Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Rename #24

Open
wants to merge 3 commits into
base: release_base_0.21
Choose a base branch
from
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion src/cpu/cpu_barrier.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -92,7 +92,7 @@ void generate(jit_generator &code, Xbyak::Reg64 reg_ctx,
code.cmp(reg_tmp, code.ptr[reg_ctx + BAR_SENSE_OFF]);
code.je(spin_label);
#ifdef DNNL_INDIRECT_JIT_AARCH64
code.CodeGeneratorAArch64::dmb(Xbyak_aarch64::ISH);
code.xa_->dmb(Xbyak_aarch64::ISH);
#endif //#ifdef DNNL_INDIRECT_JIT_AARCH64
code.CodeGenerator::L(barrier_exit_restore_label);
code.pop(reg_tmp);
Expand Down
6 changes: 3 additions & 3 deletions src/cpu/cpu_isa_traits.hpp
Original file line number Diff line number Diff line change
Expand Up @@ -109,15 +109,15 @@ template <> struct cpu_isa_traits<avx512_core_bf16>:
#ifdef __ARM_ARCH
template <>
struct cpu_isa_traits<simd> {
typedef Xbyak::Xbyak_aarch64::VReg4S Vmm;
typedef Xbyak_aarch64::VReg4S Vmm;
static constexpr int vlen_shift = 5;
static constexpr int vlen = 16;
static constexpr int n_vregs = 32;
};

template <> struct cpu_isa_traits<sve> {
typedef Xbyak::Xbyak_aarch64::ZRegS Vmm;
typedef Xbyak::Xbyak_aarch64::AdrScImm uni_ldst_addressing;
typedef Xbyak_aarch64::ZRegS Vmm;
typedef Xbyak_aarch64::AdrScImm uni_ldst_addressing;
static constexpr int vlen_shift = 5;
static constexpr int vlen = 64;
static constexpr int n_vregs = 32;
Expand Down
4 changes: 2 additions & 2 deletions src/cpu/gemm/bf16/jit_avx512_core_gemm_bf16bf16f32_kern.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -186,8 +186,8 @@ void jit_avx512_core_gemm_bf16bf16f32_kern::innerloop(int unroll_m,

mov(LoopCount_, K_);
#ifdef DNNL_INDIRECT_JIT_AARCH64
CodeGeneratorAArch64::asr(Xbyak::Xbyak_aarch64::XReg(LoopCount_.getIdx()), Xbyak::Xbyak_aarch64::XReg(LoopCount_.getIdx()), 0x3);
CodeGeneratorAArch64::cmp(Xbyak::Xbyak_aarch64::XReg(LoopCount_.getIdx()), 0);
xa_->asr(Xbyak_aarch64::XReg(LoopCount_.getIdx()), Xbyak_aarch64::XReg(LoopCount_.getIdx()), 0x3);
xa_->cmp(Xbyak_aarch64::XReg(LoopCount_.getIdx()), 0);
#else
sar(LoopCount_, 3);
#endif
Expand Down
28 changes: 14 additions & 14 deletions src/cpu/gemm/bf16/jit_avx512_core_s16_copy_an_kern.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -129,8 +129,8 @@ L(l24);
add(A, 0x60);
mov(I, M);
#ifdef DNNL_INDIRECT_JIT_AARCH64
CodeGeneratorAArch64::asr(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0x2);
CodeGeneratorAArch64::cmp(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0);
xa_->asr(Xbyak_aarch64::XReg(I.getIdx()), Xbyak_aarch64::XReg(I.getIdx()), 0x2);
xa_->cmp(Xbyak_aarch64::XReg(I.getIdx()), 0);
#else
sar(I, 0x2);
#endif
Expand Down Expand Up @@ -288,8 +288,8 @@ L(l318);
add(A, 0x40);
mov(I, M);
#ifdef DNNL_INDIRECT_JIT_AARCH64
CodeGeneratorAArch64::asr(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0x2);
CodeGeneratorAArch64::cmp(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0);
xa_->asr(Xbyak_aarch64::XReg(I.getIdx()), Xbyak_aarch64::XReg(I.getIdx()), 0x2);
