Add many tests

EnzymeAD · Jan 19, 2022 · 9090ce4 · 9090ce4
1 parent 43d74ad
commit 9090ce4
Show file tree

Hide file tree

Showing 10 changed files with 781 additions and 218 deletions.
diff --git a/.gitignore b/.gitignore
@@ -5,3 +5,4 @@ build
 .vscode
 .cache
 .DS_Store
+.clangd
diff --git a/enzyme/Enzyme/AdjointGenerator.h b/enzyme/Enzyme/AdjointGenerator.h
diff --git a/enzyme/test/Enzyme/ReverseMode/blas/cblas_daxpy_a.ll b/enzyme/test/Enzyme/ReverseMode/blas/cblas_daxpy_a.ll
@@ -0,0 +1,84 @@
+;RUN: %opt < %s %loadEnzyme -enzyme -mem2reg -instsimplify -simplifycfg -S | FileCheck %s
+
+target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-apple-macosx12.0.0"
+
+@__const.main.m = private unnamed_addr constant [3 x double] [double 1.000000e+00, double 2.000000e+00, double 3.000000e+00], align 16
+@__const.main.d_m = private unnamed_addr constant [3 x double] [double 2.000000e+00, double 4.000000e+00, double 6.000000e+00], align 16
+@__const.main.n = private unnamed_addr constant [3 x double] [double 4.000000e+00, double 5.000000e+00, double 6.000000e+00], align 16
+@__const.main.d_n = private unnamed_addr constant [3 x double] [double 8.000000e+00, double 1.000000e+01, double 1.200000e+01], align 16
+
+; Function Attrs: nounwind ssp uwtable
+define void @f(i32 %n, double %a, double* noalias %x, double* noalias %y) #0 {
+entry:
+  tail call void @cblas_daxpy(i32 %n, double %a, double* %x, i32 1, double* %y, i32 1) #4
+  ret void
+}
+
+declare void @cblas_daxpy(i32, double, double*, i32, double*, i32) local_unnamed_addr #1
+
+; Function Attrs: nounwind ssp uwtable
+define i32 @main() local_unnamed_addr #0 {
+entry:
+  %m = alloca [3 x double], align 16
+  %d_m = alloca [3 x double], align 16
+  %n = alloca [3 x double], align 16
+  %d_n = alloca [3 x double], align 16
+  %0 = bitcast [3 x double]* %m to i8*
+  call void @llvm.lifetime.start.p0i8(i64 24, i8* nonnull %0) #4
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 16 dereferenceable(24) %0, i8* noundef nonnull align 16 dereferenceable(24) bitcast ([3 x double]* @__const.main.m to i8*), i64 24, i1 false)
+  %1 = bitcast [3 x double]* %d_m to i8*
+  call void @llvm.lifetime.start.p0i8(i64 24, i8* nonnull %1) #4
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 16 dereferenceable(24) %1, i8* noundef nonnull align 16 dereferenceable(24) bitcast ([3 x double]* @__const.main.d_m to i8*), i64 24, i1 false)
+  %2 = bitcast [3 x double]* %n to i8*
+  call void @llvm.lifetime.start.p0i8(i64 24, i8* nonnull %2) #4
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 16 dereferenceable(24) %2, i8* noundef nonnull align 16 dereferenceable(24) bitcast ([3 x double]* @__const.main.n to i8*), i64 24, i1 false)
+  %3 = bitcast [3 x double]* %d_n to i8*
+  call void @llvm.lifetime.start.p0i8(i64 24, i8* nonnull %3) #4
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 16 dereferenceable(24) %3, i8* noundef nonnull align 16 dereferenceable(24) bitcast ([3 x double]* @__const.main.d_n to i8*), i64 24, i1 false)
+  %arraydecay = getelementptr inbounds [3 x double], [3 x double]* %m, i64 0, i64 0
