Vendor import of llvm trunk r161861:

http://llvm.org/svn/llvm-project/llvm/trunk@161861

10 files changed, 296 insertions, 8 deletions

diff --git a/test/Transforms/Inline/2007-04-15-InlineEH.ll b/test/Transforms/Inline/2007-04-15-InlineEH.ll
index 8fbcf92..b114537 100644
--- a/test/Transforms/Inline/2007-04-15-InlineEH.ll
+++ b/test/Transforms/Inline/2007-04-15-InlineEH.ll
@@ -1,4 +1,4 @@
-; RUN: opt < %s -inline -S | not grep {invoke void asm}
+; RUN: opt < %s -inline -S | not grep "invoke void asm"
 ; PR1335
 
 target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64"
diff --git a/test/Transforms/Inline/casts.ll b/test/Transforms/Inline/casts.ll
index 166185a..a7b051b 100644
--- a/test/Transforms/Inline/casts.ll
+++ b/test/Transforms/Inline/casts.ll
@@ -1,4 +1,4 @@
-; RUN: opt < %s -inline -S | grep {ret i32 1}
+; RUN: opt < %s -inline -S | grep "ret i32 1"
 ; ModuleID = 'short.opt.bc'
 
 define i32 @testBool(i1 %X) {
diff --git a/test/Transforms/Inline/delete-call.ll b/test/Transforms/Inline/delete-call.ll
index 3505608..7716d6a 100644
--- a/test/Transforms/Inline/delete-call.ll
+++ b/test/Transforms/Inline/delete-call.ll
@@ -1,5 +1,5 @@
-; RUN: opt %s -S  -inline -functionattrs -stats |& grep {Number of call sites deleted, not inlined}
-; RUN: opt %s -S  -inline -stats |& grep {Number of functions inlined}
+; RUN: opt %s -S  -inline -functionattrs -stats 2>&1 | grep "Number of call sites deleted, not inlined"
+; RUN: opt %s -S  -inline -stats 2>&1 | grep "Number of functions inlined"
 
 target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128-n8:16:32"
 target triple = "i386-apple-darwin9.8"
diff --git a/test/Transforms/Inline/externally_available.ll b/test/Transforms/Inline/externally_available.ll
index 08b5638..07274e7 100644
--- a/test/Transforms/Inline/externally_available.ll
+++ b/test/Transforms/Inline/externally_available.ll
@@ -1,6 +1,6 @@
 ; RUN: opt < %s -inline -constprop -S > %t
 ; RUN: not grep test_function %t
-; RUN: grep {ret i32 5} %t
+; RUN: grep "ret i32 5" %t
 
