter::jit_floor_emitter(x64::jit_generator *host, x64::cpu_isa_t ho
size_t jit_floor_emitter::get_inputs_num() const { return 1; }
void jit_floor_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_floor_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -362,9 +350,7 @@ jit_ceiling_emitter::jit_ceiling_emitter(x64::jit_generator* host, x64::cpu_isa_
size_t jit_ceiling_emitter::get_inputs_num() const { return 1; }
void jit_ceiling_emitter::emit_impl(const std::vector<size_t>& in_vec_idxs,
const std::vector<size_t>& out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
const std::vector<size_t>& out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -392,9 +378,7 @@ jit_floor_mod_emitter::jit_floor_mod_emitter(x64::jit_generator *host, x64::cpu_
size_t jit_floor_mod_emitter::get_inputs_num() const { return 2; }
void jit_floor_mod_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_floor_mod_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -444,9 +428,7 @@ jit_mod_emitter::jit_mod_emitter(x64::jit_generator *host, x64::cpu_isa_t host_i
size_t jit_mod_emitter::get_inputs_num() const { return 2; }
void jit_mod_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_mod_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -496,9 +478,7 @@ jit_maximum_emitter::jit_maximum_emitter(x64::jit_generator *host, x64::cpu_isa_
size_t jit_maximum_emitter::get_inputs_num() const { return 2; }
void jit_maximum_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_maximum_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -546,9 +526,7 @@ jit_minimum_emitter::jit_minimum_emitter(x64::jit_generator *host, x64::cpu_isa_
size_t jit_minimum_emitter::get_inputs_num() const { return 2; }
void jit_minimum_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_minimum_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -597,9 +575,7 @@ jit_squared_difference_emitter::jit_squared_difference_emitter(x64::jit_generato
size_t jit_squared_difference_emitter::get_inputs_num() const { return 2; }
void jit_squared_difference_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_squared_difference_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -654,9 +630,7 @@ jit_power_dynamic_emitter::jit_power_dynamic_emitter(x64::jit_generator *host, x
size_t jit_power_dynamic_emitter::get_inputs_num() const { return 2; }
void jit_power_dynamic_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_power_dynamic_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -767,9 +741,7 @@ jit_equal_emitter::jit_equal_emitter(x64::jit_generator *host, x64::cpu_isa_t ho
size_t jit_equal_emitter::get_inputs_num() const { return 2; }
void jit_equal_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_equal_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -828,9 +800,7 @@ jit_not_equal_emitter::jit_not_equal_emitter(x64::jit_generator *host, x64::cpu_
size_t jit_not_equal_emitter::get_inputs_num() const { return 2; }
void jit_not_equal_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_not_equal_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -889,9 +859,7 @@ jit_greater_emitter::jit_greater_emitter(x64::jit_generator *host, x64::cpu_isa_
size_t jit_greater_emitter::get_inputs_num() const { return 2; }
void jit_greater_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_greater_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -951,9 +919,7 @@ jit_greater_equal_emitter::jit_greater_equal_emitter(x64::jit_generator *host, x
size_t jit_greater_equal_emitter::get_inputs_num() const { return 2; }
void jit_greater_equal_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_greater_equal_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -1012,9 +978,7 @@ jit_less_emitter::jit_less_emitter(x64::jit_generator *host, x64::cpu_isa_t host
size_t jit_less_emitter::get_inputs_num() const { return 2; }
