176 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=x86_64-apple-macosx -mattr=+sse2 -verify-machineinstrs | FileCheck %s3 4; After tail duplication, two copies in an early exit BB can be cancelled out.5; rdar://106403636define i32 @t1(i32 %a, i32 %b) nounwind {7; CHECK-LABEL: t1:8; CHECK: ## %bb.0: ## %entry9; CHECK-NEXT: movl %edi, %eax10; CHECK-NEXT: testl %esi, %esi11; CHECK-NEXT: je LBB0_412; CHECK-NEXT: ## %bb.1: ## %while.body.preheader13; CHECK-NEXT: movl %esi, %edx14; CHECK-NEXT: .p2align 415; CHECK-NEXT: LBB0_2: ## %while.body16; CHECK-NEXT: ## =>This Inner Loop Header: Depth=117; CHECK-NEXT: movl %edx, %ecx18; CHECK-NEXT: cltd19; CHECK-NEXT: idivl %ecx20; CHECK-NEXT: testl %edx, %edx21; CHECK-NEXT: movl %ecx, %eax22; CHECK-NEXT: jne LBB0_223; CHECK-NEXT: ## %bb.3: ## %while.end24; CHECK-NEXT: movl %ecx, %eax25; CHECK-NEXT: LBB0_4:26; CHECK-NEXT: retq27entry:28 %cmp1 = icmp eq i32 %b, 029 br i1 %cmp1, label %while.end, label %while.body30 31while.body: ; preds = %entry, %while.body32 %a.addr.03 = phi i32 [ %b.addr.02, %while.body ], [ %a, %entry ]33 %b.addr.02 = phi i32 [ %rem, %while.body ], [ %b, %entry ]34 %rem = srem i32 %a.addr.03, %b.addr.0235 %cmp = icmp eq i32 %rem, 036 br i1 %cmp, label %while.end, label %while.body37 38while.end: ; preds = %while.body, %entry39 %a.addr.0.lcssa = phi i32 [ %a, %entry ], [ %b.addr.02, %while.body ]40 ret i32 %a.addr.0.lcssa41}42 43; Two movdqa (from phi-elimination) in the entry BB cancels out.44; rdar://1042816545define <8 x i16> @t2(<8 x i16> %T0, <8 x i16> %T1) nounwind readnone {46; CHECK-LABEL: t2:47; CHECK: ## %bb.0: ## %entry48; CHECK-NEXT: pshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]49; CHECK-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,1,1,2,4,5,6,7]50; CHECK-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]51; CHECK-NEXT: retq52entry:53 %tmp8 = shufflevector <8 x i16> %T0, <8 x i16> %T1, <8 x i32> < i32 undef, i32 undef, i32 7, i32 2, i32 8, i32 undef, i32 undef , i32 undef >54 ret <8 x i16> %tmp855}56 57define i32 @t3(i64 %a, i64 %b) nounwind {58; CHECK-LABEL: t3:59; CHECK: ## %bb.0: ## %entry60; CHECK-NEXT: movq %rdi, %rax61; CHECK-NEXT: testq %rsi, %rsi62; CHECK-NEXT: je LBB2_463; CHECK-NEXT: ## %bb.1: ## %while.body.preheader64; CHECK-NEXT: movq %rsi, %rdx65; CHECK-NEXT: .p2align 466; CHECK-NEXT: LBB2_2: ## %while.body67; CHECK-NEXT: ## =>This Inner Loop Header: Depth=168; CHECK-NEXT: movq %rdx, %rcx69; CHECK-NEXT: cqto70; CHECK-NEXT: idivq %rcx71; CHECK-NEXT: testq %rdx, %rdx72; CHECK-NEXT: movq %rcx, %rax73; CHECK-NEXT: jne LBB2_274; CHECK-NEXT: ## %bb.3: ## %while.end75; CHECK-NEXT: movl %ecx, %eax76; CHECK-NEXT: LBB2_4:77; CHECK-NEXT: retq78entry:79 %cmp1 = icmp eq i64 %b, 080 br i1 %cmp1, label %while.end, label %while.body81 82while.body: ; preds = %entry, %while.body83 %a.addr.03 = phi i64 [ %b.addr.02, %while.body ], [ %a, %entry ]84 %b.addr.02 = phi i64 [ %rem, %while.body ], [ %b, %entry ]85 %rem = srem i64 %a.addr.03, %b.addr.0286 %cmp = icmp eq i64 %rem, 087 br i1 %cmp, label %while.end, label %while.body88 89while.end: ; preds = %while.body, %entry90 %a.addr.0.lcssa = phi i64 [ %a, %entry ], [ %b.addr.02, %while.body ]91 %t = trunc i64 %a.addr.0.lcssa to i3292 ret i32 %t93}94 95; Check that copy propagation does not kill thing like:96; dst = copy src <-- do not kill that.97; ... = op1 undef dst98; ... = op2 dst <-- this is used here.99define <16 x float> @foo(<16 x float> %x) {100; CHECK-LABEL: foo:101; CHECK: ## %bb.0: ## %bb102; CHECK-NEXT: xorps %xmm5, %xmm5103; CHECK-NEXT: cvttps2dq %xmm3, %xmm8104; CHECK-NEXT: movaps %xmm3, %xmm4105; CHECK-NEXT: cmpltps %xmm5, %xmm4106; CHECK-NEXT: movaps {{.