diff options
Diffstat (limited to 'test/Transforms/InstCombine')
40 files changed, 1007 insertions, 83 deletions
diff --git a/test/Transforms/InstCombine/2008-05-08-StrLenSink.ll b/test/Transforms/InstCombine/2008-05-08-StrLenSink.ll index 1da28562aa..d266164fd8 100644 --- a/test/Transforms/InstCombine/2008-05-08-StrLenSink.ll +++ b/test/Transforms/InstCombine/2008-05-08-StrLenSink.ll @@ -1,4 +1,4 @@ -; RUN: opt -S -instcombine %s | FileCheck %s +; RUN: opt -S -instcombine < %s | FileCheck %s ; PR2297 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" target triple = "i386-apple-darwin8" diff --git a/test/Transforms/InstCombine/2010-03-03-ExtElim.ll b/test/Transforms/InstCombine/2010-03-03-ExtElim.ll index 2df12d670a..bb3159e1e6 100644 --- a/test/Transforms/InstCombine/2010-03-03-ExtElim.ll +++ b/test/Transforms/InstCombine/2010-03-03-ExtElim.ll @@ -1,4 +1,4 @@ -; RUN: opt -instcombine -S %s | FileCheck %s +; RUN: opt -instcombine -S < %s | FileCheck %s ; PR6486 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-f80:32:32-n8:16:32" diff --git a/test/Transforms/InstCombine/2010-05-30-memcpy-Struct.ll b/test/Transforms/InstCombine/2010-05-30-memcpy-Struct.ll index b75fa5ad40..09a96749f2 100644 --- a/test/Transforms/InstCombine/2010-05-30-memcpy-Struct.ll +++ b/test/Transforms/InstCombine/2010-05-30-memcpy-Struct.ll @@ -1,4 +1,4 @@ -; RUN: opt -instcombine %s -S -o - | FileCheck %s +; RUN: opt -instcombine -S < %s | FileCheck %s ; PR7265 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" diff --git a/test/Transforms/InstCombine/2012-04-23-Neon-Intrinsics.ll b/test/Transforms/InstCombine/2012-04-23-Neon-Intrinsics.ll index 0907c490bb..2dedd44e2b 100644 --- a/test/Transforms/InstCombine/2012-04-23-Neon-Intrinsics.ll +++ b/test/Transforms/InstCombine/2012-04-23-Neon-Intrinsics.ll @@ -50,7 +50,7 @@ entry: %b = add <4 x i32> zeroinitializer, %a ret <4 x i32> %b ; CHECK: entry: -; CHECK-NEXT: %a = tail call <4 x i32> @llvm.arm.neon.vmulls.v4i32(<4 x i16> <i16 2, i16 2, i16 2, i16 2>, <4 x i16> %x) nounwind +; CHECK-NEXT: %a = tail call <4 x i32> @llvm.arm.neon.vmulls.v4i32(<4 x i16> <i16 2, i16 2, i16 2, i16 2>, <4 x i16> %x) [[NUW:#[0-9]+]] ; CHECK-NEXT: ret <4 x i32> %a } @@ -66,3 +66,7 @@ entry: declare <4 x i32> @llvm.arm.neon.vmulls.v4i32(<4 x i16>, <4 x i16>) nounwind readnone declare <4 x i32> @llvm.arm.neon.vmullu.v4i32(<4 x i16>, <4 x i16>) nounwind readnone + +; CHECK: attributes #0 = { nounwind readnone ssp } +; CHECK: attributes #1 = { nounwind readnone } +; CHECK: attributes [[NUW]] = { nounwind } diff --git a/test/Transforms/InstCombine/2012-12-14-simp-vgep.ll b/test/Transforms/InstCombine/2012-12-14-simp-vgep.ll new file mode 100644 index 0000000000..fc29b095e5 --- /dev/null +++ b/test/Transforms/InstCombine/2012-12-14-simp-vgep.ll @@ -0,0 +1,10 @@ +; RUN: opt < %s -instcombine -S + +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" + +define <4 x i32> @foo(<4 x i32*>* %in) { + %t17 = load <4 x i32*>* %in, align 8 + %t18 = icmp eq <4 x i32*> %t17, zeroinitializer + %t19 = zext <4 x i1> %t18 to <4 x i32> + ret <4 x i32> %t19 +} diff --git a/test/Transforms/InstCombine/2013-03-05-Combine-BitcastTy-Into-Alloca.ll b/test/Transforms/InstCombine/2013-03-05-Combine-BitcastTy-Into-Alloca.ll new file mode 100644 index 0000000000..b20c3a07c0 --- /dev/null +++ b/test/Transforms/InstCombine/2013-03-05-Combine-BitcastTy-Into-Alloca.ll @@ -0,0 +1,45 @@ +; RUN: opt < %s -instcombine -S | FileCheck %s +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-macosx10.9.0" + +%struct._my_struct = type <{ [12 x i8], [4 x i8] }> + +@initval = common global %struct._my_struct zeroinitializer, align 1 + +; InstCombine will try to change the %struct._my_struct alloca into an +; allocation of an i96 because of the bitcast to create %2. That's not valid, +; as the other 32 bits of the structure still feed into the return value +define { i64, i64 } @function(i32 %x, i32 %y, i32 %z) nounwind { +; CHECK: @function +; CHECK-NEXT: entry: +; CHECK-NEXT: %retval = alloca %struct._my_struct, align 8 +; CHECK-NOT: bitcast i96* %retval to %struct._my_struct* +entry: + %retval = alloca %struct._my_struct, align 8 + %k.sroa.0.0.copyload = load i96* bitcast (%struct._my_struct* @initval to i96*), align 1 + %k.sroa.1.12.copyload = load i32* bitcast ([4 x i8]* getelementptr inbounds (%struct._my_struct* @initval, i64 0, i32 1) to i32*), align 1 + %0 = zext i32 %x to i96 + %bf.value = shl nuw nsw i96 %0, 6 + %bf.clear = and i96 %k.sroa.0.0.copyload, -288230376151711744 + %1 = zext i32 %y to i96 + %bf.value2 = shl nuw nsw i96 %1, 32 + %bf.shl3 = and i96 %bf.value2, 288230371856744448 + %bf.value.masked = and i96 %bf.value, 4294967232 + %2 = zext i32 %z to i96 + %bf.value8 = and i96 %2, 63 + %bf.clear4 = or i96 %bf.shl3, %bf.value.masked + %bf.set5 = or i96 %bf.clear4, %bf.value8 + %bf.set10 = or i96 %bf.set5, %bf.clear + %retval.0.cast7 = bitcast %struct._my_struct* %retval to i96* + store i96 %bf.set10, i96* %retval.0.cast7, align 8 + %retval.12.idx8 = getelementptr inbounds %struct._my_struct* %retval, i64 0, i32 1 + %retval.12.cast9 = bitcast [4 x i8]* %retval.12.idx8 to i32* + store i32 %k.sroa.1.12.copyload, i32* %retval.12.cast9, align 4 + %trunc = trunc i96 %bf.set10 to i64 + %.fca.0.insert = insertvalue { i64, i64 } undef, i64 %trunc, 0 + %retval.8.idx12 = getelementptr inbounds %struct._my_struct* %retval, i64 0, i32 0, i64 8 + %retval.8.cast13 = bitcast i8* %retval.8.idx12 to i64* + %retval.8.load14 = load i64* %retval.8.cast13, align 8 + %.fca.1.insert = insertvalue { i64, i64 } %.fca.0.insert, i64 %retval.8.load14, 1 + ret { i64, i64 } %.fca.1.insert +} diff --git a/test/Transforms/InstCombine/bitcast-vector-fold.ll b/test/Transforms/InstCombine/bitcast-vector-fold.ll index 8feec22917..8fd7f35b7b 100644 --- a/test/Transforms/InstCombine/bitcast-vector-fold.ll +++ b/test/Transforms/InstCombine/bitcast-vector-fold.ll @@ -31,3 +31,8 @@ define <4 x i32> @test6() { %tmp3 = bitcast <2 x double> <double 0.5, double 1.0> to <4 x i32> ret <4 x i32> %tmp3 } + +define i32 @test7() { + %tmp3 = bitcast <2 x half> <half 0xH1100, half 0xH0011> to i32 + ret i32 %tmp3 +}