+  %arraydecay1 = getelementptr inbounds [3 x double], [3 x double]* %d_m, i64 0, i64 0
+  %arraydecay2 = getelementptr inbounds [3 x double], [3 x double]* %n, i64 0, i64 0
+  %arraydecay3 = getelementptr inbounds [3 x double], [3 x double]* %d_n, i64 0, i64 0
+  %call = call double @__enzyme_autodiff(i8* bitcast (void (i32, double, double*, double*)* @f to i8*), i32 3, double 1.000000e+00, double* nonnull %arraydecay, double* nonnull %arraydecay1, double* nonnull %arraydecay2, double* nonnull %arraydecay3) #4
+  call void @llvm.lifetime.end.p0i8(i64 24, i8* nonnull %3) #4
+  call void @llvm.lifetime.end.p0i8(i64 24, i8* nonnull %2) #4
+  call void @llvm.lifetime.end.p0i8(i64 24, i8* nonnull %1) #4
+  call void @llvm.lifetime.end.p0i8(i64 24, i8* nonnull %0) #4
+  ret i32 0
+}
+
+; Function Attrs: argmemonly mustprogress nofree nosync nounwind willreturn
+declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #2
+
+; Function Attrs: argmemonly mustprogress nofree nounwind willreturn
+declare void @llvm.memcpy.p0i8.p0i8.i64(i8* noalias nocapture writeonly, i8* noalias nocapture readonly, i64, i1 immarg) #3
+
+declare double @__enzyme_autodiff(i8*, i32, double, double*, double*, double*, double*) local_unnamed_addr #1
+
+; Function Attrs: argmemonly mustprogress nofree nosync nounwind willreturn
+declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #2
+
+attributes #0 = { nounwind ssp uwtable "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="penryn" "target-features"="+cx16,+cx8,+fxsr,+mmx,+sahf,+sse,+sse2,+sse3,+sse4.1,+ssse3,+x87" "tune-cpu"="generic" }
+attributes #1 = { "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="penryn" "target-features"="+cx16,+cx8,+fxsr,+mmx,+sahf,+sse,+sse2,+sse3,+sse4.1,+ssse3,+x87" "tune-cpu"="generic" }
+attributes #2 = { argmemonly mustprogress nofree nosync nounwind willreturn }
+attributes #3 = { argmemonly mustprogress nofree nounwind willreturn }
+attributes #4 = { nounwind }
+
+!llvm.module.flags = !{!0, !1, !2, !3}
+!llvm.ident = !{!4}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{i32 7, !"PIC Level", i32 2}
+!2 = !{i32 7, !"uwtable", i32 1}
+!3 = !{i32 7, !"frame-pointer", i32 2}
+!4 = !{!"Homebrew clang version 13.0.0"}
+
+;CHECK:define internal { double } @diffef(i32 %n, double %a, double* noalias %x, double* %"x'", double* noalias %y, double* %"y'") #0 {
+;CHECK-NEXT:entry:
+;CHECK-NEXT:  tail call void @cblas_daxpy(i32 %n, double %a, double* %x, i32 1, double* %y, i32 1) #4
+;CHECK-NEXT:  call void @cblas_daxpy(i32 %n, double %a, double* %"y'", i32 1, double* %"x'", i32 1)
+;CHECK-NEXT:  %0 = call fast double @cblas_ddot(i32 %n, double* nocapture readonly %"y'", i32 1, double* nocapture readonly %x, i32 1)
+;CHECK-NEXT:  %1 = insertvalue { double } undef, double %0, 0
+;CHECK-NEXT:  ret { double } %1
+;CHECK-NEXT:}
diff --git a/enzyme/test/Enzyme/ReverseMode/blas/cblas_daxpy_x.ll b/enzyme/test/Enzyme/ReverseMode/blas/cblas_daxpy_x.ll
@@ -0,0 +1,82 @@
+;RUN: %opt < %s %loadEnzyme -enzyme -mem2reg -instsimplify -simplifycfg -S | FileCheck %s
+
+target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-apple-macosx12.0.0"
+