 
 ; test_function should not be emitted to the .s file.
diff --git a/test/Transforms/Inline/inline-byval-bonus.ll b/test/Transforms/Inline/inline-byval-bonus.ll
new file mode 100644
index 0000000..f3ed819
--- /dev/null
+++ b/test/Transforms/Inline/inline-byval-bonus.ll
@@ -0,0 +1,193 @@
+; RUN: opt -S -inline -inline-threshold=275 < %s | FileCheck %s
+; PR13095
+
+; The performance of the c-ray benchmark largely depends on the inlining of a
+; specific call to @ray_sphere. This test case is designed to verify that it's
+; inlined at -O3.
+
+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-n8:16:32:64-S128"
+target triple = "x86_64-apple-macosx10.8.0"
+
+%struct.sphere = type { %struct.vec3, double, %struct.material, %struct.sphere* }
+%struct.vec3 = type { double, double, double }
+%struct.material = type { %struct.vec3, double, double }
+%struct.ray = type { %struct.vec3, %struct.vec3 }
+%struct.spoint = type { %struct.vec3, %struct.vec3, %struct.vec3, double }
+
+define i32 @caller(%struct.sphere* %i) {
+  %shadow_ray = alloca %struct.ray, align 8
+  call void @fix(%struct.ray* %shadow_ray)
+
+  %call = call i32 @ray_sphere(%struct.sphere* %i, %struct.ray* byval align 8 %shadow_ray, %struct.spoint* null)
+  ret i32 %call
+
+; CHECK: @caller
+; CHECK-NOT: call i32 @ray_sphere
+; CHECK: ret i32
+}
+
+declare void @fix(%struct.ray*)
+
+define i32 @ray_sphere(%struct.sphere* nocapture %sph, %struct.ray* nocapture byval align 8 %ray, %struct.spoint* %sp) nounwind uwtable ssp {
+  %1 = getelementptr inbounds %struct.ray* %ray, i64 0, i32 1, i32 0
+  %2 = load double* %1, align 8
+  %3 = fmul double %2, %2
+  %4 = getelementptr inbounds %struct.ray* %ray, i64 0, i32 1, i32 1
+  %5 = load double* %4, align 8
+  %6 = fmul double %5, %5
+  %7 = fadd double %3, %6
+  %8 = getelementptr inbounds %struct.ray* %ray, i64 0, i32 1, i32 2
+  %9 = load double* %8, align 8
+  %10 = fmul double %9, %9
+  %11 = fadd double %7, %10
+  %12 = fmul double %2, 2.000000e+00
+  %13 = getelementptr inbounds %struct.ray* %ray, i64 0, i32 0, i32 0
+  %14 = load double* %13, align 8
+  %15 = getelementptr inbounds %struct.sphere* %sph, i64 0, i32 0, i32 0
+  %16 = load double* %15, align 8
+  %17 = fsub double %14, %16
+  %18 = fmul double %12, %17
+  %19 = fmul double %5, 2.000000e+00
+  %20 = getelementptr inbounds %struct.ray* %ray, i64 0, i32 0, i32 1
+  %21 = load double* %20, align 8
+  %22 = getelementptr inbounds %struct.sphere* %sph, i64 0, i32 0, i32 1
+  %23 = load double* %22, align 8
+  %24 = fsub double %21, %23
+  %25 = fmul double %19, %24
+  %26 = fadd double %18, %25
+  %27 = fmul double %9, 2.000000e+00
+  %28 = getelementptr inbounds %struct.ray* %ray, i64 0, i32 0, i32 2
+  %29 = load double* %28, align 8
+  %30 = getelementptr inbounds %struct.sphere* %sph, i64 0, i32 0, i32 2
+  %31 = load double* %30, align 8
+  %32 = fsub double %29, %31
+  %33 = fmul double %27, %32
+  %34 = fadd double %26, %33
+  %35 = fmul double %16, %16
+  %36 = fmul double %23, %23
+  %37 = fadd double %35, %36
+  %38 = fmul double %31, %31
+  %39 = fadd double %37, %38
+  %40 = fmul double %14, %14
+  %41 = fadd double %40, %39
+  %42 = fmul double %21, %21
+  %43 = fadd double %42, %41
+  %44 = fmul double %29, %29
+  %45 = fadd double %44, %43
+  %46 = fsub double -0.000000e+00, %16
+  %47 = fmul double %14, %46
+  %48 = fmul double %21, %23
+  %49 = fsub double %47, %48
+  %50 = fmul double %29, %31
+  %51 = fsub double %49, %50
+  %52 = fmul double %51, 2.000000e+00
+  %53 = fadd double %52, %45
+  %54 = getelementptr inbounds %struct.sphere* %sph, i64 0, i32 1
+  %55 = load double* %54, align 8
+  %56 = fmul double %55, %55
+  %57 = fsub double %53, %56
+  %58 = fmul double %34, %34
+  %59 = fmul double %11, 4.000000e+00
+  %60 = fmul double %59, %57
+  %61 = fsub double %58, %60
+  %62 = fcmp olt double %61, 0.000000e+00
+  br i1 %62, label %130, label %63
+
+; <label>:63                                      ; preds = %0
+  %64 = tail call double @sqrt(double %61) nounwind readnone
+  %65 = fsub double -0.000000e+00, %34
+  %66 = fsub double %64, %34
