Author: Phoebe Wang Date: 2023-09-01T08:24:05+02:00 New Revision: a0de0b440fa606805749dc7469a1b7dcae851eba
URL: https://github.com/llvm/llvm-project/commit/a0de0b440fa606805749dc7469a1b7dcae851eba DIFF: https://github.com/llvm/llvm-project/commit/a0de0b440fa606805749dc7469a1b7dcae851eba.diff LOG: [X86][BF16] Add test coverage for AVX-NE-CONVERT Split from D158952. (cherry picked from commit 30ec9473c6685d64d5caa17e2a6e8f4ccf275159) Added: Modified: llvm/lib/Target/X86/X86ISelLowering.cpp llvm/test/CodeGen/X86/bfloat.ll Removed: ################################################################################ diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp index ae9012055bbb93..7bcc181e2ca911 100644 --- a/llvm/lib/Target/X86/X86ISelLowering.cpp +++ b/llvm/lib/Target/X86/X86ISelLowering.cpp @@ -11360,7 +11360,8 @@ X86TargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) const { if (VT.getVectorElementType() == MVT::i1 && Subtarget.hasAVX512()) return LowerBUILD_VECTORvXi1(Op, DAG, Subtarget); - if (VT.getVectorElementType() == MVT::bf16 && Subtarget.hasBF16()) + if (VT.getVectorElementType() == MVT::bf16 && + (Subtarget.hasAVXNECONVERT() || Subtarget.hasBF16())) return LowerBUILD_VECTORvXbf16(Op, DAG, Subtarget); if (SDValue VectorConstant = materializeVectorConstant(Op, DAG, Subtarget)) diff --git a/llvm/test/CodeGen/X86/bfloat.ll b/llvm/test/CodeGen/X86/bfloat.ll index dff4864537bfd2..6798adaf1e5f2e 100644 --- a/llvm/test/CodeGen/X86/bfloat.ll +++ b/llvm/test/CodeGen/X86/bfloat.ll @@ -1,7 +1,8 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py ; RUN: llc < %s -mtriple=x86_64-linux-gnu | FileCheck %s --check-prefixes=CHECK,SSE2 -; RUN: llc < %s -mtriple=x86_64-linux-gnu -mattr=avx512bf16,avx512vl | FileCheck %s --check-prefixes=CHECK,F16,BF16 -; RUN: llc < %s -mtriple=x86_64-linux-gnu -mattr=avx512bf16,avx512fp16,avx512vl | FileCheck %s --check-prefixes=CHECK,F16,FP16 +; RUN: llc < %s -mtriple=x86_64-linux-gnu -mattr=avx512bf16,avx512vl | FileCheck %s --check-prefixes=CHECK,AVX,F16,BF16 +; RUN: llc < %s -mtriple=x86_64-linux-gnu -mattr=avx512bf16,avx512fp16,avx512vl | FileCheck %s --check-prefixes=CHECK,AVX,F16,FP16 +; RUN: llc < %s -mtriple=x86_64-linux-gnu -mattr=avxneconvert | FileCheck %s --check-prefixes=CHECK,AVX,AVXNC define void @add(ptr %pa, ptr %pb, ptr %pc) nounwind { ; SSE2-LABEL: add: @@ -21,22 +22,22 @@ define void @add(ptr %pa, ptr %pb, ptr %pc) nounwind { ; SSE2-NEXT: popq %rbx ; SSE2-NEXT: retq ; -; F16-LABEL: add: -; F16: # %bb.0: -; F16-NEXT: pushq %rbx -; F16-NEXT: movq %rdx, %rbx -; F16-NEXT: movzwl (%rsi), %eax -; F16-NEXT: shll $16, %eax -; F16-NEXT: vmovd %eax, %xmm0 -; F16-NEXT: movzwl (%rdi), %eax -; F16-NEXT: shll $16, %eax -; F16-NEXT: vmovd %eax, %xmm1 -; F16-NEXT: vaddss %xmm0, %xmm1, %xmm0 -; F16-NEXT: callq __truncsfbf2@PLT -; F16-NEXT: vmovd %xmm0, %eax -; F16-NEXT: movw %ax, (%rbx) -; F16-NEXT: popq %rbx -; F16-NEXT: retq +; AVX-LABEL: add: +; AVX: # %bb.0: +; AVX-NEXT: pushq %rbx +; AVX-NEXT: movq %rdx, %rbx +; AVX-NEXT: movzwl (%rsi), %eax +; AVX-NEXT: shll $16, %eax +; AVX-NEXT: vmovd %eax, %xmm0 +; AVX-NEXT: movzwl (%rdi), %eax +; AVX-NEXT: shll $16, %eax +; AVX-NEXT: vmovd %eax, %xmm1 +; AVX-NEXT: vaddss %xmm0, %xmm1, %xmm0 +; AVX-NEXT: callq __truncsfbf2@PLT +; AVX-NEXT: vmovd %xmm0, %eax +; AVX-NEXT: movw %ax, (%rbx) +; AVX-NEXT: popq %rbx +; AVX-NEXT: retq %a = load bfloat, ptr %pa %b = load bfloat, ptr %pb %add = fadd bfloat %a, %b @@ -59,19 +60,19 @@ define bfloat @add2(bfloat %a, bfloat %b) nounwind { ; SSE2-NEXT: popq %rax ; SSE2-NEXT: retq ; -; F16-LABEL: add2: -; F16: # %bb.0: -; F16-NEXT: pushq %rax -; F16-NEXT: vmovd %xmm0, %eax -; F16-NEXT: vmovd %xmm1, %ecx -; F16-NEXT: shll $16, %ecx -; F16-NEXT: vmovd %ecx, %xmm0 -; F16-NEXT: shll $16, %eax -; F16-NEXT: vmovd %eax, %xmm1 -; F16-NEXT: vaddss %xmm0, %xmm1, %xmm0 -; F16-NEXT: callq __truncsfbf2@PLT -; F16-NEXT: popq %rax -; F16-NEXT: retq +; AVX-LABEL: add2: +; AVX: # %bb.0: +; AVX-NEXT: pushq %rax +; AVX-NEXT: vmovd %xmm0, %eax +; AVX-NEXT: vmovd %xmm1, %ecx +; AVX-NEXT: shll $16, %ecx +; AVX-NEXT: vmovd %ecx, %xmm0 +; AVX-NEXT: shll $16, %eax +; AVX-NEXT: vmovd %eax, %xmm1 +; AVX-NEXT: vaddss %xmm0, %xmm1, %xmm0 +; AVX-NEXT: callq __truncsfbf2@PLT +; AVX-NEXT: popq %rax +; AVX-NEXT: retq %add = fadd bfloat %a, %b ret bfloat %add } @@ -106,34 +107,34 @@ define void @add_double(ptr %pa, ptr %pb, ptr %pc) nounwind { ; SSE2-NEXT: popq %rbp ; SSE2-NEXT: retq ; -; F16-LABEL: add_double: -; F16: # %bb.0: -; F16-NEXT: pushq %rbp -; F16-NEXT: pushq %r14 -; F16-NEXT: pushq %rbx -; F16-NEXT: movq %rdx, %rbx -; F16-NEXT: movq %rsi, %r14 -; F16-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero -; F16-NEXT: callq __truncdfbf2@PLT -; F16-NEXT: vmovd %xmm0, %ebp -; F16-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero -; F16-NEXT: callq __truncdfbf2@PLT -; F16-NEXT: vmovd %xmm0, %eax -; F16-NEXT: shll $16, %eax -; F16-NEXT: vmovd %eax, %xmm0 -; F16-NEXT: shll $16, %ebp -; F16-NEXT: vmovd %ebp, %xmm1 -; F16-NEXT: vaddss %xmm0, %xmm1, %xmm0 -; F16-NEXT: callq __truncsfbf2@PLT -; F16-NEXT: vmovd %xmm0, %eax -; F16-NEXT: shll $16, %eax -; F16-NEXT: vmovd %eax, %xmm0 -; F16-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0 -; F16-NEXT: vmovsd %xmm0, (%rbx) -; F16-NEXT: popq %rbx -; F16-NEXT: popq %r14 -; F16-NEXT: popq %rbp -; F16-NEXT: retq +; AVX-LABEL: add_double: +; AVX: # %bb.0: +; AVX-NEXT: pushq %rbp +; AVX-NEXT: pushq %r14 +; AVX-NEXT: pushq %rbx +; AVX-NEXT: movq %rdx, %rbx +; AVX-NEXT: movq %rsi, %r14 +; AVX-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero +; AVX-NEXT: callq __truncdfbf2@PLT +; AVX-NEXT: vmovd %xmm0, %ebp +; AVX-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero +; AVX-NEXT: callq __truncdfbf2@PLT +; AVX-NEXT: vmovd %xmm0, %eax +; AVX-NEXT: shll $16, %eax +; AVX-NEXT: vmovd %eax, %xmm0 +; AVX-NEXT: shll $16, %ebp +; AVX-NEXT: vmovd %ebp, %xmm1 +; AVX-NEXT: vaddss %xmm0, %xmm1, %xmm0 +; AVX-NEXT: callq __truncsfbf2@PLT +; AVX-NEXT: vmovd %xmm0, %eax +; AVX-NEXT: shll $16, %eax +; AVX-NEXT: vmovd %eax, %xmm0 +; AVX-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0 +; AVX-NEXT: vmovsd %xmm0, (%rbx) +; AVX-NEXT: popq %rbx +; AVX-NEXT: popq %r14 +; AVX-NEXT: popq %rbp +; AVX-NEXT: retq %la = load double, ptr %pa %a = fptrunc double %la to bfloat %lb = load double, ptr %pb @@ -170,30 +171,30 @@ define double @add_double2(double %da, double %db) nounwind { ; SSE2-NEXT: popq %rbx ; SSE2-NEXT: retq ; -; F16-LABEL: add_double2: -; F16: # %bb.0: -; F16-NEXT: pushq %rbx -; F16-NEXT: subq $16, %rsp -; F16-NEXT: vmovsd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill -; F16-NEXT: callq __truncdfbf2@PLT -; F16-NEXT: vmovd %xmm0, %ebx -; F16-NEXT: vmovq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 8-byte Folded Reload -; F16-NEXT: # xmm0 = mem[0],zero -; F16-NEXT: callq __truncdfbf2@PLT -; F16-NEXT: vmovd %xmm0, %eax -; F16-NEXT: shll $16, %eax -; F16-NEXT: vmovd %eax, %xmm0 -; F16-NEXT: shll $16, %ebx -; F16-NEXT: vmovd %ebx, %xmm1 -; F16-NEXT: vaddss %xmm0, %xmm1, %xmm0 -; F16-NEXT: callq __truncsfbf2@PLT -; F16-NEXT: vmovd %xmm0, %eax -; F16-NEXT: shll $16, %eax -; F16-NEXT: vmovd %eax, %xmm0 -; F16-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0 -; F16-NEXT: addq $16, %rsp -; F16-NEXT: popq %rbx -; F16-NEXT: retq +; AVX-LABEL: add_double2: +; AVX: # %bb.0: +; AVX-NEXT: pushq %rbx +; AVX-NEXT: subq $16, %rsp +; AVX-NEXT: vmovsd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill +; AVX-NEXT: callq __truncdfbf2@PLT +; AVX-NEXT: vmovd %xmm0, %ebx +; AVX-NEXT: vmovq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 8-byte Folded Reload +; AVX-NEXT: # xmm0 = mem[0],zero +; AVX-NEXT: callq __truncdfbf2@PLT +; AVX-NEXT: vmovd %xmm0, %eax +; AVX-NEXT: shll $16, %eax +; AVX-NEXT: vmovd %eax, %xmm0 +; AVX-NEXT: shll $16, %ebx +; AVX-NEXT: vmovd %ebx, %xmm1 +; AVX-NEXT: vaddss %xmm0, %xmm1, %xmm0 +; AVX-NEXT: callq __truncsfbf2@PLT +; AVX-NEXT: vmovd %xmm0, %eax +; AVX-NEXT: shll $16, %eax +; AVX-NEXT: vmovd %eax, %xmm0 +; AVX-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0 +; AVX-NEXT: addq $16, %rsp +; AVX-NEXT: popq %rbx +; AVX-NEXT: retq %a = fptrunc double %da to bfloat %b = fptrunc double %db to bfloat %add = fadd bfloat %a, %b @@ -216,19 +217,19 @@ define void @add_constant(ptr %pa, ptr %pc) nounwind { ; SSE2-NEXT: popq %rbx ; SSE2-NEXT: retq ; -; F16-LABEL: add_constant: -; F16: # %bb.0: -; F16-NEXT: pushq %rbx -; F16-NEXT: movq %rsi, %rbx -; F16-NEXT: movzwl (%rdi), %eax -; F16-NEXT: shll $16, %eax -; F16-NEXT: vmovd %eax, %xmm0 -; F16-NEXT: vaddss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 -; F16-NEXT: callq __truncsfbf2@PLT -; F16-NEXT: vmovd %xmm0, %eax -; F16-NEXT: movw %ax, (%rbx) -; F16-NEXT: popq %rbx -; F16-NEXT: retq +; AVX-LABEL: add_constant: +; AVX: # %bb.0: +; AVX-NEXT: pushq %rbx +; AVX-NEXT: movq %rsi, %rbx +; AVX-NEXT: movzwl (%rdi), %eax +; AVX-NEXT: shll $16, %eax +; AVX-NEXT: vmovd %eax, %xmm0 +; AVX-NEXT: vaddss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 +; AVX-NEXT: callq __truncsfbf2@PLT +; AVX-NEXT: vmovd %xmm0, %eax +; AVX-NEXT: movw %ax, (%rbx) +; AVX-NEXT: popq %rbx +; AVX-NEXT: retq %a = load bfloat, ptr %pa %add = fadd bfloat %a, 1.0 store bfloat %add, ptr %pc @@ -247,16 +248,16 @@ define bfloat @add_constant2(bfloat %a) nounwind { ; SSE2-NEXT: popq %rax ; SSE2-NEXT: retq ; -; F16-LABEL: add_constant2: -; F16: # %bb.0: -; F16-NEXT: pushq %rax -; F16-NEXT: vmovd %xmm0, %eax -; F16-NEXT: shll $16, %eax -; F16-NEXT: vmovd %eax, %xmm0 -; F16-NEXT: vaddss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 -; F16-NEXT: callq __truncsfbf2@PLT -; F16-NEXT: popq %rax -; F16-NEXT: retq +; AVX-LABEL: add_constant2: +; AVX: # %bb.0: +; AVX-NEXT: pushq %rax +; AVX-NEXT: vmovd %xmm0, %eax +; AVX-NEXT: shll $16, %eax +; AVX-NEXT: vmovd %eax, %xmm0 +; AVX-NEXT: vaddss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 +; AVX-NEXT: callq __truncsfbf2@PLT +; AVX-NEXT: popq %rax +; AVX-NEXT: retq %add = fadd bfloat %a, 1.0 ret bfloat %add } @@ -656,6 +657,120 @@ define <8 x bfloat> @addv(<8 x bfloat> %a, <8 x bfloat> %b) nounwind { ; FP16-NEXT: popq %r15 ; FP16-NEXT: popq %rbp ; FP16-NEXT: retq +; +; AVXNC-LABEL: addv: +; AVXNC: # %bb.0: +; AVXNC-NEXT: pushq %rbp +; AVXNC-NEXT: pushq %r15 +; AVXNC-NEXT: pushq %r14 +; AVXNC-NEXT: pushq %r13 +; AVXNC-NEXT: pushq %r12 +; AVXNC-NEXT: pushq %rbx +; AVXNC-NEXT: subq $40, %rsp +; AVXNC-NEXT: vmovdqa %xmm1, (%rsp) # 16-byte Spill +; AVXNC-NEXT: vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill +; AVXNC-NEXT: vpextrw $7, %xmm1, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm2 +; AVXNC-NEXT: vpextrw $7, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm1 +; AVXNC-NEXT: vaddss %xmm2, %xmm1, %xmm0 +; AVXNC-NEXT: callq __truncsfbf2@PLT +; AVXNC-NEXT: vmovss %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill +; AVXNC-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload +; AVXNC-NEXT: vpextrw $6, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm0 +; AVXNC-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload +; AVXNC-NEXT: vpextrw $6, %xmm1, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm1 +; AVXNC-NEXT: vaddss %xmm0, %xmm1, %xmm0 +; AVXNC-NEXT: callq __truncsfbf2@PLT +; AVXNC-NEXT: vmovd %xmm0, %ebp +; AVXNC-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload +; AVXNC-NEXT: vpextrw $5, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm0 +; AVXNC-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload +; AVXNC-NEXT: vpextrw $5, %xmm1, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm1 +; AVXNC-NEXT: vaddss %xmm0, %xmm1, %xmm0 +; AVXNC-NEXT: callq __truncsfbf2@PLT +; AVXNC-NEXT: vmovd %xmm0, %r14d +; AVXNC-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload +; AVXNC-NEXT: vpextrw $4, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm0 +; AVXNC-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload +; AVXNC-NEXT: vpextrw $4, %xmm1, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm1 +; AVXNC-NEXT: vaddss %xmm0, %xmm1, %xmm0 +; AVXNC-NEXT: callq __truncsfbf2@PLT +; AVXNC-NEXT: vmovd %xmm0, %r15d +; AVXNC-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload +; AVXNC-NEXT: vpextrw $3, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm0 +; AVXNC-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload +; AVXNC-NEXT: vpextrw $3, %xmm1, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm1 +; AVXNC-NEXT: vaddss %xmm0, %xmm1, %xmm0 +; AVXNC-NEXT: callq __truncsfbf2@PLT +; AVXNC-NEXT: vmovd %xmm0, %r12d +; AVXNC-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload +; AVXNC-NEXT: vpextrw $2, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm0 +; AVXNC-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload +; AVXNC-NEXT: vpextrw $2, %xmm1, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm1 +; AVXNC-NEXT: vaddss %xmm0, %xmm1, %xmm0 +; AVXNC-NEXT: callq __truncsfbf2@PLT +; AVXNC-NEXT: vmovd %xmm0, %r13d +; AVXNC-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload +; AVXNC-NEXT: vpextrw $1, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm0 +; AVXNC-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload +; AVXNC-NEXT: vpextrw $1, %xmm1, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm1 +; AVXNC-NEXT: vaddss %xmm0, %xmm1, %xmm0 +; AVXNC-NEXT: callq __truncsfbf2@PLT +; AVXNC-NEXT: vmovd %xmm0, %ebx +; AVXNC-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload +; AVXNC-NEXT: vmovd %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm0 +; AVXNC-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload +; AVXNC-NEXT: vmovd %xmm1, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm1 +; AVXNC-NEXT: vaddss %xmm0, %xmm1, %xmm0 +; AVXNC-NEXT: callq __truncsfbf2@PLT +; AVXNC-NEXT: vmovd %xmm0, %eax +; AVXNC-NEXT: vmovd %eax, %xmm0 +; AVXNC-NEXT: vpinsrw $1, %ebx, %xmm0, %xmm0 +; AVXNC-NEXT: vpinsrw $2, %r13d, %xmm0, %xmm0 +; AVXNC-NEXT: vpinsrw $3, %r12d, %xmm0, %xmm0 +; AVXNC-NEXT: vpinsrw $4, %r15d, %xmm0, %xmm0 +; AVXNC-NEXT: vpinsrw $5, %r14d, %xmm0, %xmm0 +; AVXNC-NEXT: vpinsrw $6, %ebp, %xmm0, %xmm0 +; AVXNC-NEXT: vpinsrw $7, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 4-byte Folded Reload +; AVXNC-NEXT: addq $40, %rsp +; AVXNC-NEXT: popq %rbx +; AVXNC-NEXT: popq %r12 +; AVXNC-NEXT: popq %r13 +; AVXNC-NEXT: popq %r14 +; AVXNC-NEXT: popq %r15 +; AVXNC-NEXT: popq %rbp +; AVXNC-NEXT: retq %add = fadd <8 x bfloat> %a, %b ret <8 x bfloat> %add } @@ -677,6 +792,19 @@ define <2 x bfloat> @pr62997(bfloat %a, bfloat %b) { ; F16-NEXT: vmovd %ecx, %xmm0 ; F16-NEXT: vpinsrw $1, %eax, %xmm0, %xmm0 ; F16-NEXT: retq +; +; AVXNC-LABEL: pr62997: +; AVXNC: # %bb.0: +; AVXNC-NEXT: vmovd %xmm1, %eax +; AVXNC-NEXT: vmovd %xmm0, %ecx +; AVXNC-NEXT: vmovd %ecx, %xmm0 +; AVXNC-NEXT: vpinsrw $1, %eax, %xmm0, %xmm0 +; AVXNC-NEXT: vmovd %xmm0, %eax +; AVXNC-NEXT: vpinsrw $0, %eax, %xmm0, %xmm0 +; AVXNC-NEXT: shrl $16, %eax +; AVXNC-NEXT: vpinsrw $0, %eax, %xmm0, %xmm1 +; AVXNC-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3] +; AVXNC-NEXT: retq %1 = insertelement <2 x bfloat> undef, bfloat %a, i64 0 %2 = insertelement <2 x bfloat> %1, bfloat %b, i64 1 ret <2 x bfloat> %2 @@ -695,6 +823,12 @@ define <32 x bfloat> @pr63017() { ; F16: # %bb.0: ; F16-NEXT: vxorps %xmm0, %xmm0, %xmm0 ; F16-NEXT: retq +; +; AVXNC-LABEL: pr63017: +; AVXNC: # %bb.0: +; AVXNC-NEXT: vxorps %xmm0, %xmm0, %xmm0 +; AVXNC-NEXT: vxorps %xmm1, %xmm1, %xmm1 +; AVXNC-NEXT: retq ret <32 x bfloat> zeroinitializer } @@ -1270,6 +1404,256 @@ define <32 x bfloat> @pr63017_2() nounwind { ; F16-NEXT: vpbroadcastw {{.*#+}} zmm0 = [49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024] ; F16-NEXT: vmovdqu16 (%rax), %zmm0 {%k1} ; F16-NEXT: retq +; +; AVXNC-LABEL: pr63017_2: +; AVXNC: # %bb.0: +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_1 +; AVXNC-NEXT: # %bb.2: # %cond.load +; AVXNC-NEXT: vpbroadcastw {{.*#+}} ymm1 = [49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024] +; AVXNC-NEXT: vpbroadcastw {{.*#+}} xmm0 = [49024,49024,49024,49024,49024,49024,49024,49024] +; AVXNC-NEXT: vpinsrw $0, (%rax), %xmm0, %xmm0 +; AVXNC-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],mem[4,5,6,7] +; AVXNC-NEXT: jmp .LBB12_3 +; AVXNC-NEXT: .LBB12_1: +; AVXNC-NEXT: vpbroadcastw {{.