\ No newline at end of file diff --git a/test/Transforms/InstCombine/bitcast.ll b/test/Transforms/InstCombine/bitcast.ll index 8f6ae7d835..1e6113256b 100644 --- a/test/Transforms/InstCombine/bitcast.ll +++ b/test/Transforms/InstCombine/bitcast.ll @@ -11,7 +11,7 @@ define i32 @test1(i64 %a) { %t3 = xor <2 x i32> %t1, %t2 %t4 = extractelement <2 x i32> %t3, i32 0 ret i32 %t4 - + ; CHECK: @test1 ; CHECK: ret i32 0 } @@ -30,7 +30,7 @@ define float @test2(<2 x float> %A, <2 x i32> %B) { %add = fadd float %tmp24, %tmp4 ret float %add - + ; CHECK: @test2 ; CHECK-NEXT: %tmp24 = extractelement <2 x float> %A, i32 0 ; CHECK-NEXT: bitcast <2 x i32> %B to <2 x float> @@ -55,7 +55,7 @@ define float @test3(<2 x float> %A, <2 x i64> %B) { %add = fadd float %tmp24, %tmp4 ret float %add - + ; CHECK: @test3 ; CHECK-NEXT: %tmp24 = extractelement <2 x float> %A, i32 1 ; CHECK-NEXT: bitcast <2 x i64> %B to <4 x float> @@ -75,7 +75,7 @@ define <2 x i32> @test4(i32 %A, i32 %B){ ; CHECK: @test4 ; CHECK-NEXT: insertelement <2 x i32> undef, i32 %A, i32 0 ; CHECK-NEXT: insertelement <2 x i32> {{.*}}, i32 %B, i32 1 - ; CHECK-NEXT: ret <2 x i32> + ; CHECK-NEXT: ret <2 x i32> } @@ -92,7 +92,7 @@ define <2 x float> @test5(float %A, float %B) { ; CHECK: @test5 ; CHECK-NEXT: insertelement <2 x float> undef, float %A, i32 0 ; CHECK-NEXT: insertelement <2 x float> {{.*}}, float %B, i32 1 - ; CHECK-NEXT: ret <2 x float> + ; CHECK-NEXT: ret <2 x float> } define <2 x float> @test6(float %A){ @@ -123,7 +123,7 @@ define i64 @Vec2(i64 %in) { } define i64 @All11(i64 %in) { - %out = and i64 %in, xor (i64 bitcast (<2 x float> bitcast (i64 -1 to <2 x float>) to i64), i64 -1) + %out = and i64 %in, xor (i64 bitcast (<2 x float> bitcast (i64 -1 to <2 x float>) to i64), i64 -1) ret i64 %out ; CHECK: @All11 ; CHECK: ret i64 0 @@ -131,9 +131,16 @@ define i64 @All11(i64 %in) { define i32 @All111(i32 %in) { - %out = and i32 %in, xor (i32 bitcast (<1 x float> bitcast (i32 -1 to <1 x float>) to i32), i32 -1) + %out = and i32 %in, xor (i32 bitcast (<1 x float> bitcast (i32 -1 to <1 x float>) to i32), i32 -1) ret i32 %out ; CHECK: @All111 ; CHECK: ret i32 0 } +define <2 x i16> @BitcastInsert(i32 %a) { + %v = insertelement <1 x i32> undef, i32 %a, i32 0 + %r = bitcast <1 x i32> %v to <2 x i16> + ret <2 x i16> %r +; CHECK: @BitcastInsert +; CHECK: bitcast i32 %a to <2 x i16> +} diff --git a/test/Transforms/InstCombine/cast.ll b/test/Transforms/InstCombine/cast.ll index b4eb69d436..de738bb7c0 100644 --- a/test/Transforms/InstCombine/cast.ll +++ b/test/Transforms/InstCombine/cast.ll @@ -473,14 +473,12 @@ define i64 @test51(i64 %A, i1 %cond) { %F = sext i32 %E to i64 ret i64 %F ; CHECK: @test51 - -; FIXME: disabled, see PR5997 -; HECK-NEXT: %C = and i64 %A, 4294967294 -; HECK-NEXT: %D = or i64 %A, 1 -; HECK-NEXT: %E = select i1 %cond, i64 %C, i64 %D -; HECK-NEXT: %sext = shl i64 %E, 32 -; HECK-NEXT: %F = ashr i64 %sext, 32 -; HECK-NEXT: ret i64 %F +; CHECK-NEXT: %C = and i64 %A, 4294967294 +; CHECK-NEXT: %D = or i64 %A, 1 +; CHECK-NEXT: %E = select i1 %cond, i64 %C, i64 %D +; CHECK-NEXT: %sext = shl i64 %E, 32 +; CHECK-NEXT: %F = ashr exact i64 %sext, 32 +; CHECK-NEXT: ret i64 %F } define i32 @test52(i64 %A) { diff --git a/test/Transforms/InstCombine/compare-signs.ll b/test/Transforms/InstCombine/compare-signs.ll index f8e4911061..72db66e3ab 100644 --- a/test/Transforms/InstCombine/compare-signs.ll +++ b/test/Transforms/InstCombine/compare-signs.ll @@ -1,4 +1,4 @@ -; RUN: opt %s -instcombine -S | FileCheck %s +; RUN: opt -instcombine -S < %s | FileCheck %s ; PR5438 ; TODO: This should also optimize down. diff --git a/test/Transforms/InstCombine/constant-expr-datalayout.ll b/test/Transforms/InstCombine/constant-expr-datalayout.ll new file mode 100644 index 0000000000..9a72c77afd --- /dev/null +++ b/test/Transforms/InstCombine/constant-expr-datalayout.ll @@ -0,0 +1,12 @@ +; RUN: opt -instcombine %s -S -o - | FileCheck %s + +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-unknown-linux-gnu" + +%test1.struct = type { i32, i32 } +@test1.aligned_glbl = global %test1.struct zeroinitializer, align 4 +define void @test1(i64 *%ptr) { + store i64 and (i64 ptrtoint (i32* getelementptr (%test1.struct* @test1.aligned_glbl, i32 0, i32 1) to i64), i64 3), i64* %ptr +; CHECK: store i64 0, i64* %ptr + ret void +} diff --git a/test/Transforms/InstCombine/debuginfo.ll b/test/Transforms/InstCombine/debuginfo.ll index f6892fc3e1..f27ab278da 100644 --- a/test/Transforms/InstCombine/debuginfo.ll +++ b/test/Transforms/InstCombine/debuginfo.ll @@ -28,22 +28,23 @@ entry: ret i8* %call, !dbg !21 } -!llvm.dbg.lv.foobar = !{!0, !7, !9} -!llvm.dbg.sp = !{!1} +!25 = metadata !{metadata !0, metadata !7, metadata !9} +!llvm.dbg.cu = !{!3} +!24 = metadata !{metadata !1} -!0 = metadata !{i32 590081, metadata !1, metadata !"__dest", metadata !2, i32 16777294, metadata !6, i32 0} ; [ DW_TAG_arg_variable ] -!1 = metadata !{i32 589870, i32 0, metadata !2, metadata !"foobar", metadata !"foobar", metadata !"", metadata !2, i32 79, metadata !4, i1 true, i1 true, i32 0, i32 0, i32 0, i32 256, i1 true, i8* (i8*, i32, i64)* @foobar} ; [ DW_TAG_subprogram ] -!2 = metadata !{i32 589865, metadata !"string.h", metadata !"Game", metadata !3} ; [ DW_TAG_file_type ] -!3 = metadata !{i32 589841, i32 0, i32 12, metadata !"bits.c", metadata !"Game", metadata !"clang version 3.0 (trunk 127710)", i1 true, i1 true, metadata !"", i32 0} ; [ DW_TAG_compile_unit ] -!4 = metadata !{i32 589845, metadata !2, metadata !"", metadata !2, i32 0, i64 0, i64 0, i32 0, i32 0, i32 0, metadata !5, i32 0, i32 0} ; [ DW_TAG_subroutine_type ] +!0 = metadata !{i32 786689, metadata !1, metadata !"