summaryrefslogtreecommitdiffstats
path: root/test/CodeGen
diff options
context:
space:
mode:
authorDan Gohman <gohman@apple.com>2009-02-13 03:58:31 +0000
committerDan Gohman <gohman@apple.com>2009-02-13 03:58:31 +0000
commit9194e8b0c827c97f396b209c7f4eeb75759e7068 (patch)
treec282871baf2c2da5cb036e3c626880bc461ab040 /test/CodeGen
parenta200c8faf852f6caf386c592ec54c723d4b86b96 (diff)
downloadexternal_llvm-9194e8b0c827c97f396b209c7f4eeb75759e7068.zip
external_llvm-9194e8b0c827c97f396b209c7f4eeb75759e7068.tar.gz
external_llvm-9194e8b0c827c97f396b209c7f4eeb75759e7068.tar.bz2
Fix the code that checked if a SCEVAddRecExpr Start contains an
addrec in a different loop to check the value being added to the accumulated Start value, not the Start value before it has the new value added to it. This prevents LSR from going crazy on the included testcase. Dale, please review. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@64440 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'test/CodeGen')
-rw-r--r--test/CodeGen/X86/iv-users-in-other-loops.ll292
1 files changed, 292 insertions, 0 deletions
diff --git a/test/CodeGen/X86/iv-users-in-other-loops.ll b/test/CodeGen/X86/iv-users-in-other-loops.ll
new file mode 100644
index 0000000..175f4c0
--- /dev/null
+++ b/test/CodeGen/X86/iv-users-in-other-loops.ll
@@ -0,0 +1,292 @@
+; RUN: llvm-as < %s | llc -march=x86-64 -f -o %t
+; RUN: grep inc %t | count 2
+; RUN: grep addq %t | count 11
+
+; IV users in each of the loops from other loops shouldn't cause LSR
+; to insert new induction variables. Previously it would create a
+; flood of new induction variables.
+
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128"
+target triple = "x86_64-unknown-linux-gnu"
+
+define void @foo(float* %A, i32 %IA, float* %B, i32 %IB, float* nocapture %C, i32 %N) nounwind {
+entry:
+ %0 = xor i32 %IA, 1 ; <i32> [#uses=1]
+ %1 = xor i32 %IB, 1 ; <i32> [#uses=1]
+ %2 = or i32 %1, %0 ; <i32> [#uses=1]
+ %3 = icmp eq i32 %2, 0 ; <i1> [#uses=1]
+ br i1 %3, label %bb2, label %bb13
+
+bb: ; preds = %bb3
+ %4 = load float* %A_addr.0, align 4 ; <float> [#uses=1]
+ %5 = load float* %B_addr.0, align 4 ; <float> [#uses=1]
+ %6 = mul float %4, %5 ; <float> [#uses=1]
+ %7 = add float %6, %Sum0.0 ; <float> [#uses=1]
+ %indvar.next154 = add i64 %B_addr.0.rec, 1 ; <i64> [#uses=1]
+ br label %bb2
+
+bb2: ; preds = %entry, %bb
+ %B_addr.0.rec = phi i64 [ %indvar.next154, %bb ], [ 0, %entry ] ; <i64> [#uses=14]
+ %Sum0.0 = phi float [ %7, %bb ], [ 0.000000e+00, %entry ] ; <float> [#uses=5]