*#+}} xmm7 = [13,14,15,16]107; CHECK-NEXT: movaps %xmm4, %xmm6108; CHECK-NEXT: orps %xmm7, %xmm6109; CHECK-NEXT: cvtdq2ps %xmm8, %xmm3110; CHECK-NEXT: andps %xmm7, %xmm3111; CHECK-NEXT: andps %xmm6, %xmm3112; CHECK-NEXT: andnps %xmm4, %xmm6113; CHECK-NEXT: cvttps2dq %xmm2, %xmm4114; CHECK-NEXT: movaps %xmm2, %xmm7115; CHECK-NEXT: cmpltps %xmm5, %xmm7116; CHECK-NEXT: movaps {{.*#+}} xmm8 = [9,10,11,12]117; CHECK-NEXT: movaps %xmm7, %xmm9118; CHECK-NEXT: orps %xmm8, %xmm9119; CHECK-NEXT: cvtdq2ps %xmm4, %xmm2120; CHECK-NEXT: andps %xmm8, %xmm2121; CHECK-NEXT: andps %xmm9, %xmm2122; CHECK-NEXT: andnps %xmm7, %xmm9123; CHECK-NEXT: cvttps2dq %xmm1, %xmm4124; CHECK-NEXT: cmpltps %xmm5, %xmm1125; CHECK-NEXT: movaps {{.*#+}} xmm7 = [5,6,7,8]126; CHECK-NEXT: movaps %xmm1, %xmm8127; CHECK-NEXT: orps %xmm7, %xmm8128; CHECK-NEXT: cvtdq2ps %xmm4, %xmm4129; CHECK-NEXT: andps %xmm7, %xmm4130; CHECK-NEXT: andps %xmm8, %xmm4131; CHECK-NEXT: andnps %xmm1, %xmm8132; CHECK-NEXT: cvttps2dq %xmm0, %xmm1133; CHECK-NEXT: cmpltps %xmm5, %xmm0134; CHECK-NEXT: movaps {{.*#+}} xmm5 = [1,2,3,4]135; CHECK-NEXT: movaps %xmm0, %xmm7136; CHECK-NEXT: orps %xmm5, %xmm7137; CHECK-NEXT: cvtdq2ps %xmm1, %xmm1138; CHECK-NEXT: andps %xmm5, %xmm1139; CHECK-NEXT: andps %xmm7, %xmm1140; CHECK-NEXT: andnps %xmm0, %xmm7141; CHECK-NEXT: movaps {{.*#+}} xmm0 = [1,1,1,1]142; CHECK-NEXT: andps %xmm0, %xmm7143; CHECK-NEXT: orps %xmm7, %xmm1144; CHECK-NEXT: andps %xmm0, %xmm8145; CHECK-NEXT: orps %xmm8, %xmm4146; CHECK-NEXT: andps %xmm0, %xmm9147; CHECK-NEXT: orps %xmm9, %xmm2148; CHECK-NEXT: andps %xmm0, %xmm6149; CHECK-NEXT: orps %xmm6, %xmm3150; CHECK-NEXT: movaps %xmm1, %xmm0151; CHECK-NEXT: movaps %xmm4, %xmm1152; CHECK-NEXT: retq153bb:154 %v3 = icmp slt <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>, zeroinitializer155 %v14 = zext <16 x i1> %v3 to <16 x i32>156 %v16 = fcmp olt <16 x float> %x, zeroinitializer157 %v17 = sext <16 x i1> %v16 to <16 x i32>158 %v18 = zext <16 x i1> %v16 to <16 x i32>159 %v19 = xor <16 x i32> %v14, %v18160 %v20 = or <16 x i32> %v17, <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16>161 %v21 = fptosi <16 x float> %x to <16 x i32>162 %v22 = sitofp <16 x i32> %v21 to <16 x float>163 %v69 = fcmp ogt <16 x float> %v22, zeroinitializer164 %v75 = and <16 x i1> %v69, %v3165 %v77 = bitcast <16 x float> %v22 to <16 x i32>166 %v79 = sext <16 x i1> %v75 to <16 x i32>167 %v80 = and <16 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16>, %v79168 %v81 = xor <16 x i32> %v77, %v80169 %v82 = and <16 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16>, %v81170 %v83 = xor <16 x i32> %v19, %v82171 %v84 = and <16 x i32> %v83, %v20172 %v85 = xor <16 x i32> %v19, %v84173 %v86 = bitcast <16 x i32> %v85 to <16 x float>174 ret <16 x float> %v86175}176