xa_->cmp(Xbyak_aarch64::XReg(I.getIdx()), 0);
#else
sar(I, 0x2);
#endif
Expand Down Expand Up @@ -410,8 +410,8 @@ L(l52c);
add(A, 0x20);
mov(I, M);
#ifdef DNNL_INDIRECT_JIT_AARCH64
CodeGeneratorAArch64::asr(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0x2);
CodeGeneratorAArch64::cmp(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0);
xa_->asr(Xbyak_aarch64::XReg(I.getIdx()), Xbyak_aarch64::XReg(I.getIdx()), 0x2);
xa_->cmp(Xbyak_aarch64::XReg(I.getIdx()), 0);
#else
sar(I, 0x2);
#endif
Expand Down Expand Up @@ -495,8 +495,8 @@ L(l674);
add(A, 0x10);
mov(I, M);
#ifdef DNNL_INDIRECT_JIT_AARCH64
CodeGeneratorAArch64::asr(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0x3);
CodeGeneratorAArch64::cmp(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0);
xa_->asr(Xbyak_aarch64::XReg(I.getIdx()), Xbyak_aarch64::XReg(I.getIdx()), 0x3);
xa_->cmp(Xbyak_aarch64::XReg(I.getIdx()), 0);
#else
sar(I, 0x3);
#endif
Expand Down Expand Up @@ -601,8 +601,8 @@ L(l7f8);
add(A, 0x8);
mov(I, M);
#ifdef DNNL_INDIRECT_JIT_AARCH64
CodeGeneratorAArch64::asr(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0x3);
CodeGeneratorAArch64::cmp(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0);
xa_->asr(Xbyak_aarch64::XReg(I.getIdx()), Xbyak_aarch64::XReg(I.getIdx()), 0x3);
xa_->cmp(Xbyak_aarch64::XReg(I.getIdx()), 0);
#else
sar(I, 0x3);
#endif
Expand Down Expand Up @@ -693,8 +693,8 @@ L(l930);
add(A, 0x4);
mov(I, M);
#ifdef DNNL_INDIRECT_JIT_AARCH64
CodeGeneratorAArch64::asr(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0x3);
CodeGeneratorAArch64::cmp(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0);
xa_->asr(Xbyak_aarch64::XReg(I.getIdx()), Xbyak_aarch64::XReg(I.getIdx()), 0x3);
xa_->cmp(Xbyak_aarch64::XReg(I.getIdx()), 0);
#else
sar(I, 0x3);
#endif
Expand Down Expand Up @@ -785,8 +785,8 @@ L(la5c);
add(A, 0x2);
mov(LDA3, M);
#ifdef DNNL_INDIRECT_JIT_AARCH64
CodeGeneratorAArch64::asr(Xbyak::Xbyak_aarch64::XReg(LDA3.getIdx()), Xbyak::Xbyak_aarch64::XReg(LDA3.getIdx()), 0x3);
CodeGeneratorAArch64::cmp(Xbyak::Xbyak_aarch64::XReg(LDA3.getIdx()), 0);
xa_->asr(Xbyak_aarch64::XReg(LDA3.getIdx()), Xbyak_aarch64::XReg(LDA3.getIdx()), 0x3);
xa_->cmp(Xbyak_aarch64::XReg(LDA3.getIdx()), 0);
#else
sar(LDA3, 0x3);
#endif
Expand Down
28 changes: 14 additions & 14 deletions src/cpu/gemm/bf16/jit_avx512_core_s16_copy_at_kern.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -135,8 +135,8 @@ L(l24);
add(A, I);
mov(I, M);
#ifdef DNNL_INDIRECT_JIT_AARCH64
CodeGeneratorAArch64::asr(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0x3);
CodeGeneratorAArch64::cmp(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0);
xa_->asr(Xbyak_aarch64::XReg(I.getIdx()), Xbyak_aarch64::XReg(I.getIdx()), 0x3);
xa_->cmp(Xbyak_aarch64::XReg(I.getIdx()), 0);
#else
sar(I, 0x3);
#endif
Expand Down Expand Up @@ -697,8 +697,8 @@ L(lbc0);
add(A, I);
mov(I, M);
#ifdef DNNL_INDIRECT_JIT_AARCH64
CodeGeneratorAArch64::asr(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0x3);
