145 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -passes=instcombine -S | FileCheck %s3 4; Narrow the select operands to eliminate the existing shuffles and replace a wide select with a narrow select.5 6define <2 x i8> @narrow_shuffle_of_select(<2 x i1> %cmp, <4 x i8> %x, <4 x i8> %y) {7; CHECK-LABEL: @narrow_shuffle_of_select(8; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x i8> [[X:%.*]], <4 x i8> poison, <2 x i32> <i32 0, i32 1>9; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x i8> [[Y:%.*]], <4 x i8> poison, <2 x i32> <i32 0, i32 1>10; CHECK-NEXT: [[R:%.*]] = select <2 x i1> [[CMP:%.*]], <2 x i8> [[TMP1]], <2 x i8> [[TMP2]]11; CHECK-NEXT: ret <2 x i8> [[R]]12;13 %widecmp = shufflevector <2 x i1> %cmp, <2 x i1> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>14 %widesel = select <4 x i1> %widecmp, <4 x i8> %x, <4 x i8> %y15 %r = shufflevector <4 x i8> %widesel, <4 x i8> undef, <2 x i32> <i32 0, i32 1>16 ret <2 x i8> %r17}18 19; The 1st shuffle is not extending with undefs, but demanded elements corrects that.20 21define <2 x i8> @narrow_shuffle_of_select_overspecified_extend(<2 x i1> %cmp, <4 x i8> %x, <4 x i8> %y) {22; CHECK-LABEL: @narrow_shuffle_of_select_overspecified_extend(23; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x i8> [[X:%.*]], <4 x i8> poison, <2 x i32> <i32 0, i32 1>24; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x i8> [[Y:%.*]], <4 x i8> poison, <2 x i32> <i32 0, i32 1>25; CHECK-NEXT: [[R:%.*]] = select <2 x i1> [[CMP:%.*]], <2 x i8> [[TMP1]], <2 x i8> [[TMP2]]26; CHECK-NEXT: ret <2 x i8> [[R]]27;28 %widecmp = shufflevector <2 x i1> %cmp, <2 x i1> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>29 %widesel = select <4 x i1> %widecmp, <4 x i8> %x, <4 x i8> %y30 %r = shufflevector <4 x i8> %widesel, <4 x i8> undef, <2 x i32> <i32 0, i32 1>31 ret <2 x i8> %r32}33 34; Verify that undef elements are acceptable for identity shuffle mask. Also check FP types.35 36define <3 x float> @narrow_shuffle_of_select_undefs(<3 x i1> %cmp, <4 x float> %x, <4 x float> %y) {37; CHECK-LABEL: @narrow_shuffle_of_select_undefs(38; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x float> [[X:%.*]], <4 x float> poison, <3 x i32> <i32 0, i32 1, i32 poison>39; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x float> [[Y:%.*]], <4 x float> poison, <3 x i32> <i32 0, i32 1, i32 poison>40; CHECK-NEXT: [[R:%.*]] = select <3 x i1> [[CMP:%.*]], <3 x float> [[TMP1]], <3 x float> [[TMP2]]41; CHECK-NEXT: ret <3 x float> [[R]]42;43 %widecmp = shufflevector <3 x i1> %cmp, <3 x i1> undef, <4 x i32> <i32 undef, i32 1, i32 2, i32 undef>44 %widesel = select <4 x i1> %widecmp, <4 x float> %x, <4 x float> %y45 %r = shufflevector <4 x float> %widesel, <4 x float> undef, <3 x i32> <i32 0, i32 1, i32 undef>46 ret <3 x float> %r47}48 49declare void @use(<4 x i8>)50declare void @use_cmp(<4 x i1>)51 52; Negative test - extra use would require more instructions than we started with.53 54define <2 x i8> @narrow_shuffle_of_select_use1(<2 x i1> %cmp, <4 x i8> %x, <4 x i8> %y) {55; CHECK-LABEL: @narrow_shuffle_of_select_use1(56; CHECK-NEXT: [[WIDECMP:%.*]] = shufflevector <2 x i1> [[CMP:%.*]], <2 x i1> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>57; CHECK-NEXT: [[WIDESEL:%.*]] = select <4 x i1> [[WIDECMP]], <4 x i8> [[X:%.*]], <4 x i8> [[Y:%.*]]58; CHECK-NEXT: call void @use(<4 x i8> [[WIDESEL]])59; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i8> [[WIDESEL]], <4 x i8> poison, <2 x i32> <i32 0, i32 1>60; CHECK-NEXT: ret <2 x i8> [[R]]61;62 %widecmp = shufflevector <2 x i1> %cmp, <2 x i1> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>63 %widesel = select <4 x i1> %widecmp, <4 x i8> %x, <4 x i8> %y64 call void @use(<4 x i8> %widesel)65 %r = shufflevector <4 x i8> %widesel, <4 x i8> undef, <2 x i32> <i32 0, i32 1>66 ret <2 x i8> %r67}68 69; Negative test - extra use would require more instructions than we started with.70 71define <2 x i8> @narrow_shuffle_of_select_use2(<2 x i1> %cmp, <4 x i8> %x, <4 x i8> %y) {72; CHECK-LABEL: @narrow_shuffle_of_select_use2(73; CHECK-NEXT: [[WIDECMP:%.*]] = shufflevector <2 x i1> [[CMP:%.