void jit_less_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_less_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -1073,9 +1037,7 @@ jit_less_equal_emitter::jit_less_equal_emitter(x64::jit_generator *host, x64::cp
size_t jit_less_equal_emitter::get_inputs_num() const { return 2; }
void jit_less_equal_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_less_equal_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -1135,9 +1097,7 @@ jit_logical_and_emitter::jit_logical_and_emitter(x64::jit_generator *host, x64::
size_t jit_logical_and_emitter::get_inputs_num() const { return 2; }
void jit_logical_and_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_logical_and_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -1217,9 +1177,7 @@ jit_logical_or_emitter::jit_logical_or_emitter(x64::jit_generator *host, x64::cp
size_t jit_logical_or_emitter::get_inputs_num() const { return 2; }
void jit_logical_or_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_logical_or_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -1298,9 +1256,7 @@ jit_logical_xor_emitter::jit_logical_xor_emitter(x64::jit_generator *host, x64::
size_t jit_logical_xor_emitter::get_inputs_num() const { return 2; }
void jit_logical_xor_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_logical_xor_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -1379,9 +1335,7 @@ jit_logical_not_emitter::jit_logical_not_emitter(x64::jit_generator *host, x64::
size_t jit_logical_not_emitter::get_inputs_num() const { return 1; }
void jit_logical_not_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_logical_not_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -1451,9 +1405,7 @@ jit_power_static_emitter::jit_power_static_emitter(x64::jit_generator *host, x64
size_t jit_power_static_emitter::get_inputs_num() const { return 1; }
void jit_power_static_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_power_static_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -1627,9 +1579,7 @@ jit_prelu_emitter::jit_prelu_emitter(x64::jit_generator *host, x64::cpu_isa_t ho
}
size_t jit_prelu_emitter::get_inputs_num() const { return 2; }
void jit_prelu_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_prelu_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -1684,9 +1634,7 @@ jit_sqrt_emitter::jit_sqrt_emitter(x64::jit_generator *host, x64::cpu_isa_t host
size_t jit_sqrt_emitter::get_inputs_num() const { return 1; }
void jit_sqrt_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_sqrt_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -1713,9 +1661,7 @@ jit_negative_emitter::jit_negative_emitter(x64::jit_generator *host, x64::cpu_is
size_t jit_negative_emitter::get_inputs_num() const { return 1; }
void jit_negative_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_negative_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -1751,10 +1697,7 @@ size_t jit_erf_emitter::get_inputs_num() const { return 1; }
void jit_erf_emitter::emit_impl(
const std::vector<size_t> &in_vec_idxs,
const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs,
const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -1932,9 +1875,7 @@ jit_soft_sign_emitter::jit_soft_sign_emitter(x64::jit_generator *host, x64::cpu_
size_t jit_soft_sign_emitter::get_inputs_num() const { return 1; }
void jit_soft_sign_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_soft_sign_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -1996,9 +1937,7 @@ void jit_is_finite_emitter::emit_isa(const std::vector<size_t> &in_vec_idxs, con
}
}
void jit_is_finite_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_is_finite_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::avx512_core) {
emit_isa<x64::avx512_core>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -2071,9 +2010,7 @@ void jit_is_inf_emitter::emit_isa(const std::vector<size_t> &in_vec_idxs, const
}
}
void jit_is_inf_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_is_inf_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::avx512_core) {
