[AArch64][GlobalISel] Run redundant_sext_inreg in the post-legalizer combiner

[paquette]

Summary

This is to ensure that we can eliminate G_ASSERT_SEXT.

In a follow-up patch, I'm going to make CallLowering emit G_ASSERT_SEXT for signext parameters.

[aemerson]

llvm/test/CodeGen/AArch64/GlobalISel/combine-sext-trunc-sextload.mir

2–3

What happens if we don't restrict the combiner to this?

[paquette]

llvm/test/CodeGen/AArch64/GlobalISel/combine-sext-trunc-sextload.mir

2–3

test_combine_sext_of_sextload_not_matching ends up getting combined.

computeNumSignBits should give us 17 for the G_SEXTLOAD in that test. That's less than 24, so the G_SEXT_INREG ends up being eliminated.

[aemerson]

llvm/test/CodeGen/AArch64/GlobalISel/combine-sext-trunc-sextload.mir

2–3

That test is a bit redundant, I think it's fine to show the G_SEXT_INREG being combined away.

llvm/lib/Target/AArch64/AArch64Combine.td

 // Post-legalization combines which are primarily optimizations.
 def AArch64PostLegalizerCombinerHelper
     : GICombinerHelper<"AArch64GenPostLegalizerCombinerHelper",
                        [copy_prop, erase_undef_store, combines_for_extload,
                         sext_trunc_sextload,
                         hoist_logic_op_with_same_opcode_hands,
                         redundant_and, xor_of_and_with_same_reg,
                         extractvecelt_pairwise_add, redundant_or,
-                        mul_const]> {
+                        mul_const, redundant_sext_inreg]> {
   let DisableRuleOption = "aarch64postlegalizercombiner-disable-rule";
 }

llvm/test/CodeGen/AArch64/GlobalISel/combine-sext-trunc-sextload.mir

 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
 # RUN: llc -mtriple aarch64 -run-pass=aarch64-postlegalizer-combiner -verify-machineinstrs %s -o - | FileCheck %s
+
 ---
 name:            test_combine_sext_trunc_of_sextload
 legalized:       true
 tracksRegLiveness: true
 body: |
   bb.0.entry:
     liveins: $x0
     ; CHECK-LABEL: name: test_combine_sext_trunc_of_sextload

 body: |
   bb.0.entry:
     liveins: $x0
     ; Here we're trying to extend from a smaller value than was extended in the load.
     ; CHECK-LABEL: name: test_combine_sext_of_sextload_not_matching
     ; CHECK: liveins: $x0
     ; CHECK: [[COPY:%[0-9]+]]:_(p0) = COPY $x0
     ; CHECK: [[SEXTLOAD:%[0-9]+]]:_(s32) = G_SEXTLOAD [[COPY]](p0) :: (load 2)
-    ; CHECK: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[SEXTLOAD]], 24
-    ; CHECK: $w0 = COPY [[SEXT_INREG]](s32)
+    ; CHECK: $w0 = COPY [[SEXTLOAD]](s32)
     %0:_(p0) = COPY $x0
     %1:_(s32) = G_SEXTLOAD %0:_(p0) :: (load 2)
     %2:_(s32) = COPY %1:_(s32)
     %3:_(s32) = G_SEXT_INREG %2:_(s32), 24
     $w0 = COPY %3(s32)
 ...

llvm/test/CodeGen/AArch64/GlobalISel/postlegalizer-combiner-redundant-sextinreg.mir

+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple aarch64 -run-pass=aarch64-postlegalizer-combiner -verify-machineinstrs %s -o - | FileCheck %s
+
+...
+---
+name:            sextload
+legalized:       true
+tracksRegLiveness: true
+body: |
+  bb.0.entry:
+    liveins: $x0
+
+    ; CHECK-LABEL: name: sextload
+    ; CHECK: liveins: $x0
+    ; CHECK: %x0:_(p0) = COPY $x0
+    ; CHECK: %sextload:_(s32) = G_SEXTLOAD %x0(p0) :: (load 2)
+    ; CHECK: $w0 = COPY %sextload(s32)
+    %x0:_(p0) = COPY $x0
+    %sextload:_(s32) = G_SEXTLOAD %x0:_(p0) :: (load 2)
+    %sext_inreg:_(s32) = G_SEXT_INREG %sextload:_(s32), 24
+    $w0 = COPY %sext_inreg(s32)
+...
+---
+name:            assert_sext
+legalized:       true
+tracksRegLiveness: true
+body: |
+  bb.0.entry:
+    liveins: $w0
+    ; The G_ASSERT_SEXT says we already sign extended from 24 bits, so the
+    ; G_SEXT_INREG is not necessary.
+
+    ; CHECK-LABEL: name: assert_sext
+    ; CHECK: liveins: $w0
+    ; CHECK: %w0:_(s32) = COPY $w0
+    ; CHECK: %assert_sext:_(s32) = G_ASSERT_SEXT %w0, 24
+    ; CHECK: $w0 = COPY %assert_sext(s32)
+    %w0:_(s32) = COPY $w0
+    %assert_sext:_(s32) = G_ASSERT_SEXT %w0, 24
+    %sext_inreg:_(s32) = G_SEXT_INREG %assert_sext:_(s32), 24
+    $w0 = COPY %sext_inreg(s32)
+...
+---
+name:            assert_sext_greater_width
+legalized:       true
+tracksRegLiveness: true
+body: |
+  bb.0.entry:
+    liveins: $w0
+    ; The G_ASSERT_SEXT is from a wider width than the G_SEXT_INREG, so we
+    ; can't eliminate it.
+
+    ; CHECK-LABEL: name: assert_sext_greater_width
+    ; CHECK: liveins: $w0
+    ; CHECK: %w0:_(s32) = COPY $w0
+    ; CHECK: %assert_sext:_(s32) = G_ASSERT_SEXT %w0, 24
+    ; CHECK: %sext_inreg:_(s32) = G_SEXT_INREG %assert_sext, 12
+    ; CHECK: $w0 = COPY %sext_inreg(s32)
+    %w0:_(s32) = COPY $w0
+    %assert_sext:_(s32) = G_ASSERT_SEXT %w0, 24
+    %sext_inreg:_(s32) = G_SEXT_INREG %assert_sext:_(s32), 12
+    $w0 = COPY %sext_inreg(s32)
+...
+---
+name:            assert_sext_smaller_width
+legalized:       true
+tracksRegLiveness: true
+body: |
+  bb.0.entry:
+    liveins: $w0
+    ; The G_ASSERT_SEXT is from a smaller width, so the G_SEXT_INREG is not
+    ; necessary.
+
+    ; CHECK-LABEL: name: assert_sext_smaller_width
+    ; CHECK: liveins: $w0
+    ; CHECK: %w0:_(s32) = COPY $w0
+    ; CHECK: %assert_sext:_(s32) = G_ASSERT_SEXT %w0, 8
+    ; CHECK: $w0 = COPY %assert_sext(s32)
+    %w0:_(s32) = COPY $w0
+    %assert_sext:_(s32) = G_ASSERT_SEXT %w0, 8
+    %sext_inreg:_(s32) = G_SEXT_INREG %assert_sext:_(s32), 12
+    $w0 = COPY %sext_inreg(s32)
+...