CodeGeneratorAArch64::cmp(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0);
xa_->asr(Xbyak_aarch64::XReg(I.getIdx()), Xbyak_aarch64::XReg(I.getIdx()), 0x3);
xa_->cmp(Xbyak_aarch64::XReg(I.getIdx()), 0);
#else
sar(I, 0x3);
#endif
Expand Down Expand Up @@ -1088,8 +1088,8 @@ L(l1398);
add(A, I);
mov(I, M);
#ifdef DNNL_INDIRECT_JIT_AARCH64
CodeGeneratorAArch64::asr(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0x3);
CodeGeneratorAArch64::cmp(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0);
xa_->asr(Xbyak_aarch64::XReg(I.getIdx()), Xbyak_aarch64::XReg(I.getIdx()), 0x3);
xa_->cmp(Xbyak_aarch64::XReg(I.getIdx()), 0);
#else
sar(I, 0x3);
#endif
Expand Down Expand Up @@ -1310,8 +1310,8 @@ L(l17c0);
mov(A, I);
mov(I, M);
#ifdef DNNL_INDIRECT_JIT_AARCH64
CodeGeneratorAArch64::asr(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0x3);
CodeGeneratorAArch64::cmp(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0);
xa_->asr(Xbyak_aarch64::XReg(I.getIdx()), Xbyak_aarch64::XReg(I.getIdx()), 0x3);
xa_->cmp(Xbyak_aarch64::XReg(I.getIdx()), 0);
#else
sar(I, 0x3);
#endif
Expand Down Expand Up @@ -1443,8 +1443,8 @@ L(l19f4);
mov(A, I);
mov(I, M);
#ifdef DNNL_INDIRECT_JIT_AARCH64
CodeGeneratorAArch64::asr(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0x3);
CodeGeneratorAArch64::cmp(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0);
xa_->asr(Xbyak_aarch64::XReg(I.getIdx()), Xbyak_aarch64::XReg(I.getIdx()), 0x3);
xa_->cmp(Xbyak_aarch64::XReg(I.getIdx()), 0);
#else
sar(I, 0x3);
#endif
Expand Down Expand Up @@ -1543,8 +1543,8 @@ L(l1b6c);
mov(A, I);
mov(I, M);
#ifdef DNNL_INDIRECT_JIT_AARCH64
CodeGeneratorAArch64::asr(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0x3);
CodeGeneratorAArch64::cmp(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0);
xa_->asr(Xbyak_aarch64::XReg(I.getIdx()), Xbyak_aarch64::XReg(I.getIdx()), 0x3);
xa_->cmp(Xbyak_aarch64::XReg(I.getIdx()), 0);
#else
sar(I, 0x3);
#endif
Expand Down Expand Up @@ -1616,8 +1616,8 @@ L(l1c50);
add(A, LDA);
mov(I, M);
#ifdef DNNL_INDIRECT_JIT_AARCH64
CodeGeneratorAArch64::asr(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0x4);
CodeGeneratorAArch64::cmp(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0);
xa_->asr(Xbyak_aarch64::XReg(I.getIdx()), Xbyak_aarch64::XReg(I.getIdx()), 0x4);
xa_->cmp(Xbyak_aarch64::XReg(I.getIdx()), 0);
#else
sar(I, 0x4);
#endif
Expand Down
16 changes: 8 additions & 8 deletions src/cpu/gemm/bf16/jit_avx512_core_s16_copy_bn_kern.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -114,8 +114,8 @@ L(l24);
mov(A, I);
mov(I, M);
#ifdef DNNL_INDIRECT_JIT_AARCH64
CodeGeneratorAArch64::asr(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0x3);
CodeGeneratorAArch64::cmp(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0);
xa_->asr(Xbyak_aarch64::XReg(I.getIdx()), Xbyak_aarch64::XReg(I.getIdx()), 0x3);
xa_->cmp(Xbyak_aarch64::XReg(I.getIdx()), 0);
#else
sar(I, 0x3);
#endif
Expand Down Expand Up @@ -247,8 +247,8 @@ L(l258);
mov(A, I);
mov(I, M);
#ifdef DNNL_INDIRECT_JIT_AARCH64