emit_isa<x64::avx512_core>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -2123,9 +2060,7 @@ void jit_is_nan_emitter::emit_isa(const std::vector<size_t> &in_vec_idxs, const
}
}
void jit_is_nan_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_is_nan_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::avx512_core) {
emit_isa<x64::avx512_core>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {
@@ -2160,9 +2095,7 @@ size_t jit_select_emitter::aux_vecs_count() const {
return 2;
}
void jit_select_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_select_emitter::emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const {
if (host_isa_ == x64::sse41) {
emit_isa<x64::sse41>(in_vec_idxs, out_vec_idxs);
} else if (host_isa_ == x64::avx2) {

133
src/plugins/intel_cpu/src/emitters/jit_eltwise_emitters.hpp
View File

@@ -20,9 +20,7 @@ public:
static std::set<InferenceEngine::Precision> get_supported_precisions();
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -39,9 +37,7 @@ public:
static std::set<InferenceEngine::Precision> get_supported_precisions();
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -61,9 +57,7 @@ public:
static std::set<InferenceEngine::Precision> get_supported_precisions();
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -81,9 +75,7 @@ public:
static std::set<InferenceEngine::Precision> get_supported_precisions();
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -101,9 +93,7 @@ public:
static std::set<InferenceEngine::Precision> get_supported_precisions();
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -120,9 +110,7 @@ public:
size_t get_inputs_num() const override;
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -138,9 +126,7 @@ public:
size_t get_inputs_num() const override;
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -156,9 +142,7 @@ public:
size_t get_inputs_num() const override;
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -176,9 +160,7 @@ public:
size_t get_inputs_num() const override;
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -197,9 +179,7 @@ public:
static std::set<InferenceEngine::Precision> get_supported_precisions();
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -217,9 +197,7 @@ public:
static std::set<InferenceEngine::Precision> get_supported_precisions();
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -238,9 +216,7 @@ public:
static std::set<InferenceEngine::Precision> get_supported_precisions();
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -257,9 +233,7 @@ public:
size_t get_inputs_num() const override;
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -276,9 +250,7 @@ public:
size_t get_inputs_num() const override;
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -298,9 +270,7 @@ public:
size_t get_inputs_num() const override;
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -320,9 +290,7 @@ public:
size_t get_inputs_num() const override;
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -342,9 +310,7 @@ public:
size_t get_inputs_num() const override;
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -364,9 +330,7 @@ public:
size_t get_inputs_num() const override;
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -387,9 +351,7 @@ public:
size_t get_inputs_num() const override;
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -409,9 +371,7 @@ public:
size_t get_inputs_num() const override;
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -431,9 +391,7 @@ public:
size_t get_inputs_num() const override;
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -453,9 +411,7 @@ public:
size_t get_inputs_num() const override;
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -474,9 +430,7 @@ public:
size_t get_inputs_num() const override;
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -496,9 +450,7 @@ public:
size_t get_inputs_num() const override;
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -521,9 +473,7 @@ public:
size_t get_inputs_num() const override;
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -541,9 +491,7 @@ public:
size_t get_inputs_num() const override;
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -557,9 +505,7 @@ public:
size_t get_inputs_num() const override;
private:
void emit_impl(const std::vector<size_t>& in, const std::vector<size_t>& out,
const std::vector<size_t>& pool, const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t>& in, const std::vector<size_t>& out) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -578,10 +524,7 @@ public:
private:
void emit_impl(
const std::vector<size_t> &in_vec_idxs,
const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs,
const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -600,9 +543,7 @@ public:
size_t get_inputs_num() const override;
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -628,9 +569,7 @@ protected:
void register_table_entries() override;
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emitContext) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -655,9 +594,7 @@ protected:
void register_table_entries() override;
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emitContext) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -684,9 +621,7 @@ protected:
void register_table_entries() override;
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emitContext) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;
@@ -703,9 +638,7 @@ public:
size_t aux_vecs_count() const override;
private:
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in_vec_idxs, const std::vector<size_t> &out_vec_idxs) const;

12
src/plugins/intel_cpu/src/emitters/jit_emitter.cpp
View File

@@ -208,17 +208,7 @@ void jit_emitter::emit_code(const std::vector<size_t> &in_idxs, const std::vecto
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs) const {
emitter_preamble(in_idxs, out_idxs, pool_vec_idxs, pool_gpr_idxs);
emit_impl(in_idxs, out_idxs, pool_vec_idxs, pool_gpr_idxs, nullptr);
emitter_postamble();
}
void jit_emitter::emit_code(const std::vector<size_t> &in_idxs, const std::vector<size_t> &out_idxs,
const std::shared_ptr<const emitter_context> &emit_context,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs) {
emitter_preamble(in_idxs, out_idxs, pool_vec_idxs, pool_gpr_idxs);
emit_impl(in_idxs, out_idxs, pool_vec_idxs, pool_gpr_idxs, emit_context.get());
emit_impl(in_idxs, out_idxs);
emitter_postamble();
}

11
src/plugins/intel_cpu/src/emitters/jit_emitter.hpp
View File

@@ -28,10 +28,6 @@ struct emitter_params {
virtual size_t hash() const = 0;
};
struct emitter_context {
virtual ~emitter_context() = default;
};
class jit_emitter : public ngraph::snippets::Emitter {
public:
jit_emitter(dnnl::impl::cpu::x64::jit_generator* host, dnnl::impl::cpu::x64::cpu_isa_t host_isa,
@@ -50,9 +46,6 @@ public:
const std::vector<size_t> &pool_vec_idxs = {}, const std::vector<size_t> &pool_gpr_idxs = {}) const override;
void emit_data() const override;
virtual void emit_code(const std::vector<size_t> &in_idxs, const std::vector<size_t> &out_idxs,
const std::shared_ptr<const emitter_context> &emit_context,
const std::vector<size_t> &pool_vec_idxs = {}, const std::vector<size_t> &pool_gpr_idxs = {});
virtual size_t get_inputs_num() const = 0;
virtual size_t aux_vecs_count() const;
emitter_in_out_map get_in_out_type() const;
@@ -101,9 +94,7 @@ protected:
_cmp_gt_os = dnnl::impl::cpu::x64::jit_generator::_cmp_nle_us,
};
virtual void emit_impl(const std::vector<size_t> &in_idxs, const std::vector<size_t> &out_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const = 0;
virtual void emit_impl(const std::vector<size_t> &in_idxs, const std::vector<size_t> &out_idxs) const = 0;
virtual void emitter_preamble(const std::vector<size_t> &in_idxs, const std::vector<size_t> &out_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs) const;

8