+@__const.main.m = private unnamed_addr constant [3 x double] [double 1.000000e+00, double 2.000000e+00, double 3.000000e+00], align 16
+@__const.main.d_m = private unnamed_addr constant [3 x double] [double 2.000000e+00, double 4.000000e+00, double 6.000000e+00], align 16
+@__const.main.n = private unnamed_addr constant [3 x double] [double 4.000000e+00, double 5.000000e+00, double 6.000000e+00], align 16
+@__const.main.d_n = private unnamed_addr constant [3 x double] [double 8.000000e+00, double 1.000000e+01, double 1.200000e+01], align 16
+
+; Function Attrs: nounwind ssp uwtable
+define void @f(i32 %n, double* noalias %x, double* noalias %y) #0 {
+entry:
+  tail call void @cblas_daxpy(i32 %n, double 1.000000e+00, double* %x, i32 1, double* %y, i32 1) #4
+  ret void
+}
+
+declare void @cblas_daxpy(i32, double, double*, i32, double*, i32) local_unnamed_addr #1
+
+; Function Attrs: nounwind ssp uwtable
+define i32 @main() local_unnamed_addr #0 {
+entry:
+  %m = alloca [3 x double], align 16
+  %d_m = alloca [3 x double], align 16
+  %n = alloca [3 x double], align 16
+  %d_n = alloca [3 x double], align 16
+  %0 = bitcast [3 x double]* %m to i8*
+  call void @llvm.lifetime.start.p0i8(i64 24, i8* nonnull %0) #4
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 16 dereferenceable(24) %0, i8* noundef nonnull align 16 dereferenceable(24) bitcast ([3 x double]* @__const.main.m to i8*), i64 24, i1 false)
+  %1 = bitcast [3 x double]* %d_m to i8*
+  call void @llvm.lifetime.start.p0i8(i64 24, i8* nonnull %1) #4
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 16 dereferenceable(24) %1, i8* noundef nonnull align 16 dereferenceable(24) bitcast ([3 x double]* @__const.main.d_m to i8*), i64 24, i1 false)
+  %2 = bitcast [3 x double]* %n to i8*
+  call void @llvm.lifetime.start.p0i8(i64 24, i8* nonnull %2) #4
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 16 dereferenceable(24) %2, i8* noundef nonnull align 16 dereferenceable(24) bitcast ([3 x double]* @__const.main.n to i8*), i64 24, i1 false)
+  %3 = bitcast [3 x double]* %d_n to i8*
+  call void @llvm.lifetime.start.p0i8(i64 24, i8* nonnull %3) #4
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 16 dereferenceable(24) %3, i8* noundef nonnull align 16 dereferenceable(24) bitcast ([3 x double]* @__const.main.d_n to i8*), i64 24, i1 false)
+  %arraydecay = getelementptr inbounds [3 x double], [3 x double]* %m, i64 0, i64 0
+  %arraydecay1 = getelementptr inbounds [3 x double], [3 x double]* %d_m, i64 0, i64 0
+  %arraydecay2 = getelementptr inbounds [3 x double], [3 x double]* %n, i64 0, i64 0
+  %arraydecay3 = getelementptr inbounds [3 x double], [3 x double]* %d_n, i64 0, i64 0
+  call void @__enzyme_autodiff(i8* bitcast (void (i32, double*, double*)* @f to i8*), i32 3, double* nonnull %arraydecay, double* nonnull %arraydecay1, double* nonnull %arraydecay2, double* nonnull %arraydecay3) #4
+  call void @llvm.lifetime.end.p0i8(i64 24, i8* nonnull %3) #4
+  call void @llvm.lifetime.end.p0i8(i64 24, i8* nonnull %2) #4
+  call void @llvm.lifetime.end.p0i8(i64 24, i8* nonnull %1) #4
+  call void @llvm.lifetime.end.p0i8(i64 24, i8* nonnull %0) #4
+  ret i32 0
