[AArch64][GlobalISel] Libcall i128 srem/urem and scalarize more vecto…

…r types. This better handles i128 scalar and vector types, and allows some of the other odd-sized-vectors to successfully lower under GISel.
dmpolukhin · Sep 2, 2024 · 2fd8f85 · 2fd8f85
1 parent 77c86d0
commit 2fd8f85
Show file tree

Hide file tree

Showing 2 changed files with 899 additions and 456 deletions.
diff --git a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
@@ -196,12 +196,12 @@ AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)
 
   getActionDefinitionsBuilder({G_SREM, G_UREM, G_SDIVREM, G_UDIVREM})
       .lowerFor({s8, s16, s32, s64, v2s64, v4s32, v2s32})
+      .libcallFor({s128})
       .widenScalarOrEltToNextPow2(0)
-      .clampScalarOrElt(0, s32, s64)
+      .minScalarOrElt(0, s32)
       .clampNumElements(0, v2s32, v4s32)
       .clampNumElements(0, v2s64, v2s64)
-      .moreElementsToNextPow2(0);
-
+      .scalarize(0);
 
   getActionDefinitionsBuilder({G_SMULO, G_UMULO})
       .widenScalarToNextPow2(0, /*Min = */ 32)