Introduce helper function for loading LowTCA

In facebook#117, I changed back the hardcoded `loadzlq` instructions
added by 1702bae to `emitLdPackedPtr` as it was before.
That was wrong because LowTCA will be 64 bits on PIE builds
but 32 bits otherwise, whereas `emitLdPackedPtr` uses the size
of `PackedPtr` which is always 64 bits in non-lowptr builds.
Pandemonium ensues as a result.

So, introduce a new helper here that uses the correct size of LowTCA.

Author: mszabo-wikia

hphp/runtime/vm/jit/code-gen-helpers.h

@@ -63,6 +63,10 @@ void emitLdPackedPtr(Vout& v, Vptr mem, Vreg reg) {
   ldLowPtrImpl(v, mem, reg, PackedPtr<T>::bits);
 }
 
+inline void emitLdTCAPtr(Vout& v, Vptr mem, Vreg reg) {
+  ldLowPtrImpl(v, mem, reg, LowTCA::bits);
+}
+
 /*
  * Store the LowPtr<T> in `reg' into `mem', with storage size `size'.
  */

hphp/runtime/vm/jit/irlower-call.cpp

@@ -156,7 +156,7 @@ void cgCall(IRLS& env, const IRInstruction* inst) {
     auto const pTabOff = safe_cast<int32_t>(Func::prologueTableOff());
     auto const ptrSize = safe_cast<int32_t>(sizeof(LowTCA));
     auto const dest = v.makeReg();
-    emitLdPackedPtr<uint8_t>(v, r_func_prologue_callee()[numArgsInclUnpack * ptrSize + pTabOff], dest);
+    emitLdTCAPtr(v, r_func_prologue_callee()[numArgsInclUnpack * ptrSize + pTabOff], dest);
     v << callphpr{dest, func_prologue_regs(withCtx)};
   } else {
     // It was not statically determined that the arguments are passed in a way
@@ -236,7 +236,7 @@ void cgCallFuncEntry(IRLS& env, const IRInstruction* inst) {
     // Load the FuncEntry address dynamically from the function.
     auto dest = v.makeReg();
     auto const funcEntryOff = safe_cast<int32_t>(Func::funcEntryOff());
-    emitLdPackedPtr<uint8_t>(v, callee[funcEntryOff], dest);
+    emitLdTCAPtr(v, callee[funcEntryOff], dest);
     // We have to use an ifdef instead of `if (use_lowptr)` here due to
     // funcIdOffset only being defined in non-lowptr mode.
 #ifdef USE_LOWPTR

hphp/runtime/vm/jit/unique-stubs.cpp

@@ -300,7 +300,7 @@ TCA emitFuncPrologueRedispatch(CodeBlock& cb, DataBlock& data, const char* name)
     ifThen(v, CC_LE, sf, [&] (Vout& v) {
       // Fast path (numArgs <= numNonVariadicParams). Call the numArgs prologue.
       auto const dest = v.makeReg();
-      emitLdPackedPtr<uint8_t>(v, callee[numArgs * ptrSize + pTabOff], dest);
+      emitLdTCAPtr(v, callee[numArgs * ptrSize + pTabOff], dest);
       v << jmpr{dest, func_prologue_regs(true)};
     });
 
@@ -364,7 +364,7 @@ TCA emitFuncPrologueRedispatch(CodeBlock& cb, DataBlock& data, const char* name)
 
     // Call the numNonVariadicParams + 1 prologue.
     auto const dest = v.makeReg();
-    emitLdPackedPtr<uint8_t>(
+    emitLdTCAPtr(
       v,
       Vreg(r_func_prologue_callee())[numNewArgs * ptrSize + pTabOff],
       dest
@@ -426,7 +426,7 @@ TCA emitFuncPrologueRedispatchUnpack(CodeBlock& main, CodeBlock& cold,
     auto const pTabOff = safe_cast<int32_t>(Func::prologueTableOff());
     auto const ptrSize = safe_cast<int32_t>(sizeof(LowTCA));
     auto const dest = v.makeReg();
-    emitLdPackedPtr<uint8_t>(v, callee[numNewArgs * ptrSize + pTabOff], dest);
+    emitLdTCAPtr(v, callee[numNewArgs * ptrSize + pTabOff], dest);
     v << tailcallstubr{dest, func_prologue_regs(true)};
   }, name);