Compiler Explorer

Source code

target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
target triple = "x86_64-apple-macosx"

; Function Attrs: mustprogress noinline nounwind ssp uwtable
define void @_Z14inner_product2PiS_S_iii(ptr noundef %B_mem, ptr noundef %C_mem, i32 noundef %M, i32 noundef %N) #0 {
entry:
  %div = sdiv i32 %M, 16
  %div1 = sdiv i32 %N, 16
  %factor.op.mul114 = shl i32 %M, 4
  %cmp115 = icmp sgt i32 %M, 15
  %cmp496 = icmp sgt i32 %N, 15
  %mul15 = shl nsw i64 0, 2
  %conv23 = sext i32 %N to i64
  %mul24 = shl nsw i64 %conv23, 2
  %wide.trip.count158 = zext nneg i32 %div to i64
  %wide.trip.count153 = zext nneg i32 %div1 to i64
  %wide.trip.count146 = zext nneg i32 0 to i64
  %0 = add nsw i64 %wide.trip.count146, -1
  %xtraiter163 = and i64 %wide.trip.count146, 3
  %1 = icmp ult i64 %0, 3
  %unroll_iter167 = and i64 %wide.trip.count146, 134217724
  %lcmp.mod165.not = icmp eq i64 %xtraiter163, 0
  br label %for.cond3.preheader.us.us

for.cond3.preheader.us.us:                        ; preds = %for.cond3.for.cond.cleanup5_crit_edge.split.us.us.us, %entry
  %indvars.iv155 = phi i64 [ 0, %entry ], [ %indvars.iv.next156, %for.cond3.for.cond.cleanup5_crit_edge.split.us.us.us ]
  %2 = trunc nuw nsw i64 %indvars.iv155 to i32
  %factor.op.mul95.us.us = mul i32 %factor.op.mul114, %2
  %factor.op.mul.us.us = mul i32 0, %2
  %idx.ext.us.us = sext i32 %factor.op.mul.us.us to i64
  %idx.ext28.us.us = sext i32 %factor.op.mul95.us.us to i64
  %add.ptr29.us.us = getelementptr inbounds i32, ptr %C_mem, i64 %idx.ext28.us.us
  br label %for.body6.us.us.us

for.body6.us.us.us:                               ; preds = %for.cond7.for.cond.cleanup9_crit_edge.us.us.us, %for.cond3.preheader.us.us
  %indvars.iv148 = phi i64 [ %indvars.iv.next149, %for.cond7.for.cond.cleanup9_crit_edge.us.us.us ], [ 0, %for.cond3.preheader.us.us ]
  %3 = tail call x86_amx @llvm.x86.tilezero.internal(i16 16, i16 64)
  %4 = tail call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx %3)
  %5 = shl nsw i64 %indvars.iv148, 4
  %invariant.gep.us.us.us = getelementptr i32, ptr %B_mem, i64 %5
  br i1 %lcmp.mod165.not, label %for.cond7.for.cond.cleanup9_crit_edge.us.us.us, label %for.body10.us.us.us.epil

for.body10.us.us.us.epil:                         ; preds = %for.body6.us.us.us
  %6 = shl nsw i64 0, 4
  %add.ptr14.us.us.us.epil = getelementptr inbounds i32, ptr null, i64 %6
  %7 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 16, i16 64, ptr %add.ptr14.us.us.us.epil, i64 %mul15)
  %8 = tail call noundef <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx %7)
  %9 = mul nuw nsw i64 %6, %conv23
  %gep.us.us.us.epil = getelementptr i32, ptr %invariant.gep.us.us.us, i64 %9
  %10 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 16, i16 64, ptr %gep.us.us.us.epil, i64 %mul24)
  %11 = tail call noundef <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx %10)
  %12 = tail call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> %4)
  %13 = tail call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> %8)
  %14 = tail call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> %11)
  %15 = tail call x86_amx @llvm.x86.tdpbf16ps.internal(i16 16, i16 64, i16 64, x86_amx %12, x86_amx %13, x86_amx %14)
  %16 = tail call noundef <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx %15)
  %indvars.iv.next142.epil = add nuw nsw i64 0, 1
  %epil.iter164.next = add i64 0, 1
  %epil.iter164.cmp.not = icmp eq i64 %epil.iter164.next, %xtraiter163
  br label %for.cond7.for.cond.cleanup9_crit_edge.us.us.us

