diff options
Diffstat (limited to 'llvm/test/Transforms/LoopStrengthReduce/AArch64')
6 files changed, 0 insertions, 358 deletions
diff --git a/llvm/test/Transforms/LoopStrengthReduce/AArch64/lit.local.cfg b/llvm/test/Transforms/LoopStrengthReduce/AArch64/lit.local.cfg deleted file mode 100644 index 675f48e199a..00000000000 --- a/llvm/test/Transforms/LoopStrengthReduce/AArch64/lit.local.cfg +++ /dev/null @@ -1,4 +0,0 @@ -config.suffixes = ['.ll'] - -if not 'AArch64' in config.root.targets: - config.unsupported = True diff --git a/llvm/test/Transforms/LoopStrengthReduce/AArch64/lsr-memcpy.ll b/llvm/test/Transforms/LoopStrengthReduce/AArch64/lsr-memcpy.ll deleted file mode 100644 index 2120b2a3bfe..00000000000 --- a/llvm/test/Transforms/LoopStrengthReduce/AArch64/lsr-memcpy.ll +++ /dev/null @@ -1,33 +0,0 @@ -; RUN: llc -mtriple=arm64-unknown-unknown -mcpu=cyclone -pre-RA-sched=list-hybrid < %s | FileCheck %s -; rdar://10232252 -; Prevent LSR of doing poor choice that cannot be folded in addressing mode - -; Remove the -pre-RA-sched=list-hybrid option after fixing: -; <rdar://problem/12702735> [ARM64][coalescer] need better register -; coalescing for simple unit tests. - -; CHECK: testCase -; CHECK: %while.body{{$}} -; CHECK: ldr [[STREG:x[0-9]+]], [{{x[0-9]+}}], #8 -; CHECK-NEXT: str [[STREG]], [{{x[0-9]+}}], #8 -; CHECK: %while.end -define i32 @testCase() nounwind ssp { -entry: - br label %while.body - -while.body: ; preds = %while.body, %entry - %len.06 = phi i64 [ 1288, %entry ], [ %sub, %while.body ] - %pDst.05 = phi i64* [ inttoptr (i64 6442450944 to i64*), %entry ], [ %incdec.ptr1, %while.body ] - %pSrc.04 = phi i64* [ inttoptr (i64 4294967296 to i64*), %entry ], [ %incdec.ptr, %while.body ] - %incdec.ptr = getelementptr inbounds i64, i64* %pSrc.04, i64 1 - %tmp = load volatile i64, i64* %pSrc.04, align 8 - %incdec.ptr1 = getelementptr inbounds i64, i64* %pDst.05, i64 1 - store volatile i64 %tmp, i64* %pDst.05, align 8 - %sub = add i64 %len.06, -8 - %cmp = icmp sgt i64 %sub, -1 - br i1 %cmp, label %while.body, label %while.end - -while.end: ; preds = %while.body - tail call void inttoptr (i64 6442450944 to void ()*)() nounwind - ret i32 0 -} diff --git a/llvm/test/Transforms/LoopStrengthReduce/AArch64/lsr-memset.ll b/llvm/test/Transforms/LoopStrengthReduce/AArch64/lsr-memset.ll deleted file mode 100644 index e84f675736d..00000000000 --- a/llvm/test/Transforms/LoopStrengthReduce/AArch64/lsr-memset.ll +++ /dev/null @@ -1,101 +0,0 @@ -; RUN: llc < %s -O3 -mtriple=arm64-unknown-unknown -mcpu=cyclone -pre-RA-sched=list-hybrid | FileCheck %s -; <rdar://problem/11635990> [arm64] [lsr] Inefficient EA/loop-exit calc in bzero_phys -; -; LSR on loop %while.cond should reassociate non-address mode -; expressions at use %cmp16 to avoid sinking computation into %while.body18. -; -; Remove the -pre-RA-sched=list-hybrid option after fixing: -; <rdar://problem/12702735> [ARM64][coalescer] need better register -; coalescing for simple