src/plugins/intel_cpu/src/emitters/jit_load_store_emitters.cpp
View File

@@ -77,9 +77,7 @@ size_t jit_load_emitter::aux_gprs_count() const {
return count;
}
void jit_load_emitter::emit_impl(const std::vector<size_t> &in_idxs, const std::vector<size_t> &out_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_load_emitter::emit_impl(const std::vector<size_t> &in_idxs, const std::vector<size_t> &out_idxs) const {
const int offset = in_idxs.size() == 2 ? in_idxs[1] : 0;
if (host_isa_ == cpu::x64::sse41) {
emit_isa<cpu::x64::sse41>(Reg64(in_idxs[0]), static_cast<int>(out_idxs[0]), offset);
@@ -616,9 +614,7 @@ void jit_store_emitter::emit_data() const {
uni_vcvtneps2bf16_->emit_data();
}
void jit_store_emitter::emit_impl(const std::vector<size_t> &in_idxs, const std::vector<size_t> &out_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const {
void jit_store_emitter::emit_impl(const std::vector<size_t> &in_idxs, const std::vector<size_t> &out_idxs) const {
const int offset = in_idxs.size() == 2 ? in_idxs[1] : 0;
if (host_isa_ == cpu::x64::sse41) {
emit_isa<cpu::x64::sse41>(static_cast<int>(in_idxs[0]), Reg64(out_idxs[0]), offset);

8
src/plugins/intel_cpu/src/emitters/jit_load_store_emitters.hpp
View File

@@ -68,9 +68,7 @@ public:
* \|/
* dst_prc
*/
void emit_impl(const std::vector<size_t> &in_idxs, const std::vector<size_t> &out_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_idxs, const std::vector<size_t> &out_idxs) const override;
size_t get_inputs_num() const override;
@@ -125,9 +123,7 @@ public:
* dst_prc
* note: FP32/I32-->BF16(x*) is supported only on at least avx512-core plateform
*/
void emit_impl(const std::vector<size_t> &in_idxs, const std::vector<size_t> &out_idxs,
const std::vector<size_t> &pool_vec_idxs, const std::vector<size_t> &pool_gpr_idxs,
const emitter_context *emit_context) const override;
void emit_impl(const std::vector<size_t> &in_idxs, const std::vector<size_t> &out_idxs) const override;
size_t get_inputs_num() const override;

111
src/plugins/intel_cpu/src/emitters/jit_snippets_emitters.cpp
View File

@@ -191,17 +191,13 @@ KernelEmitter::KernelEmitter(dnnl::impl::cpu::x64::jit_generator* h, dnnl::impl:
}
void KernelEmitter::emit_code(const std::vector<size_t> &in,
const std::vector<size_t> &out,
const std::vector<size_t> &pool,
const std::vector<size_t> &gpr) const {
validate_arguments(in, out, pool, gpr);
emit_impl(in, out, pool, gpr, nullptr);
const std::vector<size_t> &out) const {
validate_arguments(in, out);
emit_impl(in, out);
}
void KernelEmitter::validate_arguments(const std::vector<size_t> &in,
const std::vector<size_t> &out,
const std::vector<size_t> &pool,
const std::vector<size_t> &gpr) const {
const std::vector<size_t> &out) const {
if (!in.empty())
IE_THROW() << "KernelEmitter got invalid number of inputs. Expected 0, got " << in.size();
if (!out.empty())
@@ -301,10 +297,7 @@ void KernelEmitter::init_data_pointers(size_t num_inputs, size_t num_params, boo
}
}
void KernelEmitter::emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& vec_pool,
const std::vector<size_t>& gpr_pool,
const ov::intel_cpu::emitter_context *emit_context) const {
const std::vector<size_t>& out) const {
h->preamble();
Reg64 reg_indexes = Reg64(static_cast<int>(reg_indexes_idx));
@@ -342,17 +335,13 @@ LoopBeginEmitter::LoopBeginEmitter(dnnl::impl::cpu::x64::jit_generator* h, dnnl:
}
void LoopBeginEmitter::emit_code(const std::vector<size_t> &in,
const std::vector<size_t> &out,
const std::vector<size_t> &pool,
const std::vector<size_t> &gpr) const {
validate_arguments(in, out, pool, gpr);
emit_impl(in, out, pool, gpr, nullptr);
const std::vector<size_t> &out) const {
validate_arguments(in, out);
emit_impl(in, out);
}
void LoopBeginEmitter::validate_arguments(const std::vector<size_t> &in,
const std::vector<size_t> &out,
const std::vector<size_t> &pool,
const std::vector<size_t> &gpr) const {
const std::vector<size_t> &out) const {
if (in.size() != num_inputs)
