Fix one of the things in the todo file, and get a bit closer to folding

constant offsets from statics into the address arithmetic.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24999 91177308-0d34-0410-b5e6-96231b3b80d8

4 files changed, 25 insertions, 39 deletions

diff --git a/lib/Target/PowerPC/PPCAsmPrinter.cpp b/lib/Target/PowerPC/PPCAsmPrinter.cpp
index a5eb9182ee..f75e4aae22 100644
--- a/lib/Target/PowerPC/PPCAsmPrinter.cpp
+++ b/lib/Target/PowerPC/PPCAsmPrinter.cpp
@@ -331,6 +331,7 @@ void PPCAsmPrinter::printOp(const MachineOperand &MO) {
     // Computing the address of a global symbol, not calling it.
     GlobalValue *GV = MO.getGlobal();
     std::string Name = Mang->getValueName(GV);
+    int offset = MO.getOffset();
 
     // External or weakly linked global variables need non-lazily-resolved stubs
     if (!PPCGenerateStaticCode) {
diff --git a/lib/Target/PowerPC/PPCISelDAGToDAG.cpp b/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
index 4d9abdc9a7..47f73abebf 100644
--- a/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
+++ b/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
@@ -927,7 +927,7 @@ SDOperand PPCDAGToDAGISel::Select(SDOperand Op) {
     break;
   }
   case ISD::AND: {
-    unsigned Imm;
+    unsigned Imm, Imm2;
     // If this is an and of a value rotated between 0 and 31 bits and then and'd
     // with a mask, emit rlwinm
     if (isIntImmediate(N->getOperand(1), Imm) && (isShiftedMask_32(Imm) ||
@@ -947,6 +947,20 @@ SDOperand PPCDAGToDAGISel::Select(SDOperand Op) {
       return CurDAG->SelectNodeTo(N, PPC::RLWINM, MVT::i32, Val, getI32Imm(SH),
                                   getI32Imm(MB), getI32Imm(ME));
     }
+    // ISD::OR doesn't get all the bitfield insertion fun.
+    // (and (or x, c1), c2) where isRunOfOnes(~(c1^c2)) is a bitfield insert
+    if (isIntImmediate(N->getOperand(1), Imm) && 
+        N->getOperand(0).getOpcode() == ISD::OR &&
+        isIntImmediate(N->getOperand(0).getOperand(1), Imm2)) {
+      unsigned SH, MB, ME;
+      Imm = ~(Imm^Imm2);
+      if (isRunOfOnes(Imm, MB, ME)) {
+        SDOperand Tmp1 = Select(N->getOperand(0).getOperand(0));
+        SDOperand Tmp2 = Select(N->getOperand(0).getOperand(1));
+        return CurDAG->getTargetNode(PPC::RLWIMI, MVT::i32, Tmp1, Tmp2,
+                                     getI32Imm(SH), getI32Imm(MB), getI32Imm(ME));
+      }
+    }
     
     // Other cases are autogenerated.
     break;
diff --git a/lib/Target/PowerPC/PPCISelLowering.cpp b/lib/Target/PowerPC/PPCISelLowering.cpp
index 565eb0a025..29507d00e3 100644
--- a/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -362,8 +362,9 @@ SDOperand PPCTargetLowering::LowerOperation(SDOperand Op, SelectionDAG &DAG) {
     return Lo;
   }
   case ISD::GlobalAddress: {
-    GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
-    SDOperand GA = DAG.getTargetGlobalAddress(GV, MVT::i32);
+    GlobalAddressSDNode *GSDN = cast<GlobalAddressSDNode>(Op);
+    GlobalValue *GV = GSDN->getGlobal();
+    SDOperand GA = DAG.getTargetGlobalAddress(GV, MVT::i32, GSDN->getOffset());
     SDOperand Zero = DAG.getConstant(0, MVT::i32);
 
     if (PPCGenerateStaticCode) {
diff --git a/lib/Target/PowerPC/README.txt b/lib/Target/PowerPC/README.txt
index 58805562a8..1ba89aa448 100644
--- a/lib/Target/PowerPC/README.txt
+++ b/lib/Target/PowerPC/README.txt
@@ -3,10 +3,15 @@ TODO:
 * implement do-loop -> bdnz transform
 * implement powerpc-64 for darwin
 * use stfiwx in float->int
-* be able to combine sequences like the following into 2 instructions:
+
+* Fold add and sub with constant into non-extern, non-weak addresses so this:
 	lis r2, ha16(l2__ZTV4Cell)
 	la r2, lo16(l2__ZTV4Cell)(r2)
 	addi r2, r2, 8
+becomes:
+	lis r2, ha16(l2__ZTV4Cell+8)
+	la r2, lo16(l2__ZTV4Cell+8)(r2)
+
 
 * Teach LLVM how to codegen this:
 unsigned short foo(float a) { return a; }
@@ -24,10 +29,6 @@ _foo:
         rlwinm r3, r2, 0, 16, 31
         blr
 
-and:
-  extern int X, Y; int* test(int C) { return C? &X : &Y; }
-as one load when using --enable-pic.
-
 * Support 'update' load/store instructions.  These are cracked on the G5, but
   are still a codesize win.
 
@@ -170,37 +171,6 @@ things like this, rather than forcing llvm to generate the equivalent
 
 ===-------------------------------------------------------------------------===
 
-Compile this (standard bitfield insert of a constant):
-void %test(uint* %tmp1) {
-        %tmp2 = load uint* %tmp1                ; <uint> [#uses=1]
-        %tmp5 = or uint %tmp2, 257949696                ; <uint> [#uses=1]
-        %tmp6 = and uint %tmp5, 4018143231              ; <uint> [#uses=1]
-        store uint %tmp6, uint* %tmp1
-        ret void
-}
-
-to:
-
-_test:
-        lwz r0,0(r3)
-        li r2,123
-        rlwimi r0,r2,21,3,10
-        stw r0,0(r3)
-        blr
-
-instead of:
-
-_test:
-        lis r2, -4225
-        lwz r4, 0(r3)
-        ori r2, r2, 65535
-        oris r4, r4, 3936
-        and r2, r4, r2
-        stw r2, 0(r3)
-        blr
-
-===-------------------------------------------------------------------------===
-
 Compile this:
 
 int %f1(int %a, int %b) {