+  %67 = fmul double %11, 2.000000e+00
+  %68 = fdiv double %66, %67
+  %69 = fsub double %65, %64
+  %70 = fdiv double %69, %67
+  %71 = fcmp olt double %68, 1.000000e-06
+  %72 = fcmp olt double %70, 1.000000e-06
+  %or.cond = and i1 %71, %72
+  br i1 %or.cond, label %130, label %73
+
+; <label>:73                                      ; preds = %63
+  %74 = fcmp ogt double %68, 1.000000e+00
+  %75 = fcmp ogt double %70, 1.000000e+00
+  %or.cond1 = and i1 %74, %75
+  br i1 %or.cond1, label %130, label %76
+
+; <label>:76                                      ; preds = %73
+  %77 = icmp eq %struct.spoint* %sp, null
+  br i1 %77, label %130, label %78
+
+; <label>:78                                      ; preds = %76
+  %t1.0 = select i1 %71, double %70, double %68
+  %t2.0 = select i1 %72, double %t1.0, double %70
+  %79 = fcmp olt double %t1.0, %t2.0
+  %80 = select i1 %79, double %t1.0, double %t2.0
+  %81 = getelementptr inbounds %struct.spoint* %sp, i64 0, i32 3
+  store double %80, double* %81, align 8
+  %82 = fmul double %80, %2
+  %83 = fadd double %14, %82
+  %84 = getelementptr inbounds %struct.spoint* %sp, i64 0, i32 0, i32 0
+  store double %83, double* %84, align 8
+  %85 = fmul double %5, %80
+  %86 = fadd double %21, %85
+  %87 = getelementptr inbounds %struct.spoint* %sp, i64 0, i32 0, i32 1
+  store double %86, double* %87, align 8
+  %88 = fmul double %9, %80
+  %89 = fadd double %29, %88
+  %90 = getelementptr inbounds %struct.spoint* %sp, i64 0, i32 0, i32 2
+  store double %89, double* %90, align 8
+  %91 = load double* %15, align 8
+  %92 = fsub double %83, %91
+  %93 = load double* %54, align 8
+  %94 = fdiv double %92, %93
+  %95 = getelementptr inbounds %struct.spoint* %sp, i64 0, i32 1, i32 0
+  store double %94, double* %95, align 8
+  %96 = load double* %22, align 8
+  %97 = fsub double %86, %96
+  %98 = load double* %54, align 8
+  %99 = fdiv double %97, %98
+  %100 = getelementptr inbounds %struct.spoint* %sp, i64 0, i32 1, i32 1
+  store double %99, double* %100, align 8
+  %101 = load double* %30, align 8
+  %102 = fsub double %89, %101
+  %103 = load double* %54, align 8
+  %104 = fdiv double %102, %103
+  %105 = getelementptr inbounds %struct.spoint* %sp, i64 0, i32 1, i32 2
+  store double %104, double* %105, align 8
+  %106 = fmul double %2, %94
+  %107 = fmul double %5, %99
+  %108 = fadd double %106, %107
+  %109 = fmul double %9, %104
+  %110 = fadd double %108, %109
+  %111 = fmul double %110, 2.000000e+00
+  %112 = fmul double %94, %111
+  %113 = fsub double %112, %2
+  %114 = fsub double -0.000000e+00, %113
+  %115 = fmul double %99, %111
+  %116 = fsub double %115, %5
+  %117 = fsub double -0.000000e+00, %116
+  %118 = fmul double %104, %111
+  %119 = fsub double %118, %9
+  %120 = fsub double -0.000000e+00, %119
+  %.06 = getelementptr inbounds %struct.spoint* %sp, i64 0, i32 2, i32 0
+  %.18 = getelementptr inbounds %struct.spoint* %sp, i64 0, i32 2, i32 1
+  %.210 = getelementptr inbounds %struct.spoint* %sp, i64 0, i32 2, i32 2
+  %121 = fmul double %113, %113
+  %122 = fmul double %116, %116
+  %123 = fadd double %121, %122
+  %124 = fmul double %119, %119
+  %125 = fadd double %123, %124
+  %126 = tail call double @sqrt(double %125) nounwind readnone
+  %127 = fdiv double %114, %126
+  store double %127, double* %.06, align 8
+  %128 = fdiv double %117, %126
+  store double %128, double* %.18, align 8
+  %129 = fdiv double %120, %126
+  store double %129, double* %.210, align 8
+  br label %130
+
+; <label>:130                                     ; preds = %78, %76, %73, %63, %0
+  %.0 = phi i32 [ 0, %0 ], [ 0, %73 ], [ 0, %63 ], [ 1, %76 ], [ 1, %78 ]
+  ret i32 %.0
+}
+
+declare double @sqrt(double) nounwind readnone
diff --git a/test/Transforms/Inline/inline-invoke-tail.ll b/test/Transforms/Inline/inline-invoke-tail.ll
index 1f34113..e077523 100644
--- a/test/Transforms/Inline/inline-invoke-tail.ll
+++ b/test/Transforms/Inline/inline-invoke-tail.ll
@@ -1,4 +1,4 @@
-; RUN: opt < %s -inline -S | not grep {tail call void @llvm.memcpy.i32}
+; RUN: opt < %s -inline -S | not grep "tail call void @llvm.memcpy.i32"
 ; PR3550
 