*]], <2 x i1> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>74; CHECK-NEXT: call void @use_cmp(<4 x i1> [[WIDECMP]])75; CHECK-NEXT: [[WIDESEL:%.*]] = select <4 x i1> [[WIDECMP]], <4 x i8> [[X:%.*]], <4 x i8> [[Y:%.*]]76; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i8> [[WIDESEL]], <4 x i8> poison, <2 x i32> <i32 0, i32 1>77; CHECK-NEXT: ret <2 x i8> [[R]]78;79 %widecmp = shufflevector <2 x i1> %cmp, <2 x i1> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>80 call void @use_cmp(<4 x i1> %widecmp)81 %widesel = select <4 x i1> %widecmp, <4 x i8> %x, <4 x i8> %y82 %r = shufflevector <4 x i8> %widesel, <4 x i8> undef, <2 x i32> <i32 0, i32 1>83 ret <2 x i8> %r84}85 86; Negative test - mismatched types would require extra shuffling.87 88define <3 x i8> @narrow_shuffle_of_select_mismatch_types1(<2 x i1> %cmp, <4 x i8> %x, <4 x i8> %y) {89; CHECK-LABEL: @narrow_shuffle_of_select_mismatch_types1(90; CHECK-NEXT: [[WIDECMP:%.*]] = shufflevector <2 x i1> [[CMP:%.*]], <2 x i1> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>91; CHECK-NEXT: [[WIDESEL:%.*]] = select <4 x i1> [[WIDECMP]], <4 x i8> [[X:%.*]], <4 x i8> [[Y:%.*]]92; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i8> [[WIDESEL]], <4 x i8> poison, <3 x i32> <i32 0, i32 1, i32 2>93; CHECK-NEXT: ret <3 x i8> [[R]]94;95 %widecmp = shufflevector <2 x i1> %cmp, <2 x i1> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>96 %widesel = select <4 x i1> %widecmp, <4 x i8> %x, <4 x i8> %y97 %r = shufflevector <4 x i8> %widesel, <4 x i8> undef, <3 x i32> <i32 0, i32 1, i32 2>98 ret <3 x i8> %r99}100 101; Negative test - mismatched types would require extra shuffling.102 103define <3 x i8> @narrow_shuffle_of_select_mismatch_types2(<4 x i1> %cmp, <6 x i8> %x, <6 x i8> %y) {104; CHECK-LABEL: @narrow_shuffle_of_select_mismatch_types2(105; CHECK-NEXT: [[WIDECMP:%.*]] = shufflevector <4 x i1> [[CMP:%.*]], <4 x i1> poison, <6 x i32> <i32 0, i32 1, i32 2, i32 poison, i32 poison, i32 poison>106; CHECK-NEXT: [[WIDESEL:%.*]] = select <6 x i1> [[WIDECMP]], <6 x i8> [[X:%.*]], <6 x i8> [[Y:%.*]]107; CHECK-NEXT: [[R:%.*]] = shufflevector <6 x i8> [[WIDESEL]], <6 x i8> poison, <3 x i32> <i32 0, i32 1, i32 2>108; CHECK-NEXT: ret <3 x i8> [[R]]109;110 %widecmp = shufflevector <4 x i1> %cmp, <4 x i1> undef, <6 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef>111 %widesel = select <6 x i1> %widecmp, <6 x i8> %x, <6 x i8> %y112 %r = shufflevector <6 x i8> %widesel, <6 x i8> undef, <3 x i32> <i32 0, i32 1, i32 2>113 ret <3 x i8> %r114}115 116; Narrowing constants does not require creating new narrowing shuffle instructions.117 118define <2 x i8> @narrow_shuffle_of_select_consts(<2 x i1> %cmp) {119; CHECK-LABEL: @narrow_shuffle_of_select_consts(120; CHECK-NEXT: [[R:%.*]] = select <2 x i1> [[CMP:%.*]], <2 x i8> <i8 -1, i8 -2>, <2 x i8> <i8 1, i8 2>121; CHECK-NEXT: ret <2 x i8> [[R]]122;123 %widecmp = shufflevector <2 x i1> %cmp, <2 x i1> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>124 %widesel = select <4 x i1> %widecmp, <4 x i8> <i8 -1, i8 -2, i8 -3, i8 -4>, <4 x i8> <i8 1, i8 2, i8 3, i8 4>125 %r = shufflevector <4 x i8> %widesel, <4 x i8> undef, <2 x i32> <i32 0, i32 1>126 ret <2 x i8> %r127}128 129; PR38691 - https://bugs.llvm.org/show_bug.cgi?id=38691130; If the operands are widened only to be narrowed back, then all of the shuffles are unnecessary.131 132define <2 x i8> @narrow_shuffle_of_select_with_widened_ops(<2 x i1> %cmp, <2 x i8> %x, <2 x i8> %y) {133; CHECK-LABEL: @narrow_shuffle_of_select_with_widened_ops(134; CHECK-NEXT: [[R:%.*]] = select <2 x i1> [[CMP:%.*]], <2 x i8> [[X:%.*]], <2 x i8> [[Y:%.*]]135; CHECK-NEXT: ret <2 x i8> [[R]]136;137 %widex = shufflevector <2 x i8> %x, <2 x i8> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>138 %widey = shufflevector <2 x i8> %y, <2 x i8> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>139 %widecmp = shufflevector <2 x i1> %cmp, <2 x i1> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>140 %widesel = select <4 x i1> %widecmp, <4 x i8> %widex, <4 x i8> %widey141 %r = shufflevector <4 x i8> %widesel, <4 x i8> undef, <2 x i32> <i32 0, i32 1>142 ret <2 x i8> %r143}144 145