CodeGeneratorAArch64::asr(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0x3);
CodeGeneratorAArch64::cmp(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0);
xa_->asr(Xbyak_aarch64::XReg(I.getIdx()), Xbyak_aarch64::XReg(I.getIdx()), 0x3);
xa_->cmp(Xbyak_aarch64::XReg(I.getIdx()), 0);
#else
sar(I, 0x3);
#endif
Expand Down Expand Up @@ -347,8 +347,8 @@ L(l3d0);
mov(A, I);
mov(I, M);
#ifdef DNNL_INDIRECT_JIT_AARCH64
CodeGeneratorAArch64::asr(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0x3);
CodeGeneratorAArch64::cmp(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0);
xa_->asr(Xbyak_aarch64::XReg(I.getIdx()), Xbyak_aarch64::XReg(I.getIdx()), 0x3);
xa_->cmp(Xbyak_aarch64::XReg(I.getIdx()), 0);
#else
sar(I, 0x3);
#endif
Expand Down Expand Up @@ -420,8 +420,8 @@ L(l4b4);
add(A, LDA);
mov(I, M);
#ifdef DNNL_INDIRECT_JIT_AARCH64
CodeGeneratorAArch64::asr(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0x4);
CodeGeneratorAArch64::cmp(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0);
xa_->asr(Xbyak_aarch64::XReg(I.getIdx()), Xbyak_aarch64::XReg(I.getIdx()), 0x4);
xa_->cmp(Xbyak_aarch64::XReg(I.getIdx()), 0);
#else
sar(I, 0x4);
#endif
Expand Down
16 changes: 8 additions & 8 deletions src/cpu/gemm/bf16/jit_avx512_core_s16_copy_bt_kern.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -111,8 +111,8 @@ L(l24);
add(A, 0x10);
mov(I, M);
#ifdef DNNL_INDIRECT_JIT_AARCH64
CodeGeneratorAArch64::asr(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0x3);
CodeGeneratorAArch64::cmp(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0);
xa_->asr(Xbyak_aarch64::XReg(I.getIdx()), Xbyak_aarch64::XReg(I.getIdx()), 0x3);
xa_->cmp(Xbyak_aarch64::XReg(I.getIdx()), 0);
#else
sar(I, 0x3);
#endif
Expand Down Expand Up @@ -217,8 +217,8 @@ L(l1a8);
add(A, 0x8);
mov(I, M);
#ifdef DNNL_INDIRECT_JIT_AARCH64
CodeGeneratorAArch64::asr(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0x3);
CodeGeneratorAArch64::cmp(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0);
xa_->asr(Xbyak_aarch64::XReg(I.getIdx()), Xbyak_aarch64::XReg(I.getIdx()), 0x3);
xa_->cmp(Xbyak_aarch64::XReg(I.getIdx()), 0);
#else
sar(I, 0x3);
#endif
Expand Down Expand Up @@ -309,8 +309,8 @@ L(l2e0);
add(A, 0x4);
mov(I, M);
#ifdef DNNL_INDIRECT_JIT_AARCH64
CodeGeneratorAArch64::asr(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0x3);
CodeGeneratorAArch64::cmp(Xbyak::Xbyak_aarch64::XReg(I.getIdx()), 0);
xa_->asr(Xbyak_aarch64::XReg(I.getIdx()), Xbyak_aarch64::XReg(I.getIdx()), 0x3);
xa_->cmp(Xbyak_aarch64::XReg(I.getIdx()), 0);
#else
sar(I, 0x3);
#endif
Expand Down Expand Up @@ -401,8 +401,8 @@ L(l40c);
add(A, 0x2);
mov(LDA3, M);
#ifdef DNNL_INDIRECT_JIT_AARCH64
CodeGeneratorAArch64::asr(Xbyak::Xbyak_aarch64::XReg(LDA3.getIdx()), Xbyak::Xbyak_aarch64::XReg(LDA3.getIdx()), 0x3);
CodeGeneratorAArch64::cmp(Xbyak::Xbyak_aarch64::XReg(LDA3.getIdx()), 0);
xa_->asr(Xbyak_aarch64::XReg(LDA3.getIdx()), Xbyak_aarch64::XReg(LDA3.getIdx()), 0x3);
xa_->cmp(Xbyak_aarch64::XReg(LDA3.getIdx()), 0);
#else
sar(LDA3, 0x3);
#endif
Expand Down
Loading