+ %indvar146 = trunc i64 %B_addr.0.rec to i32 ; <i32> [#uses=1]
+ %N_addr.0 = sub i32 %N, %indvar146 ; <i32> [#uses=6]
+ %A_addr.0 = getelementptr float* %A, i64 %B_addr.0.rec ; <float*> [#uses=4]
+ %B_addr.0 = getelementptr float* %B, i64 %B_addr.0.rec ; <float*> [#uses=4]
+ %8 = icmp sgt i32 %N_addr.0, 0 ; <i1> [#uses=1]
+ br i1 %8, label %bb3, label %bb4
+
+bb3: ; preds = %bb2
+ %9 = ptrtoint float* %A_addr.0 to i64 ; <i64> [#uses=1]
+ %10 = and i64 %9, 15 ; <i64> [#uses=1]
+ %11 = icmp eq i64 %10, 0 ; <i1> [#uses=1]
+ br i1 %11, label %bb4, label %bb
+
+bb4: ; preds = %bb3, %bb2
+ %12 = ptrtoint float* %B_addr.0 to i64 ; <i64> [#uses=1]
+ %13 = and i64 %12, 15 ; <i64> [#uses=1]
+ %14 = icmp eq i64 %13, 0 ; <i1> [#uses=1]
+ %15 = icmp sgt i32 %N_addr.0, 15 ; <i1> [#uses=2]
+ br i1 %14, label %bb6.preheader, label %bb10.preheader
+
+bb10.preheader: ; preds = %bb4
+ br i1 %15, label %bb9, label %bb12.loopexit
+
+bb6.preheader: ; preds = %bb4
+ br i1 %15, label %bb5, label %bb8.loopexit
+
+bb5: ; preds = %bb5, %bb6.preheader
+ %indvar143 = phi i64 [ 0, %bb6.preheader ], [ %indvar.next144, %bb5 ] ; <i64> [#uses=3]
+ %vSum0.072 = phi <4 x float> [ zeroinitializer, %bb6.preheader ], [ %21, %bb5 ] ; <<4 x float>> [#uses=1]
+ %vSum1.070 = phi <4 x float> [ zeroinitializer, %bb6.preheader ], [ %29, %bb5 ] ; <<4 x float>> [#uses=1]
+ %vSum2.069 = phi <4 x float> [ zeroinitializer, %bb6.preheader ], [ %37, %bb5 ] ; <<4 x float>> [#uses=1]
+ %vSum3.067 = phi <4 x float> [ zeroinitializer, %bb6.preheader ], [ %45, %bb5 ] ; <<4 x float>> [#uses=1]
+ %indvar145 = trunc i64 %indvar143 to i32 ; <i32> [#uses=1]
+ %tmp150 = mul i32 %indvar145, -16 ; <i32> [#uses=1]
+ %N_addr.268 = add i32 %tmp150, %N_addr.0 ; <i32> [#uses=1]
+ %A_addr.273.rec = shl i64 %indvar143, 4 ; <i64> [#uses=5]
+ %B_addr.0.sum180 = add i64 %B_addr.0.rec, %A_addr.273.rec ; <i64> [#uses=2]
+ %B_addr.271 = getelementptr float* %B, i64 %B_addr.0.sum180 ; <float*> [#uses=1]
+ %A_addr.273 = getelementptr float* %A, i64 %B_addr.0.sum180 ; <float*> [#uses=1]
+ tail call void asm sideeffect ";# foo", "~{dirflag},~{fpsr},~{flags}"() nounwind
+ %16 = bitcast float* %A_addr.273 to <4 x float>* ; <<4 x float>*> [#uses=1]
+ %17 = load <4 x float>* %16, align 16 ; <<4 x float>> [#uses=1]
+ %18 = bitcast float* %B_addr.271 to <4 x float>* ; <<4 x float>*> [#uses=1]
+ %19 = load <4 x float>* %18, align 16 ; <<4 x float>> [#uses=1]
+ %20 = mul <4 x float> %17, %19 ; <<4 x float>> [#uses=1]
+ %21 = add <4 x float> %20, %vSum0.072 ; <<4 x float>> [#uses=2]