IE_THROW() << "Invalid inputs size: expected " << num_inputs << " got " << in.size();
if (out.size() != num_inputs + 1)
@@ -360,10 +349,7 @@ void LoopBeginEmitter::validate_arguments(const std::vector<size_t> &in,
}
void LoopBeginEmitter::emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const {
const std::vector<size_t>& out) const {
// todo: In dynamic case we will also need to set broadcasting info here
Reg64 reg_work_amount = Reg64(out.back());
Label for_body;
@@ -403,18 +389,14 @@ LoopEndEmitter::LoopEndEmitter(dnnl::impl::cpu::x64::jit_generator* h, dnnl::imp
}
void LoopEndEmitter::emit_code(const std::vector<size_t> &in,
const std::vector<size_t> &out,
const std::vector<size_t> &pool,
const std::vector<size_t> &gpr) const {
validate_arguments(in, out, pool, gpr);
emit_impl(in, out, pool, gpr, nullptr);
const std::vector<size_t> &out) const {
validate_arguments(in, out);
emit_impl(in, out);
}
void LoopEndEmitter::validate_arguments(const std::vector<size_t> &in,
const std::vector<size_t> &out,
const std::vector<size_t> &pool,
const std::vector<size_t> &gpr) const {
const std::vector<size_t> &out) const {
if (loop_begin->input_regs.size() != num_inputs)
IE_THROW() << "Invalid loop_begin->input_regs size: expected " << num_inputs << " got " << loop_begin->input_regs.size();
if (out.size() != num_outputs)
@@ -429,10 +411,7 @@ void LoopEndEmitter::validate_arguments(const std::vector<size_t> &in,
}
void LoopEndEmitter::emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const {
const std::vector<size_t>& out) const {
std::vector<size_t> data_ptr_reg_idxs(loop_begin->input_regs);
data_ptr_reg_idxs.reserve(num_inputs + num_outputs);
std::copy(out.begin(), out.end(), std::back_inserter(data_ptr_reg_idxs));
@@ -464,10 +443,7 @@ BroadcastMoveEmitter::BroadcastMoveEmitter(dnnl::impl::cpu::x64::jit_generator*
}
void BroadcastMoveEmitter::emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const {
const std::vector<size_t>& out) const {
if (host_isa_ == dnnl::impl::cpu::x64::sse41) {
emit_isa<dnnl::impl::cpu::x64::sse41>(in, out);
} else if (host_isa_ == dnnl::impl::cpu::x64::avx2) {
@@ -502,10 +478,7 @@ ScalarEmitter::ScalarEmitter(dnnl::impl::cpu::x64::jit_generator* h, dnnl::impl:
}
void ScalarEmitter::emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const {
const std::vector<size_t>& out) const {
if (host_isa_ == dnnl::impl::cpu::x64::sse41) {
emit_isa<dnnl::impl::cpu::x64::sse41>(in, out);
} else if (host_isa_ == dnnl::impl::cpu::x64::avx2) {
@@ -545,10 +518,7 @@ StoreEmitter::StoreEmitter(dnnl::impl::cpu::x64::jit_generator* h, dnnl::impl::c
}
void StoreEmitter::emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const {
const std::vector<size_t>& out) const {
if (host_isa_ == dnnl::impl::cpu::x64::sse41) {
emit_isa<dnnl::impl::cpu::x64::sse41>(in, out);
} else if (host_isa_ == dnnl::impl::cpu::x64::avx2) {
@@ -587,10 +557,7 @@ LoadEmitter::LoadEmitter(dnnl::impl::cpu::x64::jit_generator* h, dnnl::impl::cpu
}
void LoadEmitter::emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const {
const std::vector<size_t>& out) const {
if (host_isa_ == dnnl::impl::cpu::x64::sse41) {
emit_isa<dnnl::impl::cpu::x64::sse41>(in, out);
} else if (host_isa_ == dnnl::impl::cpu::x64::avx2) {
@@ -627,10 +594,7 @@ BroadcastLoadEmitter::BroadcastLoadEmitter(dnnl::impl::cpu::x64::jit_generator*
}
void BroadcastLoadEmitter::emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const {
const std::vector<size_t>& out) const {
if (host_isa_ == dnnl::impl::cpu::x64::sse41) {
emit_isa<dnnl::impl::cpu::x64::sse41>(in, out);
} else if (host_isa_ == dnnl::impl::cpu::x64::avx2) {
@@ -669,10 +633,7 @@ LoadConvertEmitter::LoadConvertEmitter(dnnl::impl::cpu::x64::jit_generator* h, d
}
void LoadConvertEmitter::emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const {
const std::vector<size_t>& out) const {
if (host_isa_ == dnnl::impl::cpu::x64::sse41) {
