; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+amx-bf16,+amx-fp16,+amx-complex,+amx-transpose | FileCheck %s

define void @test_amx(i32 %rv32, i64 %stride, i64 %rvalue, i8* %addr1, <4 x float> %xmm) #0 {
; CHECK-LABEL: test_amx:
; CHECK:       # %bb.0:
; CHECK-NEXT:    t2rpntlvwz0 (%rcx,%rsi), %tmm0
; CHECK-NEXT:    t2rpntlvwz0t1 (%rcx,%rsi), %tmm2
; CHECK-NEXT:    t2rpntlvwz1 (%rcx,%rsi), %tmm0
; CHECK-NEXT:    t2rpntlvwz1t1 (%rcx,%rsi), %tmm2
; CHECK-NEXT:    ttransposed %tmm3, %tmm1
; CHECK-NEXT:    ttdpbf16ps %tmm3, %tmm2, %tmm1
; CHECK-NEXT:    ttdpfp16ps %tmm6, %tmm5, %tmm4
; CHECK-NEXT:    ttcmmimfp16ps %tmm3, %tmm2, %tmm1
; CHECK-NEXT:    ttcmmrlfp16ps %tmm3, %tmm2, %tmm1
; CHECK-NEXT:    tconjtcmmimfp16ps %tmm3, %tmm2, %tmm1
; CHECK-NEXT:    tconjtfp16 %tmm2, %tmm1
; CHECK-NEXT:    retq
  call void @llvm.x86.t2rpntlvwz0(i8 1, i8* %addr1, i64 %stride)
  call void @llvm.x86.t2rpntlvwz0t1(i8 2, i8* %addr1, i64 %stride)
  call void @llvm.x86.t2rpntlvwz1(i8 1, i8* %addr1, i64 %stride)
  call void @llvm.x86.t2rpntlvwz1t1(i8 2, i8* %addr1, i64 %stride)
  call void @llvm.x86.ttransposed(i8 1, i8 3)
  call void @llvm.x86.ttdpbf16ps(i8 1, i8 2, i8 3)
  call void @llvm.x86.ttdpfp16ps(i8 4, i8 5, i8 6)
  call void @llvm.x86.ttcmmimfp16ps(i8 1, i8 2, i8 3)
  call void @llvm.x86.ttcmmrlfp16ps(i8 1, i8 2, i8 3)
  call void @llvm.x86.tconjtcmmimfp16ps(i8 1, i8 2, i8 3)
  call void @llvm.x86.tconjtfp16(i8 1, i8 2)
  ret void
}

declare void @llvm.x86.t2rpntlvwz0(i8 %tile1, i8* %addr1, i64 %stride)
declare void @llvm.x86.t2rpntlvwz0t1(i8 %tile1, i8* %addr1, i64 %stride)
declare void @llvm.x86.t2rpntlvwz1(i8 %tile1, i8* %addr1, i64 %stride)
declare void @llvm.x86.t2rpntlvwz1t1(i8 %tile1, i8* %addr1, i64 %stride)
declare void @llvm.x86.ttransposed(i8 %tile0, i8 %tile1)
declare void @llvm.x86.ttdpbf16ps(i8 %tile0, i8 %tile1, i8 %tile2)
declare void @llvm.x86.ttdpfp16ps(i8 %tile0, i8 %tile1, i8 %tile2)
declare void @llvm.x86.ttcmmimfp16ps(i8 %A, i8 %B, i8 %C)
declare void @llvm.x86.ttcmmrlfp16ps(i8 %A, i8 %B, i8 %C)
declare void @llvm.x86.tconjtcmmimfp16ps(i8 %A, i8 %B, i8 %C)
declare void @llvm.x86.tconjtfp16(i8 %A, i8 %B)