unit tests. - -; CHECK: @memset -; CHECK: %while.body18{{$}} -; CHECK: str x{{[0-9]+}}, [x{{[0-9]+}}], #8 -; First set the IVREG variable, then use it -; CHECK-NEXT: sub [[IVREG:x[0-9]+]], -; CHECK: [[IVREG]], #8 -; CHECK-NEXT: cmp [[IVREG]], #7 -; CHECK-NEXT: b.hi -define i8* @memset(i8* %dest, i32 %val, i64 %len) nounwind ssp noimplicitfloat { -entry: - %cmp = icmp eq i64 %len, 0 - br i1 %cmp, label %done, label %while.cond.preheader - -while.cond.preheader: ; preds = %entry - %conv = trunc i32 %val to i8 - br label %while.cond - -while.cond: ; preds = %while.body, %while.cond.preheader - %ptr.0 = phi i8* [ %incdec.ptr, %while.body ], [ %dest, %while.cond.preheader ] - %len.addr.0 = phi i64 [ %dec, %while.body ], [ %len, %while.cond.preheader ] - %cond = icmp eq i64 %len.addr.0, 0 - br i1 %cond, label %done, label %land.rhs - -land.rhs: ; preds = %while.cond - %0 = ptrtoint i8* %ptr.0 to i64 - %and = and i64 %0, 7 - %cmp5 = icmp eq i64 %and, 0 - br i1 %cmp5, label %if.end9, label %while.body - -while.body: ; preds = %land.rhs - %incdec.ptr = getelementptr inbounds i8, i8* %ptr.0, i64 1 - store i8 %conv, i8* %ptr.0, align 1, !tbaa !0 - %dec = add i64 %len.addr.0, -1 - br label %while.cond - -if.end9: ; preds = %land.rhs - %conv.mask = and i32 %val, 255 - %1 = zext i32 %conv.mask to i64 - %2 = shl nuw nsw i64 %1, 8 - %ins18 = or i64 %2, %1 - %3 = shl nuw nsw i64 %1, 16 - %ins15 = or i64 %ins18, %3 - %4 = shl nuw nsw i64 %1, 24 - %5 = shl nuw nsw i64 %1, 32 - %mask8 = or i64 %ins15, %4 - %6 = shl nuw nsw i64 %1, 40 - %mask5 = or i64 %mask8, %5 - %7 = shl nuw nsw i64 %1, 48 - %8 = shl nuw i64 %1, 56 - %mask2.masked = or i64 %mask5, %6 - %mask = or i64 %mask2.masked, %7 - %ins = or i64 %mask, %8 - %9 = bitcast i8* %ptr.0 to i64* - %cmp1636 = icmp ugt i64 %len.addr.0, 7 - br i1 %cmp1636, label %while.body18, label %while.body29.lr.ph - -while.body18: ; preds = %if.end9, %while.body18 - %wideptr.038 = phi i64* [ %incdec.ptr19, %while.body18 ], [ %9, %if.end9 ] - %len.addr.137 = phi i64 [ %sub, %while.body18 ], [ %len.addr.0, %if.end9 ] - %incdec.ptr19 = getelementptr inbounds i64, i64* %wideptr.038, i64 1 - store i64 %ins, i64* %wideptr.038, align 8, !tbaa !2 - %sub = add i64 %len.addr.137, -8 - %cmp16 = icmp ugt i64 %sub, 7 - br i1 %cmp16, label %while.body18, label %while.end20 - -while.end20: ; preds = %while.body18 - %cmp21 = icmp eq i64 %sub, 0 - br i1 %cmp21, label %done, label %while.body29.lr.ph - -while.body29.lr.ph: ; preds = %while.end20, %if.end9 - %len.addr.1.lcssa49 = phi i64 [ %sub, %while.end20 ], [ %len.addr.0, %if.end9 ] - %wideptr.0.lcssa48 = phi i64* [ %incdec.ptr19, %while.end20 ], [ %9, %if.end9 ] - %10 = bitcast i64* %wideptr.0.lcssa48 to i8* - br label %while.body29 - -while.body29: ; preds = %while.body29, %while.body29.lr.ph - %len.addr.235 = phi i64 [ %len.addr.1.lcssa49, %while.body29.lr.ph ], [ %dec26, %while.body29 ] - %ptr.134 = phi i8* [ %10, %while.body29.lr.ph ], [ %incdec.ptr31, %while.body29 ] - %dec26 = add i64 %len.addr.235, -1 - %incdec.ptr31 = getelementptr inbounds i8, i8* %ptr.134, i64 1 - store i8 %conv, i8* %ptr.134, align 1, !tbaa !0 - %cmp27 = icmp eq i64 %dec26, 0 - br i1 %cmp27, label %done, label %while.body29 - -done: ; preds = %while.cond, %while.body29, %while.end20, %entry - ret i8* %dest -} - -!0 = !