emit_isa<dnnl::impl::cpu::x64::sse41>(in, out);
} else if (host_isa_ == dnnl::impl::cpu::x64::avx2) {
@@ -710,10 +671,7 @@ StoreConvertEmitter::StoreConvertEmitter(dnnl::impl::cpu::x64::jit_generator* h,
}
void StoreConvertEmitter::emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const {
const std::vector<size_t>& out) const {
if (host_isa_ == dnnl::impl::cpu::x64::sse41) {
emit_isa<dnnl::impl::cpu::x64::sse41>(in, out);
} else if (host_isa_ == dnnl::impl::cpu::x64::avx2) {
@@ -858,10 +816,7 @@ void BrgemmEmitter::initBrgemm(brgemmCtx& ctx, std::unique_ptr<brgemm_kernel_t>&
}
void BrgemmEmitter::emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const {
const std::vector<size_t>& out) const {
if (host_isa_ == cpu::x64::sse41 || host_isa_ == cpu::x64::avx2) {
IE_THROW() << "BrgemmEmitter requires at least avx512_core instruction set";
} else if (host_isa_ == cpu::x64::avx512_core) {
@@ -1024,10 +979,7 @@ HorizonMaxEmitter::HorizonMaxEmitter(dnnl::impl::cpu::x64::jit_generator* h, dnn
jit_emitter(h, isa, n, Precision::FP32, emitter_in_out_map::vec_to_vec) {}
void HorizonMaxEmitter::emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const {
const std::vector<size_t>& out) const {
if (host_isa_ == dnnl::impl::cpu::x64::sse41) {
emit_isa<dnnl::impl::cpu::x64::sse41>(in, out);
} else if (host_isa_ == dnnl::impl::cpu::x64::avx2) {
@@ -1069,10 +1021,7 @@ HorizonSumEmitter::HorizonSumEmitter(dnnl::impl::cpu::x64::jit_generator* h, dnn
jit_emitter(h, isa, n, Precision::FP32, emitter_in_out_map::vec_to_vec) {}
void HorizonSumEmitter::emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const {
const std::vector<size_t>& out) const {
if (host_isa_ == dnnl::impl::cpu::x64::sse41) {
emit_isa<dnnl::impl::cpu::x64::sse41>(in, out);
} else if (host_isa_ == dnnl::impl::cpu::x64::avx2) {
@@ -1112,10 +1061,7 @@ VectorBufferEmitter::VectorBufferEmitter(dnnl::impl::cpu::x64::jit_generator* h,
jit_emitter(h, isa, n, Precision::FP32, emitter_in_out_map::vec_to_vec) {}
void VectorBufferEmitter::emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const {
const std::vector<size_t>& out) const {
if (host_isa_ == dnnl::impl::cpu::x64::sse41) {
emit_isa<dnnl::impl::cpu::x64::sse41>(in, out);
} else if (host_isa_ == dnnl::impl::cpu::x64::avx2) {
@@ -1154,10 +1100,7 @@ size_t FillEmitter::aux_gprs_count() const {
}
void FillEmitter::emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const {
const std::vector<size_t>& out) const {
if (host_isa_ == dnnl::impl::cpu::x64::sse41) {
emit_isa<dnnl::impl::cpu::x64::sse41>(in, out);
} else if (host_isa_ == dnnl::impl::cpu::x64::avx2) {

104
src/plugins/intel_cpu/src/emitters/jit_snippets_emitters.hpp
View File

@@ -80,20 +80,13 @@ public:
size_t get_inputs_num() const override {return 0;}
void emit_code(const std::vector<size_t> &in,
const std::vector<size_t> &out,
const std::vector<size_t> &pool,
const std::vector<size_t> &gpr) const override;
const std::vector<size_t> &out) const;
private:
void validate_arguments(const std::vector<size_t> &in,
const std::vector<size_t> &out,
const std::vector<size_t> &pool,
const std::vector<size_t> &gpr) const override;
const std::vector<size_t> &out) const;
void emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const override;
const std::vector<size_t>& out) const override;
void init_data_pointers(size_t, size_t, bool, const Reg64&, const Reg64&, const std::vector<Reg64>&) const;
jit_snippets_compile_args jcp;
@@ -119,22 +112,15 @@ class LoopBeginEmitter : public jit_emitter {
public:
LoopBeginEmitter(dnnl::impl::cpu::x64::jit_generator* h, dnnl::impl::cpu::x64::cpu_isa_t isa, const std::shared_ptr<ov::Node>& n);
void emit_code(const std::vector<size_t> &in,
const std::vector<size_t> &out,
const std::vector<size_t> &pool,
const std::vector<size_t> &gpr) const override;
const std::vector<size_t> &out) const;
// todo: it is purely virtual in the base class, but do we need it?