define void @test_amx2(i8* %pointer, i8* %base, i64 %stride) #0 {
; CHECK-LABEL: test_amx2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    pushq %rbp
; CHECK-NEXT:    subq $2928, %rsp # imm = 0xB70
; CHECK-NEXT:    vxorps %xmm0, %xmm0, %xmm0
; CHECK-NEXT:    vmovups %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT:    movb $1, {{[0-9]+}}(%rsp)
; CHECK-NEXT:    movb $8, {{[0-9]+}}(%rsp)
; CHECK-NEXT:    movw $8, {{[0-9]+}}(%rsp)
; CHECK-NEXT:    movb $8, {{[0-9]+}}(%rsp)
; CHECK-NEXT:    movw $8, {{[0-9]+}}(%rsp)
; CHECK-NEXT:    movb $8, {{[0-9]+}}(%rsp)
; CHECK-NEXT:    movw $8, {{[0-9]+}}(%rsp)
; CHECK-NEXT:    movb $8, {{[0-9]+}}(%rsp)
; CHECK-NEXT:    movw $8, {{[0-9]+}}(%rsp)
; CHECK-NEXT:    ldtilecfg {{[0-9]+}}(%rsp)
; CHECK-NEXT:    movw $8, %ax
; CHECK-NEXT:    tileloadd (%rsi,%rdx), %tmm0
; CHECK-NEXT:    tilezero %tmm1
; CHECK-NEXT:    tilezero %tmm2
; CHECK-NEXT:    ttdpbf16ps %tmm1, %tmm0, %tmm2
; CHECK-NEXT:    ttdpfp16ps %tmm1, %tmm0, %tmm2
; CHECK-NEXT:    ttcmmimfp16ps %tmm1, %tmm0, %tmm2
; CHECK-NEXT:    ttcmmrlfp16ps %tmm1, %tmm0, %tmm2
; CHECK-NEXT:    movabsq $64, %rbp
; CHECK-NEXT:    tilestored %tmm2, 896(%rsp,%rbp) # 1024-byte Folded Spill
; CHECK-NEXT:    tileloadd 896(%rsp,%rbp), %tmm3 # 1024-byte Folded Reload
; CHECK-NEXT:    tconjtcmmimfp16ps %tmm1, %tmm0, %tmm3
; CHECK-NEXT:    tconjtfp16 %tmm3, %tmm0
; CHECK-NEXT:    tilestored %tmm2, (%rdi,%rdx)
; CHECK-NEXT:    addq $2928, %rsp # imm = 0xB70
; CHECK-NEXT:    popq %rbp
; CHECK-NEXT:    tilerelease
; CHECK-NEXT:    vzeroupper
; CHECK-NEXT:    retq

  %a = call x86_amx @llvm.x86.tileloadd64.internal(i16 8, i16 8, i8* %base, i64 %stride)
  %b = call x86_amx @llvm.x86.tilezero.internal(i16 8, i16 8)
  %c = call x86_amx @llvm.x86.tilezero.internal(i16 8, i16 8)
  %c1 = call x86_amx @llvm.x86.ttdpbf16ps.internal(i16 8, i16 8, i16 8, x86_amx %c, x86_amx %a, x86_amx %b)
  %c2 = call x86_amx @llvm.x86.ttdpfp16ps.internal(i16 8, i16 8, i16 8, x86_amx %c1, x86_amx %a, x86_amx %b)
  %c3 = call x86_amx @llvm.x86.ttcmmimfp16ps.internal(i16 8, i16 8, i16 8, x86_amx %c2, x86_amx %a, x86_amx %b)
  %c4 = call x86_amx @llvm.x86.ttcmmrlfp16ps.internal(i16 8, i16 8, i16 8, x86_amx %c3, x86_amx %a, x86_amx %b)
  %c5 = call x86_amx @llvm.x86.tconjtcmmimfp16ps.internal(i16 8, i16 8, i16 8, x86_amx %c4, x86_amx %a, x86_amx %b)
  %c6 = call x86_amx @llvm.x86.tconjtfp16.internal(i16 8, i16 8, x86_amx %c5)

  call void @llvm.x86.tilestored64.internal(i16 8, i16 8, i8* %pointer, i64 %stride, x86_amx %c4)
  ret void
}

define void @test_amx3(i8* %pointer, i8* %base, i64 %stride) #0 {
; CHECK-LABEL: test_amx3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm0, %xmm0, %xmm0
; CHECK-NEXT:    vmovups %zmm0, -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    movb $1, -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    movb $8, -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    movw $8, -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    movb $8, -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    movw $8, -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    movb $0, -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    movw $0, -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    ldtilecfg -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    xorl %eax, %eax
; CHECK-NEXT:    movw $8, %cx
; CHECK-NEXT:    t2rpntlvwz0 (%rsi,%rdx), %tmm4
; CHECK-NEXT:    t2rpntlvwz0t1 (%rsi,%rdx), %tmm4
; CHECK-NEXT:    t2rpntlvwz1 (%rsi,%rdx), %tmm4
; CHECK-NEXT:    t2rpntlvwz1t1 (%rsi,%rdx), %tmm4
; CHECK-NEXT:    ttransposed %tmm4, %tmm0
; CHECK-NEXT:    tilestored %tmm0, (%rdi,%rdx)
; CHECK-NEXT:    tilerelease
; CHECK-NEXT:    vzeroupper
; CHECK-NEXT:    retq
  %1 = call { x86_amx, x86_amx } @llvm.x86.t2rpntlvwz0.internal(i16 8, i16 8, i16 0, i8* %base, i64 %stride)
  %2 = call { x86_amx, x86_amx } @llvm.x86.t2rpntlvwz0t1.internal(i16 8, i16 8, i16 0, i8* %base, i64 %stride)
  %3 = call { x86_amx, x86_amx } @llvm.x86.t2rpntlvwz1.internal(i16 8, i16 8, i16 0, i8* %base, i64 %stride)
  %4 = call { x86_amx, x86_amx } @llvm.x86.t2rpntlvwz1t1.internal(i16 8, i16 8, i16 0, i8* %base, i64 %stride)
  %5 = extractvalue { x86_amx, x86_amx } %4, 0
  %6 = call x86_amx @llvm.x86.ttransposed.internal(i16 8, i16 8, x86_amx %5)
  call void @llvm.x86.tilestored64.internal(i16 8, i16 8, i8* %pointer, i64 %stride, x86_amx %6)
  ret void
}