+}
+
+; Function Attrs: argmemonly mustprogress nofree nosync nounwind willreturn
+declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #2
+
+; Function Attrs: argmemonly mustprogress nofree nounwind willreturn
+declare void @llvm.memcpy.p0i8.p0i8.i64(i8* noalias nocapture writeonly, i8* noalias nocapture readonly, i64, i1 immarg) #3
+
+declare void @__enzyme_autodiff(i8*, i32, double*, double*, double*, double*) local_unnamed_addr #1
+
+; Function Attrs: argmemonly mustprogress nofree nosync nounwind willreturn
+declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #2
+
+attributes #0 = { nounwind ssp uwtable "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="penryn" "target-features"="+cx16,+cx8,+fxsr,+mmx,+sahf,+sse,+sse2,+sse3,+sse4.1,+ssse3,+x87" "tune-cpu"="generic" }
+attributes #1 = { "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="penryn" "target-features"="+cx16,+cx8,+fxsr,+mmx,+sahf,+sse,+sse2,+sse3,+sse4.1,+ssse3,+x87" "tune-cpu"="generic" }
+attributes #2 = { argmemonly mustprogress nofree nosync nounwind willreturn }
+attributes #3 = { argmemonly mustprogress nofree nounwind willreturn }
+attributes #4 = { nounwind }
+
+!llvm.module.flags = !{!0, !1, !2, !3}
+!llvm.ident = !{!4}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{i32 7, !"PIC Level", i32 2}
+!2 = !{i32 7, !"uwtable", i32 1}
+!3 = !{i32 7, !"frame-pointer", i32 2}
+!4 = !{!"Homebrew clang version 13.0.0"}
+
+;CHECK:define internal void @diffef(i32 %n, double* noalias %x, double* %"x'", double* noalias %y, double* %"y'") #0 {
+;CHECK-NEXT:entry:
+;CHECK-NEXT:  tail call void @cblas_daxpy(i32 %n, double 1.000000e+00, double* %x, i32 1, double* %y, i32 1) #4
+;CHECK-NEXT:  call void @cblas_daxpy(i32 %n, double 1.000000e+00, double* %"y'", i32 1, double* %"x'", i32 1)
+;CHECK-NEXT:  ret void
+;CHECK-NEXT:}
diff --git a/enzyme/test/Enzyme/ReverseMode/blas/cblas_dgemv_alpha.ll b/enzyme/test/Enzyme/ReverseMode/blas/cblas_dgemv_alpha.ll
@@ -0,0 +1,105 @@
+;RUN: %opt < %s %loadEnzyme -enzyme -mem2reg -instsimplify -simplifycfg -S | FileCheck %s
+
+target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-apple-macosx12.0.0"
+
+@__const.main.a = private unnamed_addr constant [4 x double] [double 1.000000e+00, double 0.000000e+00, double 0.000000e+00, double 1.000000e+00], align 16
+@__const.main.x = private unnamed_addr constant [2 x double] [double 1.000000e+00, double 1.000000e+00], align 16
+@__const.main.y = private unnamed_addr constant [3 x double] [double 1.000000e+00, double 2.000000e+00, double 3.000000e+00], align 16
+@__const.main.d_y = private unnamed_addr constant [3 x double] [double 2.000000e+00, double 4.000000e+00, double 6.000000e+00], align 16
+
+; Function Attrs: nounwind ssp uwtable
+define void @f(double* noalias %x, double* noalias %y, double* noalias %a, double %alpha) #0 {
+entry:
+  tail call void @cblas_dgemv(i32 102, i32 111, i32 2, i32 2, double %alpha, double* %a, i32 2, double* %x, i32 1, double 2.000000e+00, double* %y, i32 1) #5
+  ret void
+}
+
+; Function Attrs: argmemonly mustprogress nofree nosync nounwind willreturn
+declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #1
+
+declare void @cblas_dgemv(i32, i32, i32, i32, double, double*, i32, double*, i32, double, double*, i32) local_unnamed_addr #2
+
+; Function Attrs: argmemonly mustprogress nofree nosync nounwind willreturn
+declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #1
+
+; Function Attrs: nounwind ssp uwtable
+define i32 @main() local_unnamed_addr #0 {
+entry:
+  %a = alloca [4 x double], align 16
+  %d_a = alloca [4 x double], align 16
+  %x = alloca [2 x double], align 16
+  %d_x = alloca [2 x double], align 16
+  %y = alloca [3 x double], align 16
+  %d_y = alloca [3 x double], align 16
+  %0 = bitcast [4 x double]* %a to i8*
+  call void @llvm.lifetime.start.p0i8(i64 32, i8* nonnull %0) #5
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 16 dereferenceable(32) %0, i8* noundef nonnull align 16 dereferenceable(32) bitcast ([4 x double]* @__const.main.a to i8*), i64 32, i1 false)
+  %1 = bitcast [4 x double]* %d_a to i8*
+  call void @llvm.lifetime.start.p0i8(i64 32, i8* nonnull %1) #5
+  call void @llvm.memset.p0i8.i64(i8* noundef nonnull align 16 dereferenceable(32) %1, i8 0, i64 32, i1 false)
+  %2 = bitcast [2 x double]* %x to i8*
+  call void @llvm.lifetime.start.p0i8(i64 16, i8* nonnull %2) #5
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 16 dereferenceable(16) %2, i8* noundef nonnull align 16 dereferenceable(16) bitcast ([2 x double]* @__const.main.x to i8*), i64 16, i1 false)
+  %3 = bitcast [2 x double]* %d_x to i8*
+  call void @llvm.lifetime.start.p0i8(i64 16, i8* nonnull %3) #5
+  call void @llvm.memset.p0i8.i64(i8* noundef nonnull align 16 dereferenceable(16) %3, i8 0, i64 16, i1 false)
+  %4 = bitcast [3 x double]* %y to i8*
+  call void @llvm.lifetime.start.p0i8(i64 24, i8* nonnull %4) #5
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 16 dereferenceable(24) %4, i8* noundef nonnull align 16 dereferenceable(24) bitcast ([3 x double]* @__const.main.y to i8*), i64 24, i1 false)
+  %5 = bitcast [3 x double]* %d_y to i8*
+  call void @llvm.lifetime.start.p0i8(i64 24, i8* nonnull %5) #5
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 16 dereferenceable(24) %5, i8* noundef nonnull align 16 dereferenceable(24) bitcast ([3 x double]* @__const.main.d_y to i8*), i64 24, i1 false)
+  %arraydecay = getelementptr inbounds [2 x double], [2 x double]* %x, i64 0, i64 0
+  %arraydecay1 = getelementptr inbounds [2 x double], [2 x double]* %d_x, i64 0, i64 0
+  %arraydecay2 = getelementptr inbounds [3 x double], [3 x double]* %y, i64 0, i64 0
+  %arraydecay3 = getelementptr inbounds [3 x double], [3 x double]* %d_y, i64 0, i64 0
+  %arraydecay4 = getelementptr inbounds [4 x double], [4 x double]* %a, i64 0, i64 0
+  %arraydecay5 = getelementptr inbounds [4 x double], [4 x double]* %d_a, i64 0, i64 0
+  %call = call double @__enzyme_autodiff(i8* bitcast (void (double*, double*, double*, double)* @f to i8*), double* nonnull %arraydecay, double* nonnull %arraydecay1, double* nonnull %arraydecay2, double* nonnull %arraydecay3, double* nonnull %arraydecay4, double* nonnull %arraydecay5, double 3.000000e+00) #5
+  call void @llvm.lifetime.end.p0i8(i64 24, i8* nonnull %5) #5