+ %A_addr.273.sum163 = or i64 %A_addr.273.rec, 4 ; <i64> [#uses=1]
+ %A_addr.0.sum175 = add i64 %B_addr.0.rec, %A_addr.273.sum163 ; <i64> [#uses=2]
+ %22 = getelementptr float* %A, i64 %A_addr.0.sum175 ; <float*> [#uses=1]
+ %23 = bitcast float* %22 to <4 x float>* ; <<4 x float>*> [#uses=1]
+ %24 = load <4 x float>* %23, align 16 ; <<4 x float>> [#uses=1]
+ %25 = getelementptr float* %B, i64 %A_addr.0.sum175 ; <float*> [#uses=1]
+ %26 = bitcast float* %25 to <4 x float>* ; <<4 x float>*> [#uses=1]
+ %27 = load <4 x float>* %26, align 16 ; <<4 x float>> [#uses=1]
+ %28 = mul <4 x float> %24, %27 ; <<4 x float>> [#uses=1]
+ %29 = add <4 x float> %28, %vSum1.070 ; <<4 x float>> [#uses=2]
+ %A_addr.273.sum161 = or i64 %A_addr.273.rec, 8 ; <i64> [#uses=1]
+ %A_addr.0.sum174 = add i64 %B_addr.0.rec, %A_addr.273.sum161 ; <i64> [#uses=2]
+ %30 = getelementptr float* %A, i64 %A_addr.0.sum174 ; <float*> [#uses=1]
+ %31 = bitcast float* %30 to <4 x float>* ; <<4 x float>*> [#uses=1]
+ %32 = load <4 x float>* %31, align 16 ; <<4 x float>> [#uses=1]
+ %33 = getelementptr float* %B, i64 %A_addr.0.sum174 ; <float*> [#uses=1]
+ %34 = bitcast float* %33 to <4 x float>* ; <<4 x float>*> [#uses=1]
+ %35 = load <4 x float>* %34, align 16 ; <<4 x float>> [#uses=1]
+ %36 = mul <4 x float> %32, %35 ; <<4 x float>> [#uses=1]
+ %37 = add <4 x float> %36, %vSum2.069 ; <<4 x float>> [#uses=2]
+ %A_addr.273.sum159 = or i64 %A_addr.273.rec, 12 ; <i64> [#uses=1]
+ %A_addr.0.sum173 = add i64 %B_addr.0.rec, %A_addr.273.sum159 ; <i64> [#uses=2]
+ %38 = getelementptr float* %A, i64 %A_addr.0.sum173 ; <float*> [#uses=1]
+ %39 = bitcast float* %38 to <4 x float>* ; <<4 x float>*> [#uses=1]
+ %40 = load <4 x float>* %39, align 16 ; <<4 x float>> [#uses=1]
+ %41 = getelementptr float* %B, i64 %A_addr.0.sum173 ; <float*> [#uses=1]
+ %42 = bitcast float* %41 to <4 x float>* ; <<4 x float>*> [#uses=1]
+ %43 = load <4 x float>* %42, align 16 ; <<4 x float>> [#uses=1]
+ %44 = mul <4 x float> %40, %43 ; <<4 x float>> [#uses=1]
+ %45 = add <4 x float> %44, %vSum3.067 ; <<4 x float>> [#uses=2]
+ %.rec83 = add i64 %A_addr.273.rec, 16 ; <i64> [#uses=1]
+ %A_addr.0.sum172 = add i64 %B_addr.0.rec, %.rec83 ; <i64> [#uses=2]
+ %46 = getelementptr float* %A, i64 %A_addr.0.sum172 ; <float*> [#uses=1]
+ %47 = getelementptr float* %B, i64 %A_addr.0.sum172 ; <float*> [#uses=1]
+ %48 = add i32 %N_addr.268, -16 ; <i32> [#uses=2]
+ %49 = icmp sgt i32 %48, 15 ; <i1> [#uses=1]
+ %indvar.next144 = add i64 %indvar143, 1 ; <i64> [#uses=1]