*#+}} ymm0 = [49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024,49024] +; AVXNC-NEXT: vmovdqa %ymm0, %ymm1 +; AVXNC-NEXT: .LBB12_3: # %else +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_5 +; AVXNC-NEXT: # %bb.4: # %cond.load1 +; AVXNC-NEXT: vpinsrw $1, (%rax), %xmm0, %xmm2 +; AVXNC-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7] +; AVXNC-NEXT: .LBB12_5: # %else2 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_7 +; AVXNC-NEXT: # %bb.6: # %cond.load4 +; AVXNC-NEXT: vpinsrw $2, (%rax), %xmm0, %xmm2 +; AVXNC-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7] +; AVXNC-NEXT: .LBB12_7: # %else5 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_9 +; AVXNC-NEXT: # %bb.8: # %cond.load7 +; AVXNC-NEXT: vpinsrw $3, (%rax), %xmm0, %xmm2 +; AVXNC-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7] +; AVXNC-NEXT: .LBB12_9: # %else8 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_11 +; AVXNC-NEXT: # %bb.10: # %cond.load10 +; AVXNC-NEXT: vpinsrw $4, (%rax), %xmm0, %xmm2 +; AVXNC-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7] +; AVXNC-NEXT: .LBB12_11: # %else11 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_13 +; AVXNC-NEXT: # %bb.12: # %cond.load13 +; AVXNC-NEXT: vpinsrw $5, (%rax), %xmm0, %xmm2 +; AVXNC-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7] +; AVXNC-NEXT: .LBB12_13: # %else14 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_15 +; AVXNC-NEXT: # %bb.14: # %cond.load16 +; AVXNC-NEXT: vpinsrw $6, (%rax), %xmm0, %xmm2 +; AVXNC-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7] +; AVXNC-NEXT: .LBB12_15: # %else17 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_17 +; AVXNC-NEXT: # %bb.16: # %cond.load19 +; AVXNC-NEXT: vpinsrw $7, (%rax), %xmm0, %xmm2 +; AVXNC-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7] +; AVXNC-NEXT: .LBB12_17: # %else20 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_19 +; AVXNC-NEXT: # %bb.18: # %cond.load22 +; AVXNC-NEXT: vpbroadcastw (%rax), %ymm2 +; AVXNC-NEXT: vpblendw {{.*#+}} ymm2 = ymm2[0],ymm0[1,2,3,4,5,6,7],ymm2[8],ymm0[9,10,11,12,13,14,15] +; AVXNC-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7] +; AVXNC-NEXT: .LBB12_19: # %else23 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_21 +; AVXNC-NEXT: # %bb.20: # %cond.load25 +; AVXNC-NEXT: vpbroadcastw (%rax), %ymm2 +; AVXNC-NEXT: vpblendw {{.*#+}} ymm2 = ymm0[0],ymm2[1],ymm0[2,3,4,5,6,7,8],ymm2[9],ymm0[10,11,12,13,14,15] +; AVXNC-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7] +; AVXNC-NEXT: .LBB12_21: # %else26 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_23 +; AVXNC-NEXT: # %bb.22: # %cond.load28 +; AVXNC-NEXT: vpbroadcastw (%rax), %ymm2 +; AVXNC-NEXT: vpblendw {{.*#+}} ymm2 = ymm0[0,1],ymm2[2],ymm0[3,4,5,6,7,8,9],ymm2[10],ymm0[11,12,13,14,15] +; AVXNC-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7] +; AVXNC-NEXT: .LBB12_23: # %else29 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_25 +; AVXNC-NEXT: # %bb.24: # %cond.load31 +; AVXNC-NEXT: vpbroadcastw (%rax), %ymm2 +; AVXNC-NEXT: vpblendw {{.*#+}} ymm2 = ymm0[0,1,2],ymm2[3],ymm0[4,5,6,7,8,9,10],ymm2[11],ymm0[12,13,14,15] +; AVXNC-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7] +; AVXNC-NEXT: .LBB12_25: # %else32 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_27 +; AVXNC-NEXT: # %bb.26: # %cond.load34 +; AVXNC-NEXT: vpbroadcastw (%rax), %ymm2 +; AVXNC-NEXT: vpblendw {{.*#+}} ymm2 = ymm0[0,1,2,3],ymm2[4],ymm0[5,6,7,8,9,10,11],ymm2[12],ymm0[13,14,15] +; AVXNC-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7] +; AVXNC-NEXT: .LBB12_27: # %else35 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_29 +; AVXNC-NEXT: # %bb.28: # %cond.load37 +; AVXNC-NEXT: vpbroadcastw (%rax), %ymm2 +; AVXNC-NEXT: vpblendw {{.*#+}} ymm2 = ymm0[0,1,2,3,4],ymm2[5],ymm0[6,7,8,9,10,11,12],ymm2[13],ymm0[14,15] +; AVXNC-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7] +; AVXNC-NEXT: .LBB12_29: # %else38 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_31 +; AVXNC-NEXT: # %bb.30: # %cond.load40 +; AVXNC-NEXT: vpbroadcastw (%rax), %ymm2 +; AVXNC-NEXT: vpblendw {{.*#+}} ymm2 = ymm0[0,1,2,3,4,5],ymm2[6],ymm0[7,8,9,10,11,12,13],ymm2[14],ymm0[15] +; AVXNC-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7] +; AVXNC-NEXT: .LBB12_31: # %else41 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_33 +; AVXNC-NEXT: # %bb.32: # %cond.load43 +; AVXNC-NEXT: vpbroadcastw (%rax), %ymm2 +; AVXNC-NEXT: vpblendw {{.