Add tests for shrink wrapping and VLAs

[chill]

Summary

Split the test from https://reviews.llvm.org/D45524 and move it to the relevant target directories.

[thegameg]

Sorry for not looking thoroughly at the test in the previous patch.

test/CodeGen/AArch64/shrink-wrapping-vla.ll

17

Hmm, I'm pretty sure this is not needed. Did you encounter any issues without this line? My understanding was that the test failed because it lived in Generic instead of the target directory.

19

Related to both tests: are these attributes actually needed? (the #0, #1)

[thegameg]

local_unnamed_addr is probably not needed either.

Thanks! LGTM. You can just remove the local_unnamed_addrs and commit this.

test/CodeGen/AArch64/shrink-wrapping-vla.ll

+; Test shrink wrapping placement is correct with respect to calls to llvm.{stacksave,stackrestore}
+
+; void f(int n, int x[]) {
+;   if (n < 0)
+;     return;
+;
+;  int a[n];
+;
+;  for (int i = 0; i < n; i++)
+;    a[i] = x[n - i - 1];
+;
+;  for (int i = 0; i < n; i++)
+;    x[i] = a[i] + 1;
+; }
+;
+; RUN: llc -mtriple aarch64-linux %s -o - | FileCheck %s
+; REQUIRES: arm-registered-target
+
+define dso_local void @f(i32 %n, i32* nocapture %x) local_unnamed_addr #0 {
+entry:
+  %cmp = icmp slt i32 %n, 0
+  br i1 %cmp, label %return, label %if.end
+
+if.end:                                           ; preds = %entry
+  %0 = zext i32 %n to i64
+  %1 = tail call i8* @llvm.stacksave()
+  %vla = alloca i32, i64 %0, align 16
+  %cmp132 = icmp eq i32 %n, 0
+  br i1 %cmp132, label %for.cond.cleanup8, label %for.body.lr.ph
+
+for.body.lr.ph:                                   ; preds = %if.end
+  %sub = add i32 %n, -1
+  br label %for.body
+
+for.cond6.preheader:                              ; preds = %for.body
+  %cmp730 = icmp sgt i32 %n, 0
+  br i1 %cmp730, label %for.body9, label %for.cond.cleanup8
+
+for.body:                                         ; preds = %for.body, %for.body.lr.ph
+  %indvars.iv34 = phi i64 [ 0, %for.body.lr.ph ], [ %indvars.iv.next35, %for.body ]
+  %2 = trunc i64 %indvars.iv34 to i32
+  %sub2 = sub i32 %sub, %2
+  %idxprom = sext i32 %sub2 to i64
+  %arrayidx = getelementptr inbounds i32, i32* %x, i64 %idxprom
+  %3 = load i32, i32* %arrayidx, align 4
+  %arrayidx4 = getelementptr inbounds i32, i32* %vla, i64 %indvars.iv34
+  store i32 %3, i32* %arrayidx4, align 4
+  %indvars.iv.next35 = add nuw nsw i64 %indvars.iv34, 1
+  %exitcond37 = icmp eq i64 %indvars.iv.next35, %0
+  br i1 %exitcond37, label %for.cond6.preheader, label %for.body
+
+for.cond.cleanup8:                                ; preds = %for.body9, %if.end, %for.cond6.preheader
+  tail call void @llvm.stackrestore(i8* %1)
+  br label %return
+
+for.body9:                                        ; preds = %for.cond6.preheader, %for.body9
+  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body9 ], [ 0, %for.cond6.preheader ]
+  %arrayidx11 = getelementptr inbounds i32, i32* %vla, i64 %indvars.iv
+  %4 = load i32, i32* %arrayidx11, align 4
+  %add = add nsw i32 %4, 1
+  %arrayidx13 = getelementptr inbounds i32, i32* %x, i64 %indvars.iv
+  store i32 %add, i32* %arrayidx13, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond = icmp eq i64 %indvars.iv.next, %0
+  br i1 %exitcond, label %for.cond.cleanup8, label %for.body9
+
+return:                                           ; preds = %entry, %for.cond.cleanup8
+  ret void
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.stacksave() #1
+
+; Function Attrs: nounwind
+declare void @llvm.stackrestore(i8*) #1
+
+; Check that llvm.stackrestore() happens before CSRs are popped off the stack
+
+; CHECK-LABEL: f
+
+; CHECK:      stp x29, x30, [sp, #-16]!
+; CHECK-NEXT: mov x29, sp
+
+; VLA allocation
+; CHECK: add [[X1:x[0-9]+]], [[X1]], #15
+; CHECK: mov [[X2:x[0-9]+]], sp
+; CHECK: and [[X1]], [[X1]], #0x7fffffff0
+; Saving the SP via llvm.stacksave()
+; CHECK: mov [[SAVE:x[0-9]+]], sp
+; CHECK: sub [[X2]], [[X2]], [[X1]]
+
+; The next instruction comes from llvm.stackrestore()
+; CHECK:      mov sp, [[SAVE]]
+; Epilogue
+; CHECK-NEXT: mov sp, x29
+; CHECK-NEXT: ldp x29, x30, [sp], #16