+ br i1 %49, label %bb5, label %bb8.loopexit
+
+bb7: ; preds = %bb7, %bb8.loopexit
+ %indvar130 = phi i64 [ 0, %bb8.loopexit ], [ %indvar.next131, %bb7 ] ; <i64> [#uses=3]
+ %vSum0.260 = phi <4 x float> [ %vSum0.0.lcssa, %bb8.loopexit ], [ %55, %bb7 ] ; <<4 x float>> [#uses=1]
+ %indvar132 = trunc i64 %indvar130 to i32 ; <i32> [#uses=1]
+ %tmp133 = mul i32 %indvar132, -4 ; <i32> [#uses=1]
+ %N_addr.358 = add i32 %tmp133, %N_addr.2.lcssa ; <i32> [#uses=1]
+ %A_addr.361.rec = shl i64 %indvar130, 2 ; <i64> [#uses=3]
+ %B_addr.359 = getelementptr float* %B_addr.2.lcssa, i64 %A_addr.361.rec ; <float*> [#uses=1]
+ %A_addr.361 = getelementptr float* %A_addr.2.lcssa, i64 %A_addr.361.rec ; <float*> [#uses=1]
+ %50 = bitcast float* %A_addr.361 to <4 x float>* ; <<4 x float>*> [#uses=1]
+ %51 = load <4 x float>* %50, align 16 ; <<4 x float>> [#uses=1]
+ %52 = bitcast float* %B_addr.359 to <4 x float>* ; <<4 x float>*> [#uses=1]
+ %53 = load <4 x float>* %52, align 16 ; <<4 x float>> [#uses=1]
+ %54 = mul <4 x float> %51, %53 ; <<4 x float>> [#uses=1]
+ %55 = add <4 x float> %54, %vSum0.260 ; <<4 x float>> [#uses=2]
+ %.rec85 = add i64 %A_addr.361.rec, 4 ; <i64> [#uses=2]
+ %56 = getelementptr float* %A_addr.2.lcssa, i64 %.rec85 ; <float*> [#uses=1]
+ %57 = getelementptr float* %B_addr.2.lcssa, i64 %.rec85 ; <float*> [#uses=1]
+ %58 = add i32 %N_addr.358, -4 ; <i32> [#uses=2]
+ %59 = icmp sgt i32 %58, 3 ; <i1> [#uses=1]
+ %indvar.next131 = add i64 %indvar130, 1 ; <i64> [#uses=1]
+ br i1 %59, label %bb7, label %bb13
+
+bb8.loopexit: ; preds = %bb5, %bb6.preheader
+ %A_addr.2.lcssa = phi float* [ %A_addr.0, %bb6.preheader ], [ %46, %bb5 ] ; <float*> [#uses=3]
+ %vSum0.0.lcssa = phi <4 x float> [ zeroinitializer, %bb6.preheader ], [ %21, %bb5 ] ; <<4 x float>> [#uses=2]
+ %B_addr.2.lcssa = phi float* [ %B_addr.0, %bb6.preheader ], [ %47, %bb5 ] ; <float*> [#uses=3]
+ %vSum1.0.lcssa = phi <4 x float> [ zeroinitializer, %bb6.preheader ], [ %29, %bb5 ] ; <<4 x float>> [#uses=2]
+ %vSum2.0.lcssa = phi <4 x float> [ zeroinitializer, %bb6.preheader ], [ %37, %bb5 ] ; <<4 x float>> [#uses=2]
+ %N_addr.2.lcssa = phi i32 [ %N_addr.0, %bb6.preheader ], [ %48, %bb5 ] ; <i32> [#uses=3]
+ %vSum3.0.lcssa = phi <4 x float> [ zeroinitializer, %bb6.preheader ], [ %45, %bb5 ] ; <<4 x float>> [#uses=2]
+ %60 = icmp sgt i32 %N_addr.2.lcssa, 3 ; <i1> [#uses=1]
+ br i1 %60, label %bb7, label %bb13
+
+bb9: ; preds = %bb9, %bb10.preheader
+ %indvar106 = phi i64 [ 0, %bb10.preheader ], [ %indvar.next107, %bb9 ] ; <i64> [#uses=3]