*#+}} ymm2 = ymm0[0,1,2,3,4,5,6],ymm2[7],ymm0[8,9,10,11,12,13,14],ymm2[15] +; AVXNC-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7] +; AVXNC-NEXT: .LBB12_33: # %else44 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_35 +; AVXNC-NEXT: # %bb.34: # %cond.load46 +; AVXNC-NEXT: vpinsrw $0, (%rax), %xmm1, %xmm2 +; AVXNC-NEXT: vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7] +; AVXNC-NEXT: .LBB12_35: # %else47 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_37 +; AVXNC-NEXT: # %bb.36: # %cond.load49 +; AVXNC-NEXT: vpinsrw $1, (%rax), %xmm1, %xmm2 +; AVXNC-NEXT: vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7] +; AVXNC-NEXT: .LBB12_37: # %else50 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_39 +; AVXNC-NEXT: # %bb.38: # %cond.load52 +; AVXNC-NEXT: vpinsrw $2, (%rax), %xmm1, %xmm2 +; AVXNC-NEXT: vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7] +; AVXNC-NEXT: .LBB12_39: # %else53 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_41 +; AVXNC-NEXT: # %bb.40: # %cond.load55 +; AVXNC-NEXT: vpinsrw $3, (%rax), %xmm1, %xmm2 +; AVXNC-NEXT: vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7] +; AVXNC-NEXT: .LBB12_41: # %else56 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_43 +; AVXNC-NEXT: # %bb.42: # %cond.load58 +; AVXNC-NEXT: vpinsrw $4, (%rax), %xmm1, %xmm2 +; AVXNC-NEXT: vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7] +; AVXNC-NEXT: .LBB12_43: # %else59 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_45 +; AVXNC-NEXT: # %bb.44: # %cond.load61 +; AVXNC-NEXT: vpinsrw $5, (%rax), %xmm1, %xmm2 +; AVXNC-NEXT: vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7] +; AVXNC-NEXT: .LBB12_45: # %else62 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_47 +; AVXNC-NEXT: # %bb.46: # %cond.load64 +; AVXNC-NEXT: vpinsrw $6, (%rax), %xmm1, %xmm2 +; AVXNC-NEXT: vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7] +; AVXNC-NEXT: .LBB12_47: # %else65 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_49 +; AVXNC-NEXT: # %bb.48: # %cond.load67 +; AVXNC-NEXT: vpinsrw $7, (%rax), %xmm1, %xmm2 +; AVXNC-NEXT: vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7] +; AVXNC-NEXT: .LBB12_49: # %else68 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_51 +; AVXNC-NEXT: # %bb.50: # %cond.load70 +; AVXNC-NEXT: vpbroadcastw (%rax), %ymm2 +; AVXNC-NEXT: vpblendw {{.*#+}} ymm2 = ymm2[0],ymm1[1,2,3,4,5,6,7],ymm2[8],ymm1[9,10,11,12,13,14,15] +; AVXNC-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7] +; AVXNC-NEXT: .LBB12_51: # %else71 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_53 +; AVXNC-NEXT: # %bb.52: # %cond.load73 +; AVXNC-NEXT: vpbroadcastw (%rax), %ymm2 +; AVXNC-NEXT: vpblendw {{.*#+}} ymm2 = ymm1[0],ymm2[1],ymm1[2,3,4,5,6,7,8],ymm2[9],ymm1[10,11,12,13,14,15] +; AVXNC-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7] +; AVXNC-NEXT: .LBB12_53: # %else74 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_55 +; AVXNC-NEXT: # %bb.54: # %cond.load76 +; AVXNC-NEXT: vpbroadcastw (%rax), %ymm2 +; AVXNC-NEXT: vpblendw {{.*#+}} ymm2 = ymm1[0,1],ymm2[2],ymm1[3,4,5,6,7,8,9],ymm2[10],ymm1[11,12,13,14,15] +; AVXNC-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7] +; AVXNC-NEXT: .LBB12_55: # %else77 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_57 +; AVXNC-NEXT: # %bb.56: # %cond.load79 +; AVXNC-NEXT: vpbroadcastw (%rax), %ymm2 +; AVXNC-NEXT: vpblendw {{.*#+}} ymm2 = ymm1[0,1,2],ymm2[3],ymm1[4,5,6,7,8,9,10],ymm2[11],ymm1[12,13,14,15] +; AVXNC-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7] +; AVXNC-NEXT: .LBB12_57: # %else80 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_59 +; AVXNC-NEXT: # %bb.58: # %cond.load82 +; AVXNC-NEXT: vpbroadcastw (%rax), %ymm2 +; AVXNC-NEXT: vpblendw {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm2[4],ymm1[5,6,7,8,9,10,11],ymm2[12],ymm1[13,14,15] +; AVXNC-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7] +; AVXNC-NEXT: .LBB12_59: # %else83 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_61 +; AVXNC-NEXT: # %bb.60: # %cond.load85 +; AVXNC-NEXT: vpbroadcastw (%rax), %ymm2 +; AVXNC-NEXT: vpblendw {{.*#+}} ymm2 = ymm1[0,1,2,3,4],ymm2[5],ymm1[6,7,8,9,10,11,12],ymm2[13],ymm1[14,15] +; AVXNC-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7] +; AVXNC-NEXT: .LBB12_61: # %else86 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_63 +; AVXNC-NEXT: # %bb.62: # %cond.load88 +; AVXNC-NEXT: vpbroadcastw (%rax), %ymm2 +; AVXNC-NEXT: vpblendw {{.