{!"omnipotent char", !1} -!1 = !{!"Simple C/C++ TBAA"} -!2 = !{!"long long", !0} diff --git a/llvm/test/Transforms/LoopStrengthReduce/AArch64/lsr-reuse.ll b/llvm/test/Transforms/LoopStrengthReduce/AArch64/lsr-reuse.ll deleted file mode 100644 index a2dfe81b108..00000000000 --- a/llvm/test/Transforms/LoopStrengthReduce/AArch64/lsr-reuse.ll +++ /dev/null @@ -1,34 +0,0 @@ -; RUN: llc -mtriple=arm64-unknown-unknown -print-lsr-output < %s 2>&1 | FileCheck %s - -declare void @foo(i64) - -; Verify that redundant adds aren't inserted by LSR. -; CHECK-LABEL: @bar( -define void @bar(double* %A) { -entry: - br label %while.cond - -while.cond: -; CHECK-LABEL: while.cond: -; CHECK: add i64 %lsr.iv, 1 -; CHECK-NOT: add i64 %lsr.iv, 1 -; CHECK-LABEL: land.rhs: - %indvars.iv28 = phi i64 [ %indvars.iv.next29, %land.rhs ], [ 50, %entry ] - %cmp = icmp sgt i64 %indvars.iv28, 0 - br i1 %cmp, label %land.rhs, label %while.end - -land.rhs: - %indvars.iv.next29 = add nsw i64 %indvars.iv28, -1 - %arrayidx = getelementptr inbounds double, double* %A, i64 %indvars.iv.next29 - %Aload = load double, double* %arrayidx, align 8 - %cmp1 = fcmp oeq double %Aload, 0.000000e+00 - br i1 %cmp1, label %while.cond, label %if.end - -while.end: - %indvars.iv28.lcssa = phi i64 [ %indvars.iv28, %while.cond ] - tail call void @foo(i64 %indvars.iv28.lcssa) - br label %if.end - -if.end: - ret void -} diff --git a/llvm/test/Transforms/LoopStrengthReduce/AArch64/req-regs.ll b/llvm/test/Transforms/LoopStrengthReduce/AArch64/req-regs.ll deleted file mode 100644 index c877ace7abf..00000000000 --- a/llvm/test/Transforms/LoopStrengthReduce/AArch64/req-regs.ll +++ /dev/null @@ -1,70 +0,0 @@ -; RUN: llc -mcpu=cyclone -debug-only=loop-reduce < %s 2>&1 | FileCheck %s -; REQUIRES: asserts - -; LSR used to fail here due to a bug in the ReqRegs test. -; CHECK: The chosen solution requires -; CHECK-NOT: No Satisfactory Solution - -target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128" -target triple = "arm64-apple-ios" - -define void @do_integer_add(i64 %iterations, i8* nocapture readonly %cookie) { -entry: - %N = bitcast i8* %cookie to i32* - %0 = load i32, i32* %N, align 4 - %add = add nsw i32 %0, 57 - %cmp56 = icmp eq i64 %iterations, 0 - br i1 %cmp56, label %while.end, label %for.cond.preheader.preheader - -for.cond.preheader.preheader: ; preds = %entry - br label %for.cond.preheader - -while.cond.loopexit: ; preds = %for.body - %add21.lcssa = phi i32 [ %add21, %for.body ] - %dec58 = add i64 %dec58.in, -1 - %cmp = icmp eq i64 %dec58, 0 - br i1 %cmp, label %while.end.loopexit, label %for.cond.preheader - -for.cond.preheader: ; preds = %for.cond.preheader.preheader, %while.cond.loopexit - %dec58.in = phi i64 [ %dec58, %while.cond.loopexit ], [ %iterations, %for.cond.preheader.preheader ] - %a.057 = phi i32 [ %add21.lcssa, %while.cond.loopexit ], [ %add, %for.cond.preheader.preheader ] - br label %for.body - -for.body: ; preds = %for.body, %for.cond.preheader - %a.154 = phi i32 [ %a.057, %for.cond.preheader ], [ %add21, %for.body ] - %i.053 = phi i32 [ 1, %for.cond.preheader ], [ %inc, %for.body ] - %inc = add nsw i32 %i.053, 1 - %add2 = shl i32 %a.154, 1 - %add3 = add nsw i32 %add2, %i.053 - %add4 = shl i32 %add3, 1 - %add5 = add nsw i32 %add4, %i.053 - %add6 = shl i32 %add5, 1 - %add7 = add nsw i32 %add6, %i.053 - %add8 = shl i32 %add7, 1 - %add9 = add nsw i32 %add8, %i.053 - %add10 = shl i32 %add9, 1 - %add11 = add nsw i32 %add10, %i.053 - %add12 = shl i32 %add11, 1 - %add13 = add nsw i32 %add12, %i.053 - %add14 = shl i32 %add13, 1 - %add15 = add nsw i32 %add14, %i.053 - %add16 = shl i32 %add15, 1 - %add17 = add nsw i32 %add16, %i.053 - %add18 = shl i32 %add17, 1 - %add19 = add nsw i32 %add18, %i.053 - %add20 = shl i32 %add19, 1 - %add21 = add nsw i32 %add20, %i.053 - %exitcond = icmp eq i32 %inc, 1001 - br i1 %exitcond, label %while.cond.loopexit, label %for.body - -while.end.loopexit: ; preds = %while.cond.loopexit - %add21.lcssa.lcssa = phi i32 [ %add21.lcssa, %while.cond.loopexit ] - br label %while.end - -while.end: ; preds = %while.end.loopexit, %entry - %a.0.lcssa = phi i32 [ %add, %entry ], [ %add21.lcssa.lcssa, %while.end.loopexit ] - tail call void @use_int(i32 %a.0.lcssa) - ret void -} - -declare void @use_int(i32) diff --git a/llvm/test/Transforms/LoopStrengthReduce/AArch64/small-constant.ll b/llvm/test/Transforms/LoopStrengthReduce/AArch64/small-constant.ll deleted file mode 100644 index 04ad762df99..00000000000 --- a/llvm/test/Transforms/LoopStrengthReduce/AArch64/small-constant.ll +++ /dev/null @@ -1,116 +0,0 @@ -; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py - -; RUN: llc < %s -mtriple=aarch64-unknown-unknown | FileCheck %s - -; Test LSR for giving small constants, which get re-associated as unfolded -; offset, a chance to get combined with loop-invariant registers (same as -; large constants which do not fit as add immediate operands). LSR -; favors here to bump the base pointer outside the loop. - -; float test(float *arr, long long start, float threshold) { -; for (long long i = start; i != 0; ++i) { -; float x = arr[i + 7]; -; if (x > threshold) -; return x; -; } -; return -7; -; } -define float @test1(float* nocapture readonly %arr, i64 %start, float %threshold) { -; CHECK-LABEL: test1: -; CHECK: // %bb.0: // %entry -; CHECK-NEXT: fmov s2, #-7.00000000 -; CHECK-NEXT: cbz x1, .LBB0_5 -; CHECK-NEXT: // %bb.1: // %for.body.preheader -; CHECK-NEXT: add x8, x0, #28 // =28 -; CHECK-NEXT: .LBB0_2: // %for.body -; CHECK-NEXT: // =>This Inner Loop Header: Depth=1 -; CHECK-NEXT: ldr s1, [x8, x1, lsl #2] -; CHECK-NEXT: fcmp s1, s0 -; CHECK-NEXT: b.gt .LBB0_6 -; CHECK-NEXT: // %bb.3: // %for.cond -; CHECK-NEXT: // in Loop: Header=BB0_2 Depth=1 -; CHECK-NEXT: add x1, x1, #1 // =1 -; CHECK-NEXT: cbnz x1, .LBB0_2 -; CHECK-NEXT: // %bb.4: -; CHECK-NEXT: mov v0.16b, v2.16b -; CHECK-NEXT: ret -; CHECK-NEXT: .LBB0_5: -; CHECK-NEXT: mov v0.16b, v2.16b -; CHECK-NEXT: ret -; CHECK-NEXT: .LBB0_6: // %cleanup2 -; CHECK-NEXT: mov v0.16b, v1.16b -; CHECK-NEXT: ret -entry: - %cmp11 = icmp eq i64 %start, 0 - br i1 %cmp11, label %cleanup2, label %for.body - -for.cond: ; preds = %for.body - %cmp = icmp eq i64 %inc, 0 - br i1 %cmp, label %cleanup2, label %for.body - -for.body: ; preds = %entry, %for.cond - %i.012 = phi i64 [ %inc, %for.cond ], [ %start, %entry ] - %add = add nsw i64 %i.012, 7 - %arrayidx = getelementptr inbounds float, float* %arr, i64 %add - %0 = load float, float* %arrayidx, align 4 - %cmp1 = fcmp ogt float %0, %threshold - %inc = add nsw i64 %i.012, 1 - br i1 %cmp1, label %cleanup2, label %for.cond - -cleanup2: ; preds = %for.cond, %for.body, %entry - %1 = phi float [ -7.000000e+00, %entry ], [ %0, %for.body ], [ -7.000000e+00, %for.cond ] - ret float %1 -} - -; Same as test1, except i has another use: -; if (x > threshold) ---> if (x > threshold + i) -define float @test2(float* nocapture readonly %arr, i64 %start, float %threshold) { -; CHECK-LABEL: test2: -; CHECK: // %bb.0: // %entry -; CHECK-NEXT: fmov s2, #-7.00000000 -; CHECK-NEXT: cbz x1, .LBB1_5 -; CHECK-NEXT: // %bb.1: // %for.body.preheader -; CHECK-NEXT: add x8, x0, #28 // =28 -; CHECK-NEXT: .LBB1_2: // %for.body -; CHECK-NEXT: // =>This Inner Loop Header: Depth=1 -; CHECK-NEXT: ldr s1, [x8, x1, lsl #2] -; CHECK-NEXT: scvtf s3, x1 -; CHECK-NEXT: fadd s3, s3, s0 -; CHECK-NEXT: fcmp s1, s3 -; CHECK-NEXT: b.gt .LBB1_6 -; CHECK-NEXT: // %bb.3: // %for.cond -; CHECK-NEXT: // in Loop: Header=BB1_2 Depth=1 -; CHECK-NEXT: add x1, x1, #1 // =1 -; CHECK-NEXT: cbnz x1, .LBB1_2 -; CHECK-NEXT: // %bb.4: -; CHECK-NEXT: mov v0.16b, v2.16b -; CHECK-NEXT: ret -; CHECK-NEXT: .LBB1_5: -; CHECK-NEXT: mov v0.16b, v2.16b -; CHECK-NEXT: ret -; CHECK-NEXT: .LBB1_6: // %cleanup4 -; CHECK-NEXT: mov v0.16b, v1.16b -; CHECK-NEXT: ret -entry: - %cmp14 = icmp eq i64 %start, 0 - br i1 %cmp14, label %cleanup4, label %for.body - -for.cond: ; preds = %for.body - %cmp = icmp eq i64 %inc, 0 - br i1 %cmp, label %cleanup4, label %for.body - -for.body: ; preds = %entry, %for.cond - %i.015 = phi i64 [ %inc, %for.cond ], [ %start, %entry ] - %add = add nsw i64 %i.015, 7 - %arrayidx = getelementptr inbounds float, float* %arr, i64 %add - %0 = load float, float* %arrayidx, align 4 - %conv = sitofp i64 %i.015 to float - %add1 = fadd float %conv, %threshold - %cmp2 = fcmp ogt float %0, %add1 - %inc = add nsw i64 %i.015, 1 - br i1 %cmp2, label %cleanup4, label %for.cond - -cleanup4: ; preds = %for.cond, %for.body, %entry - %1 = phi float [ -7.000000e+00, %entry ], [ %0, %for.body ], [ -7.000000e+00, %for.cond ] - ret float %1 -} |
