[PATCH] D110667: [AArch64][GlobalISel] Run overlapping_and after legalization

Jessica Paquette via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Tue Sep 28 16:03:04 PDT 2021


paquette created this revision.
paquette added reviewers: aemerson, jroelofs.
Herald added subscribers: hiraditya, kristof.beyls, rovka.
paquette requested review of this revision.
Herald added a project: LLVM.

When we have code with truncates, those truncates may be changed into G_ANDs with constants. These may, in turn, feed into other G_AND instructions.

Running this combine post-legalize allows us to optimize examples like this one:

https://godbolt.org/z/zrGY4dfEW

SDAG currently optimizes the example above so that there is only one `and`. GISel doesn't optimize it, because the G_AND we'd optimize here is translated as a G_TRUNC. Later, that G_TRUNC is turned into a G_AND during legalization.


https://reviews.llvm.org/D110667

Files:
  llvm/lib/Target/AArch64/AArch64Combine.td
  llvm/test/CodeGen/AArch64/GlobalISel/opt-overlapping-and-postlegalize.mir


Index: llvm/test/CodeGen/AArch64/GlobalISel/opt-overlapping-and-postlegalize.mir
===================================================================
--- /dev/null
+++ llvm/test/CodeGen/AArch64/GlobalISel/opt-overlapping-and-postlegalize.mir
@@ -0,0 +1,30 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -debugify-and-strip-all-safe -mtriple arm64-apple-ios -O0 -run-pass=aarch64-postlegalizer-combiner --aarch64postlegalizercombinerhelper-only-enable-rule="overlapping_and" -global-isel -verify-machineinstrs %s -o - | FileCheck %s
+# REQUIRES: asserts
+
+# Test running the overlapping_and combine post-legalization.
+
+...
+---
+name:            test
+legalized:       true
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $w0
+    ; CHECK-LABEL: name: test
+    ; CHECK: liveins: $w0
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: %copy:_(s32) = COPY $w0
+    ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 192
+    ; CHECK-NEXT: %and2:_(s32) = G_AND %copy, [[C]]
+    ; CHECK-NEXT: $w0 = COPY %and2(s32)
+    ; CHECK-NEXT: RET_ReallyLR implicit $w0
+    %copy:_(s32) = COPY $w0
+    %cst_neg_64:_(s32) = G_CONSTANT i32 -64
+    %and1:_(s32) = G_AND %copy, %cst_neg_64
+    %cst_255:_(s32) = G_CONSTANT i32 255
+    %and2:_(s32) = G_AND %and1, %cst_255
+    $w0 = COPY %and2(s32)
+    RET_ReallyLR implicit $w0
+...
Index: llvm/lib/Target/AArch64/AArch64Combine.td
===================================================================
--- llvm/lib/Target/AArch64/AArch64Combine.td
+++ llvm/lib/Target/AArch64/AArch64Combine.td
@@ -213,6 +213,6 @@
                         icmp_to_true_false_known_bits, merge_unmerge,
                         select_combines, fold_merge_to_zext,
                         constant_fold, identity_combines,
-                        ptr_add_immed_chain]> {
+                        ptr_add_immed_chain, overlapping_and]> {
   let DisableRuleOption = "aarch64postlegalizercombiner-disable-rule";
 }


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D110667.375730.patch
Type: text/x-patch
Size: 2000 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20210928/715f4dc5/attachment.bin>


More information about the llvm-commits mailing list