+  call void @llvm.lifetime.end.p0i8(i64 24, i8* nonnull %4) #5
+  call void @llvm.lifetime.end.p0i8(i64 16, i8* nonnull %3) #5
+  call void @llvm.lifetime.end.p0i8(i64 16, i8* nonnull %2) #5
+  call void @llvm.lifetime.end.p0i8(i64 32, i8* nonnull %1) #5
+  call void @llvm.lifetime.end.p0i8(i64 32, i8* nonnull %0) #5
+  ret i32 0
+}
+
+; Function Attrs: argmemonly mustprogress nofree nounwind willreturn
+declare void @llvm.memcpy.p0i8.p0i8.i64(i8* noalias nocapture writeonly, i8* noalias nocapture readonly, i64, i1 immarg) #3
+
+; Function Attrs: argmemonly mustprogress nofree nounwind willreturn writeonly
+declare void @llvm.memset.p0i8.i64(i8* nocapture writeonly, i8, i64, i1 immarg) #4
+
+declare double @__enzyme_autodiff(i8*, double*, double*, double*, double*, double*, double*, double) local_unnamed_addr #2
+
+attributes #0 = { nounwind ssp uwtable "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="penryn" "target-features"="+cx16,+cx8,+fxsr,+mmx,+sahf,+sse,+sse2,+sse3,+sse4.1,+ssse3,+x87" "tune-cpu"="generic" }
+attributes #1 = { argmemonly mustprogress nofree nosync nounwind willreturn }
+attributes #2 = { "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="penryn" "target-features"="+cx16,+cx8,+fxsr,+mmx,+sahf,+sse,+sse2,+sse3,+sse4.1,+ssse3,+x87" "tune-cpu"="generic" }
+attributes #3 = { argmemonly mustprogress nofree nounwind willreturn }
+attributes #4 = { argmemonly mustprogress nofree nounwind willreturn writeonly }
+attributes #5 = { nounwind }
+
+!llvm.module.flags = !{!0, !1, !2, !3}
+!llvm.ident = !{!4}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{i32 7, !"PIC Level", i32 2}
+!2 = !{i32 7, !"uwtable", i32 1}
+!3 = !{i32 7, !"frame-pointer", i32 2}
+!4 = !{!"Homebrew clang version 13.0.0"}
+
+;CHECK:define internal { double } @diffef(double* noalias %x, double* %"x'", double* noalias %y, double* %"y'", double* noalias %a, double* %"a'", double %alpha) #0 {
+;CHECK-NEXT:entry:
+;CHECK-NEXT:  %malloccall = tail call i8* @malloc(i64 mul (i64 ptrtoint (double* getelementptr (double, double* null, i32 1) to i64), i64 2))
+;CHECK-NEXT:  %0 = bitcast i8* %malloccall to double*
+;CHECK-NEXT:  tail call void @cblas_dgemv(i32 102, i32 111, i32 2, i32 2, double %alpha, double* %a, i32 2, double* %x, i32 1, double 2.000000e+00, double* %y, i32 1) #5
+;CHECK-NEXT:  call void @cblas_dscal(i32 2, double 2.000000e+00, double* %"y'", i32 1)
+;CHECK-NEXT:  call void @cblas_dgemv(i32 102, i32 112, i32 2, i32 2, double %alpha, double* %a, i32 2, double* %"y'", i32 1, double 1.000000e+00, double* %"x'", i32 1)
+;CHECK-NEXT:  call void @cblas_dger(i32 102, i32 2, i32 2, double %alpha, double* %"y'", i32 1, double* %"x'", i32 1, double* %a, i32 2)
+;CHECK-NEXT:  call void @cblas_dgemv(i32 102, i32 111, i32 2, i32 2, double 1.000000e+00, double* %a, i32 2, double* %x, i32 1, double 0.000000e+00, double* %0, i32 1)
+;CHECK-NEXT:  %1 = call fast double @cblas_ddot(i32 2, double* nocapture readonly %"y'", i32 1, double* nocapture readonly %0, i32 1)
+;CHECK-NEXT:  %2 = insertvalue { double } undef, double %1, 0
+;CHECK-NEXT:  ret { double } %2
+;CHECK-NEXT:}