__dest", metadata !2, i32 16777294, metadata !6, i32 0, null} ; [ DW_TAG_arg_variable ] +!1 = metadata !{i32 786478, i32 0, metadata !2, metadata !"foobar", metadata !"foobar", metadata !"", metadata !2, i32 79, metadata !4, i1 true, i1 true, i32 0, i32 0, i32 0, i32 256, i1 true, i8* (i8*, i32, i64)* @foobar, null, null, metadata !25, i32 79} ; [ DW_TAG_subprogram ] +!2 = metadata !{i32 786473, metadata !"string.h", metadata !"Game", metadata !3} ; [ DW_TAG_file_type ] +!3 = metadata !{i32 786449, i32 0, i32 12, metadata !"bits.c", metadata !"Game", metadata !"clang version 3.0 (trunk 127710)", i1 true, i1 true, metadata !"", i32 0, null, null, metadata !24, null, null} ; [ DW_TAG_compile_unit ] +!4 = metadata !{i32 786453, metadata !2, metadata !"", metadata !2, i32 0, i64 0, i64 0, i32 0, i32 0, i32 0, metadata !5, i32 0, i32 0} ; [ DW_TAG_subroutine_type ] !5 = metadata !{metadata !6} -!6 = metadata !{i32 589839, metadata !3, metadata !"", null, i32 0, i64 64, i64 64, i64 0, i32 0, null} ; [ DW_TAG_pointer_type ] -!7 = metadata !{i32 590081, metadata !1, metadata !"__val", metadata !2, i32 33554510, metadata !8, i32 0} ; [ DW_TAG_arg_variable ] -!8 = metadata !{i32 589860, metadata !3, metadata !"int", null, i32 0, i64 32, i64 32, i64 0, i32 0, i32 5} ; [ DW_TAG_base_type ] -!9 = metadata !{i32 590081, metadata !1, metadata !"__len", metadata !2, i32 50331726, metadata !10, i32 0} ; [ DW_TAG_arg_variable ] +!6 = metadata !{i32 786447, metadata !3, metadata !"", null, i32 0, i64 64, i64 64, i64 0, i32 0, null} ; [ DW_TAG_pointer_type ] +!7 = metadata !{i32 786689, metadata !1, metadata !"__val", metadata !2, i32 33554510, metadata !8, i32 0, null} ; [ DW_TAG_arg_variable ] +!8 = metadata !{i32 786468, metadata !3, metadata !"int", null, i32 0, i64 32, i64 32, i64 0, i32 0, i32 5} ; [ DW_TAG_base_type ] +!9 = metadata !{i32 786689, metadata !1, metadata !"__len", metadata !2, i32 50331726, metadata !10, i32 0, null} ; [ DW_TAG_arg_variable ] !10 = metadata !{i32 589846, metadata !3, metadata !"size_t", metadata !2, i32 80, i64 0, i64 0, i64 0, i32 0, metadata !11} ; [ DW_TAG_typedef ] !11 = metadata !{i32 589846, metadata !3, metadata !"__darwin_size_t", metadata !2, i32 90, i64 0, i64 0, i64 0, i32 0, metadata !12} ; [ DW_TAG_typedef ] -!12 = metadata !{i32 589860, metadata !3, metadata !"long unsigned int", null, i32 0, i64 64, i64 64, i64 0, i32 0, i32 7} ; [ DW_TAG_base_type ] +!12 = metadata !{i32 786468, metadata !3, metadata !"long unsigned int", null, i32 0, i64 64, i64 64, i64 0, i32 0, i32 7} ; [ DW_TAG_base_type ] !13 = metadata !{metadata !"any pointer", metadata !14} !14 = metadata !{metadata !"omnipotent char", metadata !15} !15 = metadata !{metadata !"Simple C/C++ TBAA", null} @@ -53,5 +54,5 @@ entry: !19 = metadata !{metadata !"long", metadata !14} !20 = metadata !{i32 78, i32 54, metadata !1, null} !21 = metadata !{i32 80, i32 3, metadata !22, null} -!22 = metadata !