+ %vSum0.339 = phi <4 x float> [ zeroinitializer, %bb10.preheader ], [ %75, %bb9 ] ; <<4 x float>> [#uses=1]
+ %vSum1.237 = phi <4 x float> [ zeroinitializer, %bb10.preheader ], [ %80, %bb9 ] ; <<4 x float>> [#uses=1]
+ %vSum2.236 = phi <4 x float> [ zeroinitializer, %bb10.preheader ], [ %85, %bb9 ] ; <<4 x float>> [#uses=1]
+ %vSum3.234 = phi <4 x float> [ zeroinitializer, %bb10.preheader ], [ %90, %bb9 ] ; <<4 x float>> [#uses=1]
+ %indvar108 = trunc i64 %indvar106 to i32 ; <i32> [#uses=1]
+ %tmp113 = mul i32 %indvar108, -16 ; <i32> [#uses=1]
+ %N_addr.435 = add i32 %tmp113, %N_addr.0 ; <i32> [#uses=1]
+ %A_addr.440.rec = shl i64 %indvar106, 4 ; <i64> [#uses=5]
+ %B_addr.0.sum = add i64 %B_addr.0.rec, %A_addr.440.rec ; <i64> [#uses=2]
+ %B_addr.438 = getelementptr float* %B, i64 %B_addr.0.sum ; <float*> [#uses=1]
+ %A_addr.440 = getelementptr float* %A, i64 %B_addr.0.sum ; <float*> [#uses=1]
+ %61 = bitcast float* %B_addr.438 to i8* ; <i8*> [#uses=1]
+ %62 = tail call <4 x float> @llvm.x86.sse.loadu.ps(i8* %61) nounwind readonly ; <<4 x float>> [#uses=1]
+ %B_addr.438.sum169 = or i64 %A_addr.440.rec, 4 ; <i64> [#uses=1]
+ %B_addr.0.sum187 = add i64 %B_addr.0.rec, %B_addr.438.sum169 ; <i64> [#uses=2]
+ %63 = getelementptr float* %B, i64 %B_addr.0.sum187 ; <float*> [#uses=1]
+ %64 = bitcast float* %63 to i8* ; <i8*> [#uses=1]
+ %65 = tail call <4 x float> @llvm.x86.sse.loadu.ps(i8* %64) nounwind readonly ; <<4 x float>> [#uses=1]
+ %B_addr.438.sum168 = or i64 %A_addr.440.rec, 8 ; <i64> [#uses=1]
+ %B_addr.0.sum186 = add i64 %B_addr.0.rec, %B_addr.438.sum168 ; <i64> [#uses=2]
+ %66 = getelementptr float* %B, i64 %B_addr.0.sum186 ; <float*> [#uses=1]
+ %67 = bitcast float* %66 to i8* ; <i8*> [#uses=1]
+ %68 = tail call <4 x float> @llvm.x86.sse.loadu.ps(i8* %67) nounwind readonly ; <<4 x float>> [#uses=1]
+ %B_addr.438.sum167 = or i64 %A_addr.440.rec, 12 ; <i64> [#uses=1]
+ %B_addr.0.sum185 = add i64 %B_addr.0.rec, %B_addr.438.sum167 ; <i64> [#uses=2]
+ %69 = getelementptr float* %B, i64 %B_addr.0.sum185 ; <float*> [#uses=1]
+ %70 = bitcast float* %69 to i8* ; <i8*> [#uses=1]
+ %71 = tail call <4 x float> @llvm.x86.sse.loadu.ps(i8* %70) nounwind readonly ; <<4 x float>> [#uses=1]
+ %72 = bitcast float* %A_addr.440 to <4 x float>* ; <<4 x float>*> [#uses=1]
+ %73 = load <4 x float>* %72, align 16 ; <<4 x float>> [#uses=1]
+ %74 = mul <4 x float> %73, %62 ; <<4 x float>> [#uses=1]
+ %75 = add <4 x float> %74, %vSum0.339 ; <<4 x float>> [#uses=2]
+ %76 = getelementptr float* %A, i64 %B_addr.0.sum187 ; <float*> [#uses=1]
+ %77 = bitcast float* %76 to <4 x float>* ; <<4 x float>*> [#uses=1]
+ %78 = load <4 x float>* %77, align 16 ; <<4 x float>> [#uses=1]
+ %79 = mul <4 x float> %78, %65 ; <<4 x float>> [#uses=1]
+ %80 = add <4 x float> %79, %vSum1.237 ; <<4 x float>> [#uses=2]
+ %81 = getelementptr float* %A, i64 %B_addr.0.sum186 ; <float*> [#uses=1]
+ %82 = bitcast float* %81 to <4 x float>* ; <<4 x float>*> [#uses=1]
+ %83 = load <4 x float>* %82, align 16 ; <<4 x float>> [#uses=1]
+ %84 = mul <4 x float> %83, %68 ; <<4 x float>> [#uses=1]
+ %85 = add <4 x float> %84, %vSum2.236 ; <<4 x float>> [#uses=2]
+ %86 = getelementptr float* %A, i64 %B_addr.0.sum185 ; <float*> [#uses=1]
+ %87 = bitcast float* %86 to <4 x float>* ; <<4 x float>*> [#uses=1]
+ %88 = load <4 x float>* %87, align 16 ; <<4 x float>> [#uses=1]
+ %89 = mul <4 x float> %88, %71 ; <<4 x float>> [#uses=1]
+ %90 = add <4 x float> %89, %vSum3.234 ; <<4 x float>> [#uses=2]
+ %.rec89 = add i64 %A_addr.440.rec, 16 ; <i64> [#uses=1]
+ %A_addr.0.sum170 = add i64 %B_addr.0.rec, %.rec89 ; <i64> [#uses=2]
+ %91 = getelementptr float* %A, i64 %A_addr.0.sum170 ; <float*> [#uses=1]
+ %92 = getelementptr float* %B, i64 %A_addr.0.sum170 ; <float*> [#uses=1]
+ %93 = add i32 %N_addr.435, -16 ; <i32> [#uses=2]
+ %94 = icmp sgt i32 %93, 15 ; <i1> [#uses=1]
+ %indvar.next107 = add i64 %indvar106, 1 ; <i64> [#uses=1]
+ br i1 %94, label %bb9, label %bb12.loopexit
+
+bb11: ; preds = %bb11, %bb12.loopexit
+ %indvar = phi i64 [ 0, %bb12.loopexit ], [ %indvar.next, %bb11 ] ; <i64> [#uses=3]
+ %vSum0.428 = phi <4 x float> [ %vSum0.3.lcssa, %bb12.loopexit ], [ %100, %bb11 ] ; <<4 x float>> [#uses=1]
+ %indvar96 = trunc i64 %indvar to i32 ; <i32> [#uses=1]
+ %tmp = mul i32 %indvar96, -4 ; <i32> [#uses=1]
+ %N_addr.526 = add i32 %tmp, %N_addr.4.lcssa ; <i32> [#uses=1]
+ %A_addr.529.rec = shl i64 %indvar, 2 ; <i64> [#uses=3]
+ %B_addr.527 = getelementptr float* %B_addr.4.lcssa, i64 %A_addr.529.rec ; <float*> [#uses=1]
+ %A_addr.529 = getelementptr float* %A_addr.4.lcssa, i64 %A_addr.529.rec ; <float*> [#uses=1]
+ %95 = bitcast float* %B_addr.527 to i8* ; <i8*> [#uses=1]
+ %96 = tail call <4 x float> @llvm.x86.sse.loadu.ps(i8* %95) nounwind readonly ; <<4 x float>> [#uses=1]
+ %97 = bitcast float* %A_addr.529 to <4 x float>* ; <<4 x float>*> [#uses=1]
+ %98 = load <4 x float>* %97, align 16 ; <<4 x float>> [#uses=1]
+ %99 = mul <4 x float> %98, %96 ; <<4 x float>> [#uses=1]