define void @test_amx_spill(i8* %pointer, i8* %base, i64 %stride) #0 {
; CHECK-LABEL: test_amx_spill:
; CHECK:       # %bb.0:
; CHECK-NEXT:    subq $6088, %rsp # imm = 0x17C8
; CHECK-NEXT:    vxorps %xmm0, %xmm0, %xmm0
; CHECK-NEXT:    vmovups %zmm0, -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    movb $1, -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    movb $8, -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    movw $8, -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    movb $8, -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    movw $8, -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    movb $8, -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    movw $8, -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    movb $8, -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    movw $8, -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    movb $8, -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    movw $8, -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    ldtilecfg -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    movw $8, %ax
; CHECK-NEXT:    tileloadd (%rsi,%rdx), %tmm0
; CHECK-NEXT:    t2rpntlvwz0 (%rsi,%rdx), %tmm4
; CHECK-NEXT:    t2rpntlvwz0t1 (%rsi,%rdx), %tmm6
; CHECK-NEXT:    movabsq $64, %rcx
; CHECK-NEXT:    tilestored %tmm6, 4032(%rsp,%rcx) # 1024-byte Folded Spill
; CHECK-NEXT:    tilestored %tmm7, 5056(%rsp,%rcx) # 1024-byte Folded Spill
; CHECK-NEXT:    t2rpntlvwz1 (%rsi,%rdx), %tmm6
; CHECK-NEXT:    tilestored %tmm6, 1984(%rsp,%rcx) # 1024-byte Folded Spill
; CHECK-NEXT:    tilestored %tmm7, 3008(%rsp,%rcx) # 1024-byte Folded Spill
; CHECK-NEXT:    t2rpntlvwz1t1 (%rsi,%rdx), %tmm6
; CHECK-NEXT:    tilestored %tmm6, -64(%rsp,%rcx) # 1024-byte Folded Spill
; CHECK-NEXT:    tilestored %tmm7, 960(%rsp,%rcx) # 1024-byte Folded Spill
; CHECK-NEXT:    t2rpntlvwz0 (%rsi,%rdx), %tmm6
; CHECK-NEXT:    tilestored %tmm4, (%rsi,%rdx)
; CHECK-NEXT:    tilestored %tmm5, (%rsi,%rdx)
; CHECK-NEXT:    tileloadd 4032(%rsp,%rcx), %tmm4 # 1024-byte Folded Reload
; CHECK-NEXT:    tileloadd 5056(%rsp,%rcx), %tmm5 # 1024-byte Folded Reload
; CHECK-NEXT:    tilestored %tmm4, (%rsi,%rdx)
; CHECK-NEXT:    tilestored %tmm5, (%rsi,%rdx)
; CHECK-NEXT:    tileloadd 1984(%rsp,%rcx), %tmm4 # 1024-byte Folded Reload
; CHECK-NEXT:    tileloadd 3008(%rsp,%rcx), %tmm5 # 1024-byte Folded Reload
; CHECK-NEXT:    tilestored %tmm4, (%rsi,%rdx)
; CHECK-NEXT:    tilestored %tmm5, (%rsi,%rdx)
; CHECK-NEXT:    tileloadd -64(%rsp,%rcx), %tmm4 # 1024-byte Folded Reload
; CHECK-NEXT:    tileloadd 960(%rsp,%rcx), %tmm5 # 1024-byte Folded Reload
; CHECK-NEXT:    tilestored %tmm4, (%rsi,%rdx)
; CHECK-NEXT:    tilestored %tmm5, (%rsi,%rdx)
; CHECK-NEXT:    tilestored %tmm6, (%rsi,%rdx)
; CHECK-NEXT:    tilestored %tmm7, (%rsi,%rdx)
; CHECK-NEXT:    addq $6088, %rsp # imm = 0x17C8
; CHECK-NEXT:    tilerelease
; CHECK-NEXT:    vzeroupper
; CHECK-NEXT:    retq
  %a = call x86_amx @llvm.x86.tileloadd64.internal(i16 8, i16 8, i8* %base, i64 %stride)
  %b1 = call { x86_amx, x86_amx } @llvm.x86.t2rpntlvwz0.internal(i16 8, i16 8, i16 8, i8* %base, i64 %stride)
  %b2 = call { x86_amx, x86_amx } @llvm.x86.t2rpntlvwz0t1.internal(i16 8, i16 8, i16 8, i8* %base, i64 %stride)
  %b3 = call { x86_amx, x86_amx } @llvm.x86.t2rpntlvwz1.internal(i16 8, i16 8, i16 8, i8* %base, i64 %stride)
  %b4 = call { x86_amx, x86_amx } @llvm.x86.t2rpntlvwz1t1.internal(i16 8, i16 8, i16 8, i8* %base, i64 %stride)
  %b5 = call { x86_amx, x86_amx } @llvm.x86.t2rpntlvwz0.internal(i16 8, i16 8, i16 8, i8* %base, i64 %stride)
  %e11 = extractvalue { x86_amx, x86_amx } %b1, 0
  %e12 = extractvalue { x86_amx, x86_amx } %b1, 1
  %e21 = extractvalue { x86_amx, x86_amx } %b2, 0
  %e22 = extractvalue { x86_amx, x86_amx } %b2, 1
  %e31 = extractvalue { x86_amx, x86_amx } %b3, 0
  %e32 = extractvalue { x86_amx, x86_amx } %b3, 1
  %e41 = extractvalue { x86_amx, x86_amx } %b4, 0
  %e42 = extractvalue { x86_amx, x86_amx } %b4, 1
  %e51 = extractvalue { x86_amx, x86_amx } %b5, 0
  %e52 = extractvalue { x86_amx, x86_amx } %b5, 1
  call void @llvm.x86.tilestored64.internal(i16 8, i16 8, i8* %base, i64 %stride, x86_amx %e11)
  call void @llvm.x86.tilestored64.internal(i16 8, i16 8, i8* %base, i64 %stride, x86_amx %e12)
  call void @llvm.x86.tilestored64.internal(i16 8, i16 8, i8* %base, i64 %stride, x86_amx %e21)
  call void @llvm.x86.tilestored64.internal(i16 8, i16 8, i8* %base, i64 %stride, x86_amx %e22)
  call void @llvm.x86.tilestored64.internal(i16 8, i16 8, i8* %base, i64 %stride, x86_amx %e31)
  call void @llvm.x86.tilestored64.internal(i16 8, i16 8, i8* %base, i64 %stride, x86_amx %e32)
  call void @llvm.x86.tilestored64.internal(i16 8, i16 8, i8* %base, i64 %stride, x86_amx %e41)
  call void @llvm.x86.tilestored64.internal(i16 8, i16 8, i8* %base, i64 %stride, x86_amx %e42)
  call void @llvm.x86.tilestored64.internal(i16 8, i16 8, i8* %base, i64 %stride, x86_amx %e51)
  call void @llvm.x86.tilestored64.internal(i16 8, i16 8, i8* %base, i64 %stride, x86_amx %e52)
  ret void
}

