[llvm-commits] [llvm] r171608 - in /llvm/trunk: lib/Target/X86/X86InstrArithmetic.td test/CodeGen/X86/early-ifcvt.ll

[Michael Gottesman]

Thanks Craig!

(I tried to recommit several times only to be thwarted by llvm.org being down = ( ).

Michael

On Jan 5, 2013, at 2:39 AM, Craig Topper <craig.topper at gmail.com> wrote:

> Author: ctopper
> Date: Sat Jan  5 01:39:25 2013
> New Revision: 171608
> 
> URL: http://llvm.org/viewvc/llvm-project?rev=171608&view=rev
> Log:
> Recommit r171461 which was incorrectly reverted. Mark DIV/IDIV instructions hasSideEffects=1 because they can trap when dividing by 0. This is needed to keep early if conversion from moving them across basic blocks.
> 
> Modified:
>    llvm/trunk/lib/Target/X86/X86InstrArithmetic.td
>    llvm/trunk/test/CodeGen/X86/early-ifcvt.ll
> 
> Modified: llvm/trunk/lib/Target/X86/X86InstrArithmetic.td
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86InstrArithmetic.td?rev=171608&r1=171607&r2=171608&view=diff
> ==============================================================================
> --- llvm/trunk/lib/Target/X86/X86InstrArithmetic.td (original)
> +++ llvm/trunk/lib/Target/X86/X86InstrArithmetic.td Sat Jan  5 01:39:25 2013
> @@ -266,7 +266,7 @@
> 
> 
> // unsigned division/remainder
> -let hasSideEffects = 0 in {
> +let hasSideEffects = 1 in { // so that we don't speculatively execute
> let Defs = [AL,EFLAGS,AX], Uses = [AX] in
> def DIV8r  : I<0xF6, MRM6r, (outs),  (ins GR8:$src),    // AX/r8 = AL,AH
>                "div{b}\t$src", [], IIC_DIV8_REG>;
> 
> Modified: llvm/trunk/test/CodeGen/X86/early-ifcvt.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/early-ifcvt.ll?rev=171608&r1=171607&r2=171608&view=diff
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/early-ifcvt.ll (original)
> +++ llvm/trunk/test/CodeGen/X86/early-ifcvt.ll Sat Jan  5 01:39:25 2013
> @@ -142,3 +142,34 @@
> }
> 
> declare void @BZ2_bz__AssertH__fail()
> +
> +; Make sure we don't speculate on div/idiv instructions
> +; CHECK: test_idiv
> +; CHECK-NOT: cmov
> +define i32 @test_idiv(i32 %a, i32 %b) nounwind uwtable readnone ssp {
> +  %1 = icmp eq i32 %b, 0
> +  br i1 %1, label %4, label %2
> +
> +; <label>:2                                       ; preds = %0
> +  %3 = sdiv i32 %a, %b
> +  br label %4
> +
> +; <label>:4                                       ; preds = %0, %2
> +  %5 = phi i32 [ %3, %2 ], [ %a, %0 ]
> +  ret i32 %5
> +}
> +
> +; CHECK: test_div
> +; CHECK-NOT: cmov
> +define i32 @test_div(i32 %a, i32 %b) nounwind uwtable readnone ssp {
> +  %1 = icmp eq i32 %b, 0
> +  br i1 %1, label %4, label %2
> +
> +; <label>:2                                       ; preds = %0
> +  %3 = udiv i32 %a, %b
> +  br label %4
> +
> +; <label>:4                                       ; preds = %0, %2
> +  %5 = phi i32 [ %3, %2 ], [ %a, %0 ]
> +  ret i32 %5
> +}
> 
> 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits