Big shift test

[chfast]

Summary

A test for patch D4978

[lhames]

Hi Paweł,

Is this JIT specific? What platforms are you seeing problems on? I suspect it should just be a CodeGen test for those platforms.

• Lang.

[loladiro]

This is a non-JIT-specific codegen bug (see the linked patch). I only noticed it on X86, but I suspect other platforms may have similar semantics, so it may be an issue there anyway. The fix certainly isn't architecture specific, so it might make sense to have this run on all platforms.

[chfast]

I agree with @loladiro, it looks like generic CodeGen bug. Placing it in ExecutionEngine/MCJIT wasn't the best idea, but I've found that lli based test are there. Should I move it to some better place? Like test/CodeGen/Generic?

[lhames]

It's tricky. This may be a generic codegen bug, but it will manifest differently on different platforms. I'd be inclined to pick a popular target where it's currently broken (X86?), fix it there, then add the test to test/CodeGen/<Target>.

[chfast]

In D7752#140854, @lhames wrote:

It's tricky. This may be a generic codegen bug, but it will manifest differently on different platforms. I'd be inclined to pick a popular target where it's currently broken (X86?), fix it there, then add the test to test/CodeGen/<Target>.

This test is a quite different from other Target tests because it checks runtime result. Maybe we should come back to that when the actual bug is fixed.

[loladiro]

I'd like to include this test case with the bugfix. Can we decide on where it should live?

[chfast]

In D7752#150579, @loladiro wrote:

I'd like to include this test case with the bugfix. Can we decide on where it should live?

I would call it a "runtime" test. But I haven't found similar tests in test/ directory.

[chfast]

A new test that checks X86_64 assembly

[loladiro]

LGTM. If you want, you can combine this with the fix and commit. Otherwise I'll get to it later today or tomorrow.

[chfast]

Commit rL235370

test/CodeGen/X86/shift-i256.ll

-; RUN: llc < %s -march=x86
-; RUN: llc < %s -march=x86-64
+; RUN: llc < %s -march=x86        | FileCheck %s
+; RUN: llc < %s -march=x86-64 -O0 | FileCheck %s -check-prefix=CHECK-X64
+; RUN: llc < %s -march=x86-64 -O2 | FileCheck %s -check-prefix=CHECK-X64
 
-define void @t(i256 %x, i256 %a, i256* nocapture %r) nounwind readnone {
+; CHECK-LABEL: shift1
+define void @shift1(i256 %x, i256 %a, i256* nocapture %r) nounwind readnone {
 entry:
   %0 = ashr i256 %x, %a
   store i256 %0, i256* %r
         ret void
 }
+
+; CHECK-LABEL: shift2
+define i256 @shift2(i256 %c) nounwind
+{
+  %b = shl i256 1, %c  ; %c must not be a constant
+  ; Special case when %c is 0:
+  ; CHECK-X64: testb [[REG:%r[0-9]+b]], [[REG]]
+  ; CHECK-X64: cmoveq
+  ret i256 %b
+}