+ %100 = add <4 x float> %99, %vSum0.428 ; <<4 x float>> [#uses=2]
+ %.rec91 = add i64 %A_addr.529.rec, 4 ; <i64> [#uses=2]
+ %101 = getelementptr float* %A_addr.4.lcssa, i64 %.rec91 ; <float*> [#uses=1]
+ %102 = getelementptr float* %B_addr.4.lcssa, i64 %.rec91 ; <float*> [#uses=1]
+ %103 = add i32 %N_addr.526, -4 ; <i32> [#uses=2]
+ %104 = icmp sgt i32 %103, 3 ; <i1> [#uses=1]
+ %indvar.next = add i64 %indvar, 1 ; <i64> [#uses=1]
+ br i1 %104, label %bb11, label %bb13
+
+bb12.loopexit: ; preds = %bb9, %bb10.preheader
+ %A_addr.4.lcssa = phi float* [ %A_addr.0, %bb10.preheader ], [ %91, %bb9 ] ; <float*> [#uses=3]
+ %vSum0.3.lcssa = phi <4 x float> [ zeroinitializer, %bb10.preheader ], [ %75, %bb9 ] ; <<4 x float>> [#uses=2]
+ %B_addr.4.lcssa = phi float* [ %B_addr.0, %bb10.preheader ], [ %92, %bb9 ] ; <float*> [#uses=3]
+ %vSum1.2.lcssa = phi <4 x float> [ zeroinitializer, %bb10.preheader ], [ %80, %bb9 ] ; <<4 x float>> [#uses=2]
+ %vSum2.2.lcssa = phi <4 x float> [ zeroinitializer, %bb10.preheader ], [ %85, %bb9 ] ; <<4 x float>> [#uses=2]
+ %N_addr.4.lcssa = phi i32 [ %N_addr.0, %bb10.preheader ], [ %93, %bb9 ] ; <i32> [#uses=3]
+ %vSum3.2.lcssa = phi <4 x float> [ zeroinitializer, %bb10.preheader ], [ %90, %bb9 ] ; <<4 x float>> [#uses=2]
+ %105 = icmp sgt i32 %N_addr.4.lcssa, 3 ; <i1> [#uses=1]
+ br i1 %105, label %bb11, label %bb13
+
+bb13: ; preds = %bb12.loopexit, %bb11, %bb8.loopexit, %bb7, %entry
+ %Sum0.1 = phi float [ 0.000000e+00, %entry ], [ %Sum0.0, %bb7 ], [ %Sum0.0, %bb8.loopexit ], [ %Sum0.0, %bb11 ], [ %Sum0.0, %bb12.loopexit ] ; <float> [#uses=1]
+ %vSum3.1 = phi <4 x float> [ zeroinitializer, %entry ], [ %vSum3.0.lcssa, %bb7 ], [ %vSum3.0.lcssa, %bb8.loopexit ], [ %vSum3.2.lcssa, %bb11 ], [ %vSum3.2.lcssa, %bb12.loopexit ] ; <<4 x float>> [#uses=1]
+ %N_addr.1 = phi i32 [ %N, %entry ], [ %N_addr.2.lcssa, %bb8.loopexit ], [ %58, %bb7 ], [ %N_addr.4.lcssa, %bb12.loopexit ], [ %103, %bb11 ] ; <i32> [#uses=2]
+ %vSum2.1 = phi <4 x float> [ zeroinitializer, %entry ], [ %vSum2.0.lcssa, %bb7 ], [ %vSum2.0.lcssa, %bb8.loopexit ], [ %vSum2.2.lcssa, %bb11 ], [ %vSum2.2.lcssa, %bb12.loopexit ] ; <<4 x float>> [#uses=1]
+ %vSum1.1 = phi <4 x float> [ zeroinitializer, %entry ], [ %vSum1.0.lcssa, %bb7 ], [ %vSum1.0.lcssa, %bb8.loopexit ], [ %vSum1.2.lcssa, %bb11 ], [ %vSum1.2.lcssa, %bb12.loopexit ] ; <<4 x float>> [#uses=1]
+ %B_addr.1 = phi float* [ %B, %entry ], [ %B_addr.2.lcssa, %bb8.loopexit ], [ %57, %bb7 ], [ %B_addr.4.lcssa, %bb12.loopexit ], [ %102, %bb11 ] ; <float*> [#uses=1]