test/CodeGen/X86/shrink-wrapping-vla.ll

+; Test shrink wrapping placement is correct with respect to calls to llvm.{stacksave,stackrestore}
+
+; void f(int n, int x[]) {
+;   if (n < 0)
+;     return;
+;
+;  int a[n];
+;
+;  for (int i = 0; i < n; i++)
+;    a[i] = x[n - i - 1];
+;
+;  for (int i = 0; i < n; i++)
+;    x[i] = a[i] + 1;
+; }
+;
+; RUN: llc -mtriple x86_64-linux %s -o - | FileCheck %s --check-prefix=CHECK
+; REQUIRES: x86-registered-target
+
+
+define dso_local void @f(i32 %n, i32* nocapture %x) local_unnamed_addr #0 {
+entry:
+  %cmp = icmp slt i32 %n, 0
+  br i1 %cmp, label %return, label %if.end
+
+if.end:                                           ; preds = %entry
+  %0 = zext i32 %n to i64
+  %1 = tail call i8* @llvm.stacksave()
+  %vla = alloca i32, i64 %0, align 16
+  %cmp132 = icmp eq i32 %n, 0
+  br i1 %cmp132, label %for.cond.cleanup8, label %for.body.lr.ph
+
+for.body.lr.ph:                                   ; preds = %if.end
+  %sub = add i32 %n, -1
+  br label %for.body
+
+for.cond6.preheader:                              ; preds = %for.body
+  %cmp730 = icmp sgt i32 %n, 0
+  br i1 %cmp730, label %for.body9, label %for.cond.cleanup8
+
+for.body:                                         ; preds = %for.body, %for.body.lr.ph
+  %indvars.iv34 = phi i64 [ 0, %for.body.lr.ph ], [ %indvars.iv.next35, %for.body ]
+  %2 = trunc i64 %indvars.iv34 to i32
+  %sub2 = sub i32 %sub, %2
+  %idxprom = sext i32 %sub2 to i64
+  %arrayidx = getelementptr inbounds i32, i32* %x, i64 %idxprom
+  %3 = load i32, i32* %arrayidx, align 4
+  %arrayidx4 = getelementptr inbounds i32, i32* %vla, i64 %indvars.iv34
+  store i32 %3, i32* %arrayidx4, align 4
+  %indvars.iv.next35 = add nuw nsw i64 %indvars.iv34, 1
+  %exitcond37 = icmp eq i64 %indvars.iv.next35, %0
+  br i1 %exitcond37, label %for.cond6.preheader, label %for.body
+
+for.cond.cleanup8:                                ; preds = %for.body9, %if.end, %for.cond6.preheader
+  tail call void @llvm.stackrestore(i8* %1)
+  br label %return
+
+for.body9:                                        ; preds = %for.cond6.preheader, %for.body9
+  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body9 ], [ 0, %for.cond6.preheader ]
+  %arrayidx11 = getelementptr inbounds i32, i32* %vla, i64 %indvars.iv
+  %4 = load i32, i32* %arrayidx11, align 4
+  %add = add nsw i32 %4, 1
+  %arrayidx13 = getelementptr inbounds i32, i32* %x, i64 %indvars.iv
+  store i32 %add, i32* %arrayidx13, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond = icmp eq i64 %indvars.iv.next, %0
+  br i1 %exitcond, label %for.cond.cleanup8, label %for.body9
+
+return:                                           ; preds = %entry, %for.cond.cleanup8
+  ret void
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.stacksave() #1
+
+; Function Attrs: nounwind
+declare void @llvm.stackrestore(i8*) #1
+
+; Check that llvm.stackrestore() happens before CSRs are popped off the stack
+
+; CHECK-LABEL: f
+
+; CHECK: pushq %rbp
+; CHECK: movq %rsp, %rbp
+
+; Saving the SP via llvm.stacksave()
+; CHECK: movq %rsp, [[SAVE:%r[a-z0-9]+]]
+; VLA allocation
+; CHECK: movq %rsp, [[TMP:%r[a-z0-9]+]]
+; CHECK: subq %r{{[a-z0-9]+}}, [[TMP]]
+; CHECK: movq [[TMP]], %rsp
+
+; The next instruction comes from llvm.stackrestore()
+; CHECK:      movq [[SAVE]], %rsp
+; Epilogue
+; CHECK-NEXT: movq %rbp, %rsp
+; CHECK-NEXT: popq %rbp