size_t get_inputs_num() const override {return 0;}
private:
void validate_arguments(const std::vector<size_t> &in,
const std::vector<size_t> &out,
const std::vector<size_t> &pool,
const std::vector<size_t> &gpr) const override;
const std::vector<size_t> &out) const;
void emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const override;
const std::vector<size_t>& out) const override;
std::shared_ptr<ngraph::snippets::op::LoopBegin> loop_begin;
size_t num_inputs = 0;
@@ -146,23 +132,16 @@ class LoopEndEmitter : public jit_emitter {
public:
LoopEndEmitter(dnnl::impl::cpu::x64::jit_generator* h, dnnl::impl::cpu::x64::cpu_isa_t isa, const std::shared_ptr<ov::Node>& n);
void emit_code(const std::vector<size_t> &in,
const std::vector<size_t> &out,
const std::vector<size_t> &pool,
const std::vector<size_t> &gpr) const override;
const std::vector<size_t> &out) const;
// todo: it is purely virtual in the base class, but do we need it?
size_t get_inputs_num() const override {return 0;}
private:
void validate_arguments(const std::vector<size_t> &in,
const std::vector<size_t> &out,
const std::vector<size_t> &pool,
const std::vector<size_t> &gpr) const override;
const std::vector<size_t> &out) const;
void emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const override;
const std::vector<size_t>& out) const override;
std::shared_ptr<ngraph::snippets::op::LoopBegin> loop_begin;
std::shared_ptr<ngraph::snippets::op::LoopEnd> loop_end;
@@ -189,10 +168,7 @@ public:
private:
void emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const override {
const std::vector<size_t>& out) const override {
}
};
@@ -204,10 +180,7 @@ public:
private:
void emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const override;
const std::vector<size_t>& out) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in, const std::vector<size_t> &out) const;
@@ -227,10 +200,7 @@ protected:
private:
void emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const override;
const std::vector<size_t>& out) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in, const std::vector<size_t> &out) const;
@@ -268,10 +238,7 @@ public:
private:
void emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const override;
const std::vector<size_t>& out) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in, const std::vector<size_t> &out) const;
@@ -289,10 +256,7 @@ public:
private:
void emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const override;
const std::vector<size_t>& out) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in, const std::vector<size_t> &out) const;
@@ -310,10 +274,7 @@ public:
private:
void emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const override;
const std::vector<size_t>& out) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in, const std::vector<size_t> &out) const;
@@ -327,10 +288,7 @@ public:
private:
void emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const override;
const std::vector<size_t>& out) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in, const std::vector<size_t> &out) const;
@@ -348,10 +306,7 @@ public:
private:
void emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const override;
const std::vector<size_t>& out) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in, const std::vector<size_t> &out) const;
@@ -369,10 +324,7 @@ public:
private:
void emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const override;
const std::vector<size_t>& out) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in, const std::vector<size_t> &out) const;
@@ -422,10 +374,7 @@ protected:
private:
void emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const override;
const std::vector<size_t>& out) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in, const std::vector<size_t> &out) const;
@@ -443,10 +392,7 @@ protected:
private:
void emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const override;
const std::vector<size_t>& out) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in, const std::vector<size_t> &out) const;
@@ -460,10 +406,7 @@ public:
private:
void emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const override;
const std::vector<size_t>& out) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in, const std::vector<size_t> &out) const;
@@ -480,10 +423,7 @@ protected:
private:
void emit_impl(const std::vector<size_t>& in,
const std::vector<size_t>& out,
const std::vector<size_t>& pool,
const std::vector<size_t>& gpr,
const ov::intel_cpu::emitter_context *emit_context) const override;
const std::vector<size_t>& out) const override;
template <dnnl::impl::cpu::x64::cpu_isa_t isa>
void emit_isa(const std::vector<size_t> &in, const std::vector<size_t> &out) const;