*#+}} ymm2 = ymm1[0,1,2,3,4,5],ymm2[6],ymm1[7,8,9,10,11,12,13],ymm2[14],ymm1[15] +; AVXNC-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7] +; AVXNC-NEXT: .LBB12_63: # %else89 +; AVXNC-NEXT: xorl %eax, %eax +; AVXNC-NEXT: testb %al, %al +; AVXNC-NEXT: jne .LBB12_65 +; AVXNC-NEXT: # %bb.64: # %cond.load91 +; AVXNC-NEXT: vpbroadcastw (%rax), %ymm2 +; AVXNC-NEXT: vpblendw {{.*#+}} ymm2 = ymm1[0,1,2,3,4,5,6],ymm2[7],ymm1[8,9,10,11,12,13,14],ymm2[15] +; AVXNC-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7] +; AVXNC-NEXT: .LBB12_65: # %else92 +; AVXNC-NEXT: retq %1 = call <32 x bfloat> @llvm.masked.load.v32bf16.p0(ptr poison, i32 2, <32 x i1> poison, <32 x bfloat> <bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80, bfloat 0xRBF80>) ret <32 x bfloat> %1 } @@ -1295,6 +1679,13 @@ define <32 x bfloat> @pr62997_3(<32 x bfloat> %0, bfloat %1) { ; F16-NEXT: vpinsrw $1, %eax, %xmm0, %xmm1 ; F16-NEXT: vinserti32x4 $0, %xmm1, %zmm0, %zmm0 ; F16-NEXT: retq +; +; AVXNC-LABEL: pr62997_3: +; AVXNC: # %bb.0: +; AVXNC-NEXT: vmovd %xmm2, %eax +; AVXNC-NEXT: vpinsrw $1, %eax, %xmm0, %xmm2 +; AVXNC-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7] +; AVXNC-NEXT: retq %3 = insertelement <32 x bfloat> %0, bfloat %1, i64 1 ret <32 x bfloat> %3 } @@ -1328,6 +1719,25 @@ define <4 x float> @pr64460_1(<4 x bfloat> %a) { ; F16-NEXT: vpxor %xmm1, %xmm1, %xmm1 ; F16-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3] ; F16-NEXT: retq +; +; AVXNC-LABEL: pr64460_1: +; AVXNC: # %bb.0: +; AVXNC-NEXT: vpextrw $1, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm1 +; AVXNC-NEXT: vmovd %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm2 +; AVXNC-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[2,3] +; AVXNC-NEXT: vpextrw $2, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm2 +; AVXNC-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1],xmm2[0],xmm1[3] +; AVXNC-NEXT: vpextrw $3, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm0 +; AVXNC-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0] +; AVXNC-NEXT: retq %b = fpext <4 x bfloat> %a to <4 x float> ret <4 x float> %b } @@ -1377,6 +1787,41 @@ define <8 x float> @pr64460_2(<8 x bfloat> %a) { ; F16-NEXT: vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero ; F16-NEXT: vpslld $16, %ymm0, %ymm0 ; F16-NEXT: retq +; +; AVXNC-LABEL: pr64460_2: +; AVXNC: # %bb.0: +; AVXNC-NEXT: vpextrw $5, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm1 +; AVXNC-NEXT: vpextrw $4, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm2 +; AVXNC-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[2,3] +; AVXNC-NEXT: vpextrw $6, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm2 +; AVXNC-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1],xmm2[0],xmm1[3] +; AVXNC-NEXT: vpextrw $7, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm2 +; AVXNC-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1,2],xmm2[0] +; AVXNC-NEXT: vpextrw $1, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm2 +; AVXNC-NEXT: vmovd %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm3 +; AVXNC-NEXT: vinsertps {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[2,3] +; AVXNC-NEXT: vpextrw $2, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm3 +; AVXNC-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1],xmm3[0],xmm2[3] +; AVXNC-NEXT: vpextrw $3, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm0 +; AVXNC-NEXT: vinsertps {{.*#+}} xmm0 = xmm2[0,1,2],xmm0[0] +; AVXNC-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 +; AVXNC-NEXT: retq %b = fpext <8 x bfloat> %a to <8 x float> ret <8 x float> %b } @@ -1461,6 +1906,74 @@ define <16 x float> @pr64460_3(<16 x bfloat> %a) { ; F16-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero ; F16-NEXT: vpslld $16, %zmm0, %zmm0 ; F16-NEXT: retq +; +; AVXNC-LABEL: pr64460_3: +; AVXNC: # %bb.0: +; AVXNC-NEXT: vpextrw $5, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm1 +; AVXNC-NEXT: vpextrw $4, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm2 +; AVXNC-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[2,3] +; AVXNC-NEXT: vpextrw $6, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm2 +; AVXNC-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1],xmm2[0],xmm1[3] +; AVXNC-NEXT: vpextrw $7, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm2 +; AVXNC-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1,2],xmm2[0] +; AVXNC-NEXT: vpextrw $1, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm2 +; AVXNC-NEXT: vmovd %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm3 +; AVXNC-NEXT: vinsertps {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[2,3] +; AVXNC-NEXT: vpextrw $2, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm3 +; AVXNC-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1],xmm3[0],xmm2[3] +; AVXNC-NEXT: vpextrw $3, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm3 +; AVXNC-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],xmm3[0] +; AVXNC-NEXT: vinsertf128 $1, %xmm1, %ymm2, %ymm2 +; AVXNC-NEXT: vextracti128 $1, %ymm0, %xmm0 +; AVXNC-NEXT: vpextrw $5, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm1 +; AVXNC-NEXT: vpextrw $4, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm3 +; AVXNC-NEXT: vinsertps {{.*#+}} xmm1 = xmm3[0],xmm1[0],xmm3[2,3] +; AVXNC-NEXT: vpextrw $6, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm3 +; AVXNC-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1],xmm3[0],xmm1[3] +; AVXNC-NEXT: vpextrw $7, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm3 +; AVXNC-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1,2],xmm3[0] +; AVXNC-NEXT: vpextrw $1, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm3 +; AVXNC-NEXT: vmovd %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm4 +; AVXNC-NEXT: vinsertps {{.*#+}} xmm3 = xmm4[0],xmm3[0],xmm4[2,3] +; AVXNC-NEXT: vpextrw $2, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm4 +; AVXNC-NEXT: vinsertps {{.*#+}} xmm3 = xmm3[0,1],xmm4[0],xmm3[3] +; AVXNC-NEXT: vpextrw $3, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm0 +; AVXNC-NEXT: vinsertps {{.*#+}} xmm0 = xmm3[0,1,2],xmm0[0] +; AVXNC-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm1 +; AVXNC-NEXT: vmovaps %ymm2, %ymm0 +; AVXNC-NEXT: retq %b = fpext <16 x bfloat> %a to <16 x float> ret <16 x float> %b } @@ -1517,6 +2030,49 @@ define <8 x double> @pr64460_4(<8 x bfloat> %a) { ; F16-NEXT: vpslld $16, %ymm0, %ymm0 ; F16-NEXT: vcvtps2pd %ymm0, %zmm0 ; F16-NEXT: retq +; +; AVXNC-LABEL: pr64460_4: +; AVXNC: # %bb.0: +; AVXNC-NEXT: vpextrw $3, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm1 +; AVXNC-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1 +; AVXNC-NEXT: vpextrw $2, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm2 +; AVXNC-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2 +; AVXNC-NEXT: vmovlhps {{.*#+}} xmm1 = xmm2[0],xmm1[0] +; AVXNC-NEXT: vpextrw $1, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm2 +; AVXNC-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2 +; AVXNC-NEXT: vmovd %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm3 +; AVXNC-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3 +; AVXNC-NEXT: vmovlhps {{.*#+}} xmm2 = xmm3[0],xmm2[0] +; AVXNC-NEXT: vinsertf128 $1, %xmm1, %ymm2, %ymm2 +; AVXNC-NEXT: vpextrw $7, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm1 +; AVXNC-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1 +; AVXNC-NEXT: vpextrw $6, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm3 +; AVXNC-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3 +; AVXNC-NEXT: vmovlhps {{.*#+}} xmm1 = xmm3[0],xmm1[0] +; AVXNC-NEXT: vpextrw $5, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm3 +; AVXNC-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3 +; AVXNC-NEXT: vpextrw $4, %xmm0, %eax +; AVXNC-NEXT: shll $16, %eax +; AVXNC-NEXT: vmovd %eax, %xmm0 +; AVXNC-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0 +; AVXNC-NEXT: vmovlhps {{.*#+}} xmm0 = xmm0[0],xmm3[0] +; AVXNC-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm1 +; AVXNC-NEXT: vmovaps %ymm2, %ymm0 +; AVXNC-NEXT: retq %b = fpext <8 x bfloat> %a to <8 x double> ret <8 x double> %b } _______________________________________________ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
