[llvm] a3a896d - [VE] Optimize LEA combinations

[Kazushi Marukawa via llvm-commits]

Author: Kazushi (Jam) Marukawa
Date: 2020-12-21T22:21:10+09:00
New Revision: a3a896d1cdc0fd2f87de4787120eaac08e69eb5f

URL: https://github.com/llvm/llvm-project/commit/a3a896d1cdc0fd2f87de4787120eaac08e69eb5f
DIFF: https://github.com/llvm/llvm-project/commit/a3a896d1cdc0fd2f87de4787120eaac08e69eb5f.diff

LOG: [VE] Optimize LEA combinations

Change to optimize references of elements of aggregate data.  Also
add regression tests.

Reviewed By: simoll

Differential Revision: https://reviews.llvm.org/D93627

Added: 
    llvm/test/CodeGen/VE/Scalar/lea-opt.ll

Modified: 
    llvm/lib/Target/VE/VEInstrInfo.td

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/VE/VEInstrInfo.td b/llvm/lib/Target/VE/VEInstrInfo.td
index fce3bf06b9d3..debd00ff6f96 100644
--- a/llvm/lib/Target/VE/VEInstrInfo.td
+++ b/llvm/lib/Target/VE/VEInstrInfo.td
@@ -1608,6 +1608,8 @@ def vehi_only : OutPatFrag<(ops node:$hi),
                            (LEASLzii 0, 0, $hi)>;
 def vehi_lo : OutPatFrag<(ops node:$hi, node:$lo),
                          (LEASLrii $lo, 0, $hi)>;
+def vehi_lo_imm : OutPatFrag<(ops node:$hi, node:$lo, node:$idx),
+                             (LEASLrii $lo, $idx, $hi)>;
 def vehi_baselo : OutPatFrag<(ops node:$base, node:$hi, node:$lo),
                              (LEASLrri $base, $lo, $hi)>;
 foreach type = [ "tblockaddress", "tconstpool", "texternalsym", "tglobaladdr",
@@ -1615,6 +1617,8 @@ foreach type = [ "tblockaddress", "tconstpool", "texternalsym", "tglobaladdr",
   def : Pat<(VElo !cast<SDNode>(type):$lo), (velo_only $lo)>;
   def : Pat<(VEhi !cast<SDNode>(type):$hi), (vehi_only $hi)>;
   def : Pat<(add (VEhi !cast<SDNode>(type):$hi), I64:$lo), (vehi_lo $hi, $lo)>;
+  def : Pat<(add (add (VEhi !cast<SDNode>(type):$hi), I64:$lo), simm7:$val),
+            (vehi_lo_imm $hi, $lo, (LO7 $val))>;
   def : Pat<(add I64:$base, (add (VEhi !cast<SDNode>(type):$hi), I64:$lo)),
             (vehi_baselo $base, $hi, $lo)>;
 }

diff  --git a/llvm/test/CodeGen/VE/Scalar/lea-opt.ll b/llvm/test/CodeGen/VE/Scalar/lea-opt.ll
new file mode 100644
index 000000000000..356b27653f4f
--- /dev/null
+++ b/llvm/test/CodeGen/VE/Scalar/lea-opt.ll
@@ -0,0 +1,63 @@
+; RUN: llc < %s -mtriple=ve | FileCheck %s
+; RUN: llc < %s -mtriple=ve -relocation-model=pic \
+; RUN:     | FileCheck %s --check-prefix=PIC
+
+;;; Tests for lea instruction and its optimizations
+
+%struct.buffer = type { i64, [1 x i8] }
+
+ at data = internal global i8 0, align 1
+ at buf = internal global %struct.buffer zeroinitializer, align 8
+
+; Function Attrs: norecurse nounwind readnone
+define nonnull i8* @lea_basic() {
+; CHECK-LABEL: lea_basic:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s0, data at lo
+; CHECK-NEXT:    and %s0, %s0, (32)0
+; CHECK-NEXT:    lea.sl %s0, data at hi(, %s0)
+; CHECK-NEXT:    b.l.t (, %s10)
+;
+; PIC-LABEL: lea_basic:
+; PIC:       # %bb.0:
+; PIC-NEXT:    st %s15, 24(, %s11)
+; PIC-NEXT:    st %s16, 32(, %s11)
+; PIC-NEXT:    lea %s15, _GLOBAL_OFFSET_TABLE_ at pc_lo(-24)
+; PIC-NEXT:    and %s15, %s15, (32)0
+; PIC-NEXT:    sic %s16
+; PIC-NEXT:    lea.sl %s15, _GLOBAL_OFFSET_TABLE_ at pc_hi(%s16, %s15)
+; PIC-NEXT:    lea %s0, data at gotoff_lo
+; PIC-NEXT:    and %s0, %s0, (32)0
+; PIC-NEXT:    lea.sl %s0, data at gotoff_hi(%s0, %s15)
+; PIC-NEXT:    ld %s16, 32(, %s11)
+; PIC-NEXT:    ld %s15, 24(, %s11)
+; PIC-NEXT:    b.l.t (, %s10)
+  ret i8* @data
+}
+
+; Function Attrs: norecurse nounwind readnone
+define i8* @lea_offset() {
+; CHECK-LABEL: lea_offset:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s0, buf at lo
+; CHECK-NEXT:    and %s0, %s0, (32)0
+; CHECK-NEXT:    lea.sl %s0, buf at hi(8, %s0)
+; CHECK-NEXT:    b.l.t (, %s10)
+;
+; PIC-LABEL: lea_offset:
+; PIC:       # %bb.0:
+; PIC-NEXT:    st %s15, 24(, %s11)
+; PIC-NEXT:    st %s16, 32(, %s11)
+; PIC-NEXT:    lea %s15, _GLOBAL_OFFSET_TABLE_ at pc_lo(-24)
+; PIC-NEXT:    and %s15, %s15, (32)0
+; PIC-NEXT:    sic %s16
+; PIC-NEXT:    lea.sl %s15, _GLOBAL_OFFSET_TABLE_ at pc_hi(%s16, %s15)
+; PIC-NEXT:    lea %s0, buf at gotoff_lo
+; PIC-NEXT:    and %s0, %s0, (32)0
+; PIC-NEXT:    lea.sl %s0, buf at gotoff_hi(, %s0)
+; PIC-NEXT:    lea %s0, 8(%s0, %s15)
+; PIC-NEXT:    ld %s16, 32(, %s11)
+; PIC-NEXT:    ld %s15, 24(, %s11)
+; PIC-NEXT:    b.l.t (, %s10)
+  ret i8* getelementptr inbounds (%struct.buffer, %struct.buffer* @buf, i64 0, i32 1, i64 0)
+}