for.cond7.for.cond.cleanup9_crit_edge.us.us.us:   ; preds = %for.body10.us.us.us.epil, %for.body6.us.us.us
  %.lcssa = phi <256 x i32> [ undef, %for.body6.us.us.us ], [ %16, %for.body10.us.us.us.epil ]
  %add.ptr32.us.us.us = getelementptr inbounds i32, ptr %add.ptr29.us.us, i64 %5
  %17 = tail call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> %.lcssa)
  tail call void @llvm.x86.tilestored64.internal(i16 16, i16 64, ptr %add.ptr32.us.us.us, i64 %mul24, x86_amx %17)
  %indvars.iv.next149 = add nuw nsw i64 %indvars.iv148, 1
  %exitcond154.not = icmp eq i64 %indvars.iv.next149, %wide.trip.count153
  br i1 %exitcond154.not, label %for.cond3.for.cond.cleanup5_crit_edge.split.us.us.us, label %for.body6.us.us.us

for.cond3.for.cond.cleanup5_crit_edge.split.us.us.us: ; preds = %for.cond7.for.cond.cleanup9_crit_edge.us.us.us
  %indvars.iv.next156 = add nuw nsw i64 %indvars.iv155, 1
  %exitcond159.not = icmp eq i64 %indvars.iv.next156, %wide.trip.count158
  br label %for.cond3.preheader.us.us
  
}

; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
declare void @llvm.lifetime.start.p0(i64 immarg, ptr nocapture) #1

; Function Attrs: nocallback nofree nounwind willreturn memory(argmem: write)
declare void @llvm.memset.p0.i64(ptr nocapture writeonly, i8, i64, i1 immarg) #2

; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
declare void @llvm.lifetime.end.p0(i64 immarg, ptr nocapture) #1

; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
declare float @llvm.fmuladd.f32(float, float, float) #3

; Function Attrs: nounwind
declare x86_amx @llvm.x86.tilezero.internal(i16, i16) #4

; Function Attrs: nocallback nofree nosync nounwind willreturn memory(none)
declare <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx) #5

; Function Attrs: nounwind
declare x86_amx @llvm.x86.tileloadd64.internal(i16, i16, ptr, i64) #4

; Function Attrs: nounwind
declare x86_amx @llvm.x86.tdpbf16ps.internal(i16, i16, i16, x86_amx, x86_amx, x86_amx) #4

; Function Attrs: nocallback nofree nosync nounwind willreturn memory(none)
declare x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32>) #5

; Function Attrs: nounwind
declare void @llvm.x86.tilestored64.internal(i16, i16, ptr, i64, x86_amx) #4

; uselistorder directives
uselistorder ptr @llvm.x86.cast.tile.to.vector.v256i32, { 3, 2, 1, 0 }
uselistorder ptr @llvm.x86.tileloadd64.internal, { 1, 0 }
uselistorder ptr @llvm.x86.cast.vector.to.tile.v256i32, { 3, 2, 1, 0 }

attributes #0 = { mustprogress noinline nounwind ssp uwtable "frame-pointer"="all" "min-legal-vector-width"="8192" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="sapphirerapids" "target-features"="+adx,+aes,+amx-bf16,+amx-int8,+amx-tile,+avx,+avx2,+avx512bf16,+avx512bitalg,+avx512bw,+avx512cd,+avx512dq,+avx512f,+avx512fp16,+avx512ifma,+avx512vbmi,+avx512vbmi2,+avx512vl,+avx512vnni,+avx512vpopcntdq,+avxvnni,+bmi,+bmi2,+cldemote,+clflushopt,+clwb,+cmov,+crc32,+cx16,+cx8,+enqcmd,+evex512,+f16c,+fma,+fsgsbase,+fxsr,+gfni,+invpcid,+lzcnt,+mmx,+movbe,+movdir64b,+movdiri,+pclmul,+pconfig,+pku,+popcnt,+prfchw,+ptwrite,+rdpid,+rdrnd,+rdseed,+sahf,+serialize,+sgx,+sha,+shstk,+sse,+sse2,+sse3,+sse4.1,+sse4.2,+ssse3,+tsxldtrk,+uintr,+vaes,+vpclmulqdq,+waitpkg,+wbnoinvd,+x87,+xsave,+xsavec,+xsaveopt,+xsaves" }
attributes #1 = { nocallback nofree nosync nounwind willreturn memory(argmem: readwrite) }
attributes #2 = { nocallback nofree nounwind willreturn memory(argmem: write) }
attributes #3 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
attributes #4 = { nounwind }
attributes #5 = { nocallback nofree nosync nounwind willreturn memory(none) }