 define internal void @foo(i32* %p, i32* %q) {
diff --git a/test/Transforms/Inline/inline-optsize.ll b/test/Transforms/Inline/inline-optsize.ll
new file mode 100644
index 0000000..20d7426
--- /dev/null
+++ b/test/Transforms/Inline/inline-optsize.ll
@@ -0,0 +1,33 @@
+; RUN: opt -S -Oz %s | FileCheck %s -check-prefix=OZ
+; RUN: opt -S -O2 %s | FileCheck %s -check-prefix=O2
+
+; The inline threshold for a function with the optsize attribute is currently
+; the same as the global inline threshold for -Os. Check that the optsize
+; function attribute don't alter the function specific inline threshold if the
+; global inline threshold is lower (as for -Oz).
+
+@a = global i32 4
+
+; This function should be larger than the inline threshold for -Oz (25), but
+; smaller than the inline threshold for optsize (75).
+define i32 @inner() {
+  %a1 = load volatile i32* @a
+  %x1 = add i32 %a1,  %a1
+  %a2 = load volatile i32* @a
+  %x2 = add i32 %x1, %a2
+  %a3 = load volatile i32* @a
+  %x3 = add i32 %x2, %a3
+  %a4 = load volatile i32* @a
+  %x4 = add i32 %x3, %a4
+  %a5 = load volatile i32* @a
+  %x5 = add i32 %x3, %a5
+  ret i32 %x5
+}
+
+; @inner() should be inlined for -O2 but not for -Oz.
+; OZ: call
+; O2-NOT: call
+define i32 @outer() optsize {
+   %r = call i32 @inner()
+   ret i32 %r
+}
diff --git a/test/Transforms/Inline/inline_constprop.ll b/test/Transforms/Inline/inline_constprop.ll
index dc35b60..0b48a72 100644
--- a/test/Transforms/Inline/inline_constprop.ll
+++ b/test/Transforms/Inline/inline_constprop.ll
@@ -110,3 +110,65 @@ bb.merge:
 bb.false:
   ret i32 %sub
 }
+
+
+define i32 @PR13412.main() {
+; This is a somewhat complicated three layer subprogram that was reported to
+; compute the wrong value for a branch due to assuming that an argument
+; mid-inline couldn't be equal to another pointer.
+;
+; After inlining, the branch should point directly to the exit block, not to
+; the intermediate block.
+; CHECK: @PR13412.main
+; CHECK: br i1 true, label %[[TRUE_DEST:.*]], label %[[FALSE_DEST:.*]]
+; CHECK: [[FALSE_DEST]]:
+; CHECK-NEXT: call void @PR13412.fail()
+; CHECK: [[TRUE_DEST]]:
+; CHECK-NEXT: ret i32 0
+
+entry:
+  %i1 = alloca i64
+  store i64 0, i64* %i1
+  %arraydecay = bitcast i64* %i1 to i32*
+  %call = call i1 @PR13412.first(i32* %arraydecay, i32* %arraydecay)
+  br i1 %call, label %cond.end, label %cond.false
+
+cond.false:
+  call void @PR13412.fail()
+  br label %cond.end
+
+cond.end:
+  ret i32 0
+}
+
+define internal i1 @PR13412.first(i32* %a, i32* %b) {
+entry:
+  %call = call i32* @PR13412.second(i32* %a, i32* %b)
+  %cmp = icmp eq i32* %call, %b
+  ret i1 %cmp
+}
+
+declare void @PR13412.fail()
+
+define internal i32* @PR13412.second(i32* %a, i32* %b) {
+entry:
+  %sub.ptr.lhs.cast = ptrtoint i32* %b to i64
+  %sub.ptr.rhs.cast = ptrtoint i32* %a to i64
+  %sub.ptr.sub = sub i64 %sub.ptr.lhs.cast, %sub.ptr.rhs.cast
+  %sub.ptr.div = ashr exact i64 %sub.ptr.sub, 2
+  %cmp = icmp ugt i64 %sub.ptr.div, 1
+  br i1 %cmp, label %if.then, label %if.end3
+
+if.then:
+  %0 = load i32* %a
+  %1 = load i32* %b
+  %cmp1 = icmp eq i32 %0, %1
+  br i1 %cmp1, label %return, label %if.end3
+
+if.end3:
+  br label %return
+
+return:
+  %retval.0 = phi i32* [ %b, %if.end3 ], [ %a, %if.then ]
+  ret i32* %retval.0
+}
diff --git a/test/Transforms/Inline/inline_prune.ll b/test/Transforms/Inline/inline_prune.ll
index 658a422..4c1574d 100644
--- a/test/Transforms/Inline/inline_prune.ll
+++ b/test/Transforms/Inline/inline_prune.ll
@@ -1,5 +1,5 @@
 ; RUN: opt < %s -inline -S | \
-; RUN:    not grep {callee\[12\](}
+; RUN:    not grep "callee[12]("
 ; RUN: opt < %s -inline -S | not grep mul
 
 define internal i32 @callee1(i32 %A, i32 %B) {
diff --git a/test/Transforms/Inline/invoke_test-1.ll b/test/Transforms/Inline/invoke_test-1.ll
index e0e6d60..922351f 100644
--- a/test/Transforms/Inline/invoke_test-1.ll
+++ b/test/Transforms/Inline/invoke_test-1.ll
@@ -2,7 +2,7 @@
 ; instructions
 
 ; RUN: opt < %s -inline -S | \
-; RUN:   not grep {call\[^e\]}
+; RUN:   not grep "call[^e]"
 
 declare void @might_throw()