[llvm-branch-commits] [llvm] b351efc - [PowerPC] Do not emit XXSPLTI32DX for sub 64-bit constants

[Tom Stellard via llvm-branch-commits]

Author: Nemanja Ivanovic
Date: 2021-02-03T14:07:33-08:00
New Revision: b351efcae08a59c0cafa123a92b24c5f2300202b

URL: https://github.com/llvm/llvm-project/commit/b351efcae08a59c0cafa123a92b24c5f2300202b
DIFF: https://github.com/llvm/llvm-project/commit/b351efcae08a59c0cafa123a92b24c5f2300202b.diff

LOG: [PowerPC] Do not emit XXSPLTI32DX for sub 64-bit constants

If the APInt returned by BuildVectorSDNode::isConstantSplat() is narrower than
64 bits, the result produced by XXSPLTI32DX is incorrect. The result returned
by the function appears to be incorrect and we'll investigate/fix it in a
follow-up commit. However, since this causes miscompiles, we must
temporarily disable emitting this instruction for such values.

(cherry picked from commit 54e570d94af995ff58287a8288389641910a8239)

Added: 
    

Modified: 
    llvm/lib/Target/PowerPC/PPCISelLowering.cpp
    llvm/test/CodeGen/PowerPC/p10-splatImm32.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
index 9215c17cb94b..663ee15db11e 100644
--- a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -8613,7 +8613,8 @@ SDValue PPCTargetLowering::LowerBUILD_VECTOR(SDValue Op,
           PPCISD::XXSPLTI_SP_TO_DP, dl, MVT::v2f64,
           DAG.getTargetConstant(APSplatBits.getZExtValue(), dl, MVT::i32));
       return DAG.getBitcast(Op.getValueType(), SplatNode);
-    } else { // We may lose precision, so we have to use XXSPLTI32DX.
+    } else if (APSplatBits.getBitWidth() == 64) {
+      // We may lose precision, so we have to use XXSPLTI32DX.
 
       uint32_t Hi =
           (uint32_t)((APSplatBits.getZExtValue() & 0xFFFFFFFF00000000LL) >> 32);

diff  --git a/llvm/test/CodeGen/PowerPC/p10-splatImm32.ll b/llvm/test/CodeGen/PowerPC/p10-splatImm32.ll
index 420a96dc1495..081cae729acf 100644
--- a/llvm/test/CodeGen/PowerPC/p10-splatImm32.ll
+++ b/llvm/test/CodeGen/PowerPC/p10-splatImm32.ll
@@ -100,3 +100,25 @@ define dso_local <8 x i16> @test_xxsplti32dx_9() {
 entry:
   ret <8 x i16> <i16 291, i16 undef, i16 undef, i16 364, i16 undef, i16 1, i16 173, i16 undef>
 }
+
+define dso_local <16 x i8> @test_xxsplti32dx_10() {
+; CHECK-LABEL: test_xxsplti32dx_10:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xxlxor vs34, vs34, vs34
+; CHECK-NEXT:    xxsplti32dx vs34, 0, 1207959552
+; CHECK-NEXT:    blr
+entry:
+  ret <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 72, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 72>
+}
+
+; FIXME: It appears that there is something wrong with the computation
+;        of the 64-bit constant to splat so we cannot emit xxsplti32dx for
+;        this test case for now.
+define dso_local <16 x i8> @constSplatBug() {
+; CHECK-LABEL: constSplatBug:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    plxv vs34, .LCPI10_0 at PCREL(0), 1
+; CHECK-NEXT:    blr
+entry:
+  ret <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 71, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 71>
+}