declare x86_amx @llvm.x86.tileloadd64.internal(i16, i16, i8*, i64)
declare void @llvm.x86.tilestored64.internal(i16, i16, i8*, i64, x86_amx)
declare { x86_amx, x86_amx } @llvm.x86.t2rpntlvwz0.internal(i16, i16, i16, i8*, i64)
declare { x86_amx, x86_amx } @llvm.x86.t2rpntlvwz0t1.internal(i16, i16, i16, i8*, i64)
declare { x86_amx, x86_amx } @llvm.x86.t2rpntlvwz1.internal(i16, i16, i16, i8*, i64)
declare { x86_amx, x86_amx } @llvm.x86.t2rpntlvwz1t1.internal(i16, i16, i16, i8*, i64)
declare x86_amx @llvm.x86.ttransposed.internal(i16, i16, x86_amx)
declare x86_amx @llvm.x86.ttdpbf16ps.internal(i16, i16, i16, x86_amx, x86_amx, x86_amx)
declare x86_amx @llvm.x86.ttdpfp16ps.internal(i16, i16, i16, x86_amx, x86_amx, x86_amx)
declare x86_amx @llvm.x86.ttcmmimfp16ps.internal(i16, i16, i16, x86_amx, x86_amx, x86_amx)
declare x86_amx @llvm.x86.ttcmmrlfp16ps.internal(i16, i16, i16, x86_amx, x86_amx, x86_amx)
declare x86_amx @llvm.x86.tconjtcmmimfp16ps.internal(i16, i16, i16, x86_amx, x86_amx, x86_amx)
declare x86_amx @llvm.x86.tconjtfp16.internal(i16, i16, x86_amx)

attributes #0 = { nounwind }