{i32 589835, metadata !23, i32 80, i32 3, metadata !2, i32 7} ; [ DW_TAG_lexical_block ] -!23 = metadata !{i32 589835, metadata !1, i32 79, i32 1, metadata !2, i32 6} ; [ DW_TAG_lexical_block ] +!22 = metadata !{i32 786443, metadata !23, i32 80, i32 3, metadata !2, i32 7} ; [ DW_TAG_lexical_block ] +!23 = metadata !{i32 786443, metadata !1, i32 79, i32 1, metadata !2, i32 6} ; [ DW_TAG_lexical_block ] diff --git a/test/Transforms/InstCombine/devirt.ll b/test/Transforms/InstCombine/devirt.ll index 6189dc2af4..9c7cf5d697 100644 --- a/test/Transforms/InstCombine/devirt.ll +++ b/test/Transforms/InstCombine/devirt.ll @@ -1,4 +1,4 @@ -; RUN: opt -instcombine -S -o - %s | FileCheck %s +; RUN: opt -instcombine -S < %s | FileCheck %s ; CHECK-NOT: getelementptr ; CHECK-NOT: ptrtoint diff --git a/test/Transforms/InstCombine/exact.ll b/test/Transforms/InstCombine/exact.ll index 14741e3c1c..88ca88c3b9 100644 --- a/test/Transforms/InstCombine/exact.ll +++ b/test/Transforms/InstCombine/exact.ll @@ -99,9 +99,9 @@ define i1 @ashr_icmp2(i64 %X) nounwind { ; PR9998 ; Make sure we don't transform the ashr here into an sdiv ; CHECK: @pr9998 -; CHECK: = and i32 %V, 1 -; CHECK: %Z = icmp ne -; CHECK: ret i1 %Z +; CHECK: [[BIT:%[A-Za-z0-9.]+]] = and i32 %V, 1 +; CHECK-NEXT: [[CMP:%[A-Za-z0-9.]+]] = icmp ne i32 [[BIT]], 0 +; CHECK-NEXT: ret i1 [[CMP]] define i1 @pr9998(i32 %V) nounwind { entry: %W = shl i32 %V, 31 @@ -112,6 +112,7 @@ entry: } + ; CHECK: @udiv_icmp1 ; CHECK: icmp ne i64 %X, 0 define i1 @udiv_icmp1(i64 %X) nounwind { diff --git a/test/Transforms/InstCombine/fast-math.ll b/test/Transforms/InstCombine/fast-math.ll index b6a15677bb..3e32a2e4dd 100644 --- a/test/Transforms/InstCombine/fast-math.ll +++ b/test/Transforms/InstCombine/fast-math.ll @@ -3,19 +3,17 @@ ; testing-case "float fold(float a) { return 1.2f * a * 2.3f; }" ; 1.2f and 2.3f is supposed to be fold. define float @fold(float %a) { -fold: %mul = fmul fast float %a, 0x3FF3333340000000 %mul1 = fmul fast float %mul, 0x4002666660000000 ret float %mul1 -; CHECK: fold -; CHECK: fmul float %a, 0x4006147AE0000000 +; CHECK: @fold +; CHECK: fmul fast float %a, 0x4006147AE0000000 } ; Same testing-case as the one used in fold() except that the operators have ; fixed FP mode. define float @notfold(float %a) { -notfold: -; CHECK: notfold +; CHECK: @notfold ; CHECK: %mul = fmul fast float %a, 0x3FF3333340000000 %mul = fmul fast float %a, 0x3FF3333340000000 %mul1 = fmul float %mul, 0x4002666660000000 @@ -23,10 +21,332 @@ notfold: } define float @fold2(float %a) { -fold2: -; CHECK: fold2 -; CHECK: fmul float %a, 0x4006147AE0000000 +; CHECK: @fold2 +; CHECK: fmul fast float %a, 0x4006147AE0000000 %mul = fmul float %a, 0x3FF3333340000000 %mul1 = fmul fast float %mul, 0x4002666660000000 ret float %mul1 } + +; C * f1 + f1 = (C+1) * f1 +define double @fold3(double %f1) { + %t1 = fmul fast double 2.000000e+00, %f1 + %t2 = fadd fast double %f1, %t1 + ret double %t2 +; CHECK: @fold3 +; CHECK: fmul fast double %f1, 3.000000e+00 +} + +; (C1 - X) + (C2 - Y) => (C1+C2) - (X + Y) +define float @fold4(float %f1, float %f2) { + %sub = fsub float 4.000000e+00, %f1 + %sub1 = fsub float 5.000000e+00, %f2 + %add = fadd fast float %sub, %sub1 + ret float %add +; CHECK: @fold4 +; CHECK: %1 = fadd fast float %f1, %f2 +; CHECK: fsub fast float 9.000000e+00, %1 +} + +; (X + C1) + C2 => X + (C1 + C2) +define float @fold5(float %f1, float %f2) { + %add = fadd float %f1, 4.000000e+00 + %add1 = fadd fast float %add, 5.000000e+00 + ret float %add1 +; CHECK: @fold5 +; CHECK: fadd fast float %f1, 9.000000e+00 +} + +; (X + X) + X => 3.0 * X +define float @fold6(float %f1) { + %t1 = fadd fast float %f1, %f1 + %t2 = fadd fast float %f1, %t1 + ret float %t2 +; CHECK: @fold6 +; CHECK: fmul fast float %f1, 3.000000e+00 +} + +; C1 * X + (X + X) = (C1 + 2) * X +define float @fold7(float %f1) { + %t1 = fmul fast float %f1, 5.000000e+00 + %t2 = fadd fast float %f1, %f1 + %t3 = fadd fast float %t1, %t2 + ret float %t3 +; CHECK: @fold7 +; CHECK: fmul fast float %f1, 7.000000e+00 +} + +; (X + X) + (X + X) => 4.0 * X +define float @fold8(float %f1) { + %t1 = fadd fast float %f1, %f1 + %t2 = fadd fast float %f1, %f1 + %t3 = fadd fast float %t1, %t2 + ret float %t3 +; CHECK: fold8 +; CHECK: fmul fast float %f1, 4.000000e+00 +} + +; X - (X + Y) => 0 - Y +define float @fold9(float %f1, float %f2) { + %t1 = fadd float %f1, %f2 + %t3 = fsub fast float %f1, %t1 + ret float %t3 + +; CHECK: @fold9 +; CHECK: fsub fast float 0.000000e+00, %f2 +} + +; Let C3 = C1 + C2. (f1 + C1) + (f2 + C2) => (f1 + f2) + C3 instead of +; "(f1 + C3) + f2" or "(f2 + C3) + f1". Placing constant-addend at the +; top of resulting simplified expression tree may potentially reveal some +; optimization opportunities in the super-expression trees. +; +define float @fold10(float %f1, float %f2) { + %t1 = fadd fast float 2.000000e+00, %f1 + %t2 = fsub fast float %f2, 3.000000e+00 + %t3 = fadd fast float %t1, %t2 + ret float %t3 +; CHECK: @fold10 +; CHECK: %t3 = fadd fast float %t2, -1.000000e+00 +; CHECK: ret float %t3 +} + +; once cause Crash/miscompilation +define float @fail1(float %f1, float %f2) { + %conv3 = fadd fast float %f1, -1.000000e+00 + %add = fadd fast float %conv3, %conv3 + %add2 = fadd fast float %add, %conv3 + ret float %add2 +; CHECK: @fail1 +; CHECK: ret +} + +define double @fail2(double %f1, double %f2) { + %t1 = fsub fast double %f1, %f2 + %t2 = fadd fast double %f1, %f2 + %t3 = fsub fast double %t1, %t2 + ret double %t3 +; CHECK: @fail2 +; CHECK: ret +} + +; ========================================================================= +; +; Testing-cases about fmul begin +; +; ========================================================================= + +; ((X*C1) + C2) * C3 => (X * (C1*C3)) + (C2*C3) (i.e. distribution) +define float @fmul_distribute1(float %f1) { + %t1 = fmul float %f1, 6.0e+3 + %t2 = fadd float %t1, 2.0e+3 + %t3 = fmul fast float %t2, 5.0e+3 + ret float %t3 +; CHECK: @fmul_distribute1 +; CHECK: %1 = fmul fast float %f1, 3.000000e+07 +; CHECK: %t3 = fadd fast float %1, 1.000000e+07 +} + +; (X/C1 + C2) * C3 => X/(C1/C3) + C2*C3 +define double @fmul_distribute2(double %f1, double %f2) { + %t1 = fdiv double %f1, 3.0e+0 + %t2 = fadd double %t1, 5.0e+1 + ; 0x10000000000000 = DBL_MIN + %t3 = fmul fast double %t2, 0x10000000000000 + ret double %t3 + +; CHECK: @fmul_distribute2 +; CHECK: %1 = fdiv fast double %f1, 0x7FE8000000000000 +; CHECK: fadd fast double %1, 0x69000000000000 +} + +; 5.0e-1 * DBL_MIN yields denormal, so "(f1*3.0 + 5.0e-1) * DBL_MIN" cannot +; be simplified into f1 * (3.0*DBL_MIN) + (5.0e-1*DBL_MIN) +define double @fmul_distribute3(double %f1) { + %t1 = fdiv double %f1, 3.0e+0 + %t2 = fadd double %t1, 5.0e-1 + %t3 = fmul fast double %t2, 0x10000000000000 + ret double %t3 + +; CHECK: @fmul_distribute3 +; CHECK: fmul fast double %t2, 0x10000000000000 +} + +; ((X*C1) + C2) * C3 => (X * (C1*C3)) + (C2*C3) (i.e. distribution) +define float @fmul_distribute4(float %f1) { + %t1 = fmul float %f1, 6.0e+3 + %t2 = fsub float 2.0e+3, %t1 + %t3 = fmul fast float %t2, 5.0e+3 + ret float %t3 +; CHECK: @fmul_distribute4 +; CHECK: %1 = fmul fast float %f1, 3.000000e+07 +; CHECK: %t3 = fsub fast float 1.000000e+07, %1 +} + +; C1/X * C2 => (C1*C2) / X +define float @fmul2(float %f1) { + %t1 = fdiv float 2.0e+3, %f1 + %t3 = fmul fast float %t1, 6.0e+3 + ret float %t3 +; CHECK: @fmul2 +; CHECK: fdiv fast float 1.200000e+07, %f1 +} + +; X/C1 * C2 => X * (C2/C1) (if C2/C1 is normal Fp) +define float @fmul3(float %f1, float %f2) { + %t1 = fdiv float %f1, 2.0e+3 + %t3 = fmul fast float %t1, 6.0e+3 + ret float %t3 +; CHECK: @fmul3 +; CHECK: fmul fast float %f1, 3.000000e+00 +} + +; Rule "X/C1 * C2 => X * (C2/C1) is not applicable if C2/C1 is either a special +; value of a denormal. The 0x3810000000000000 here take value FLT_MIN +; +define float @fmul4(float %f1, float %f2) { + %t1 = fdiv float %f1, 2.0e+3 + %t3 = fmul fast float %t1, 0x3810000000000000 + ret float %t3 +; CHECK: @fmul4 +; CHECK: fmul fast float %t1, 0x3810000000000000 +} + +; X / C1 * C2 => X / (C2/C1) if C1/C2 is either a special value of a denormal, +; and C2/C1 is a normal value. +; +define float @fmul5(float %f1, float %f2) { + %t1 = fdiv float %f1, 3.0e+0 + %t3 = fmul fast float %t1, 0x3810000000000000 + ret float %t3 +; CHECK: @fmul5 +; CHECK: fdiv fast float %f1, 0x47E8000000000000 +} + +; (X*Y) * X => (X*X) * Y +define float @fmul6(float %f1, float %f2) { + %mul = fmul float %f1, %f2 + %mul1 = fmul fast float %mul, %f1 + ret float %mul1 +; CHECK: @fmul6 +; CHECK: fmul fast float %f1, %f1 +} + +; "(X*Y) * X => (X*X) * Y" is disabled if "X*Y" has multiple uses +define float @fmul7(float %f1, float %f2) { + %mul = fmul float %f1, %f2 + %mul1 = fmul fast float %mul, %f1 + %add = fadd float %mul1, %mul + ret float %add +; CHECK: @fmul7 +; CHECK: fmul fast float %mul, %f1 +} + +; ================================== |