[PATCH] D35750: [x86] Teach the x86 backend about general fast rep+movs and rep+stos features of modern x86 CPUs, and use this feature to drastically reduce the number of places we actually emit memset and memcpy library calls.

[Eric Christopher via Phabricator via llvm-commits]

echristo added inline comments.


================
Comment at: lib/Target/X86/X86.td:287
+// lowerings for string operations when calling the library function would have
+// too high of cost.
+def FeatureFastRepStrOps
----------------
"too high of a cost"


================
Comment at: lib/Target/X86/X86SelectionDAGInfo.cpp:73
+
+static std::pair<SDValue, MVT> getUnscaledSizeAndVT(SelectionDAG &DAG,
+                                                    const SDLoc &DL,
----------------
Comment please on what this is for :)


================
Comment at: lib/Target/X86/X86SelectionDAGInfo.cpp:115
+
+static std::pair<SDValue, unsigned> getWidenedValueAndReg(SelectionDAG &DAG,
+                                                          const SDLoc &DL,
----------------
Ditto.


================
Comment at: lib/Target/X86/X86SelectionDAGInfo.cpp:164-171
   if ((Align & 3) != 0 || !ConstantSize ||
       ConstantSize->getZExtValue() > Subtarget.getMaxInlineSizeThreshold()) {
-    // Check to see if there is a specialized entry-point for memory zeroing.
+    // When we have a fast REP+STOS CPU and either have ERMSB + 16-byte
+    // alignment or PIC overhead for a library call, bypass the library call
+    // entirely.
+    if (Subtarget.hasFastRepStrOps() &&
+        (Subtarget.isPositionIndependent() ||
----------------
This seems like this all wants to be subsumed in a couple of different checks?

i.e. I don't think with the rep str ops that the max inline size threshold is as important anymore. I worry that we don't actually care about whether or not pic here and should probably just use the inline expansion either way.


================
Comment at: lib/Target/X86/X86SelectionDAGInfo.cpp:319-327
+  if (!ConstantSize ||
+      (!AlwaysInline &&
+       ConstantSize->getZExtValue() > Subtarget.getMaxInlineSizeThreshold())) {
+    // When we have a fast REP+STOS CPU and either have ERMSB + 16-byte
+    // alignment or PIC overhead for a library call, bypass the library call
+    // entirely.
+    if (Subtarget.hasFastRepStrOps() &&
----------------
Ditto.


https://reviews.llvm.org/D35750