+ %vSum0.1 = phi <4 x float> [ zeroinitializer, %entry ], [ %vSum0.0.lcssa, %bb8.loopexit ], [ %55, %bb7 ], [ %vSum0.3.lcssa, %bb12.loopexit ], [ %100, %bb11 ] ; <<4 x float>> [#uses=1]
+ %A_addr.1 = phi float* [ %A, %entry ], [ %A_addr.2.lcssa, %bb8.loopexit ], [ %56, %bb7 ], [ %A_addr.4.lcssa, %bb12.loopexit ], [ %101, %bb11 ] ; <float*> [#uses=1]
+ %106 = add <4 x float> %vSum0.1, %vSum2.1 ; <<4 x float>> [#uses=1]
+ %107 = add <4 x float> %vSum1.1, %vSum3.1 ; <<4 x float>> [#uses=1]
+ %108 = add <4 x float> %106, %107 ; <<4 x float>> [#uses=4]
+ %tmp23 = extractelement <4 x float> %108, i32 0 ; <float> [#uses=1]
+ %tmp21 = extractelement <4 x float> %108, i32 1 ; <float> [#uses=1]
+ %109 = add float %tmp23, %tmp21 ; <float> [#uses=1]
+ %tmp19 = extractelement <4 x float> %108, i32 2 ; <float> [#uses=1]
+ %tmp17 = extractelement <4 x float> %108, i32 3 ; <float> [#uses=1]
+ %110 = add float %tmp19, %tmp17 ; <float> [#uses=1]
+ %111 = add float %109, %110 ; <float> [#uses=1]
+ %Sum0.254 = add float %111, %Sum0.1 ; <float> [#uses=2]
+ %112 = icmp sgt i32 %N_addr.1, 0 ; <i1> [#uses=1]
+ br i1 %112, label %bb.nph56, label %bb16
+
+bb.nph56: ; preds = %bb13
+ %tmp. = zext i32 %N_addr.1 to i64 ; <i64> [#uses=1]
+ br label %bb14
+
+bb14: ; preds = %bb14, %bb.nph56
+ %indvar117 = phi i64 [ 0, %bb.nph56 ], [ %indvar.next118, %bb14 ] ; <i64> [#uses=3]
+ %Sum0.255 = phi float [ %Sum0.254, %bb.nph56 ], [ %Sum0.2, %bb14 ] ; <float> [#uses=1]
+ %tmp.122 = sext i32 %IB to i64 ; <i64> [#uses=1]
+ %B_addr.652.rec = mul i64 %indvar117, %tmp.122 ; <i64> [#uses=1]
+ %tmp.124 = sext i32 %IA to i64 ; <i64> [#uses=1]
+ %A_addr.653.rec = mul i64 %indvar117, %tmp.124 ; <i64> [#uses=1]
+ %B_addr.652 = getelementptr float* %B_addr.1, i64 %B_addr.652.rec ; <float*> [#uses=1]
+ %A_addr.653 = getelementptr float* %A_addr.1, i64 %A_addr.653.rec ; <float*> [#uses=1]
+ %113 = load float* %A_addr.653, align 4 ; <float> [#uses=1]
+ %114 = load float* %B_addr.652, align 4 ; <float> [#uses=1]
+ %115 = mul float %113, %114 ; <float> [#uses=1]
+ %Sum0.2 = add float %115, %Sum0.255 ; <float> [#uses=2]
+ %indvar.next118 = add i64 %indvar117, 1 ; <i64> [#uses=2]
+ %exitcond = icmp eq i64 %indvar.next118, %tmp. ; <i1> [#uses=1]
+ br i1 %exitcond, label %bb16, label %bb14
+
+bb16: ; preds = %bb14, %bb13
+ %Sum0.2.lcssa = phi float [ %Sum0.254, %bb13 ], [ %Sum0.2, %bb14 ] ; <float> [#uses=1]
+ store float %Sum0.2.lcssa, float* %C, align 4
+ ret void
+}
+
+declare <4 x float> @llvm.x86.sse.loadu.ps(i8*) nounwind readonly