[MemCpyOpt] Only replace memcpy with bitcast if address spaces match

[jprice]

Summary

Without this check, invalid bitcast instructions that change the address space are generated for the added test.

Similar to D30902 by @escha, just in a different MemCpyOpt optimisation.

[arsenm]

LGTM

[jprice]

Thanks for the review! Would you be able to commit this on my behalf?

[arsenm]

r299866

lib/Transforms/Scalar/MemCpyOptimizer.cpp

bool MemCpyOptPass::processByValArgument(CallSite CS, unsigned ArgNo) {
   // source of the memcpy to the alignment we need.  If we fail, we bail out.
   AssumptionCache &AC = LookupAssumptionCache();
   DominatorTree &DT = LookupDomTree();
   if (MDep->getAlignment() < ByValAlign &&
       getOrEnforceKnownAlignment(MDep->getSource(), ByValAlign, DL,
                                  CS.getInstruction(), &AC, &DT) < ByValAlign)
     return false;
 
+  // The address space of the memcpy source must match the byval argument
+  if (MDep->getSource()->getType()->getPointerAddressSpace() !=
+      ByValArg->getType()->getPointerAddressSpace())
+    return false;
+
   // Verify that the copied-from memory doesn't change in between the memcpy and
   // the byval call.
   //    memcpy(a <- b)
   //    *b = 42;
   //    foo(*a)
   // It would be invalid to transform the second memcpy into foo(*b).
   //
   // NOTE: This is conservative, it will stop on any read from the source loc,

test/Transforms/MemCpyOpt/memcpy.ll

define void @test4(i8 *%P) {
   %a = bitcast %1* %A to i8*
   call void @llvm.memcpy.p0i8.p0i8.i64(i8* %a, i8* %P, i64 8, i32 4, i1 false)
   call void @test4a(i8* align 1 byval %a)
   ret void
 ; CHECK-LABEL: @test4(
 ; CHECK-NEXT: call void @test4a(
 }
 
+; Make sure we don't remove the memcpy if the source address space doesn't match the byval argument
+define void @test4_addrspace(i8 addrspace(1)* %P) {
+  %A = alloca %1
+  %a = bitcast %1* %A to i8*
+  call void @llvm.memcpy.p0i8.p1i8.i64(i8* %a, i8 addrspace(1)* %P, i64 8, i32 4, i1 false)
+  call void @test4a(i8* align 1 byval %a)
+  ret void
+; CHECK-LABEL: @test4_addrspace(
+; CHECK: call void @llvm.memcpy.p0i8.p1i8.i64(
+; CHECK-NEXT: call void @test4a(
+}
+
 declare void @test4a(i8* align 1 byval)
 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i32, i1) nounwind
+declare void @llvm.memcpy.p0i8.p1i8.i64(i8* nocapture, i8 addrspace(1)* nocapture, i64, i32, i1) nounwind
 declare void @llvm.memcpy.p1i8.p1i8.i64(i8 addrspace(1)* nocapture, i8 addrspace(1)* nocapture, i64, i32, i1) nounwind
 
 %struct.S = type { i128, [4 x i8]}
 
 @sS = external global %struct.S, align 16
 
 declare void @test5a(%struct.S* align 16 byval) nounwind ssp