[PowerPC] Implement the vector extend sign instruction pattern match

Power9 has instructions to implement the semantics of SIGN_EXTEND_INREG for vector type.
Mark it as legal and add the match pattern.

Differential Revision: https://reviews.llvm.org/D69601

diff --git a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
index 0159e8c63c8547b1d25f520413695baa72407f2a..4d9288ce7982434c6d5db896c18bf229379f73de 100644 (file)
--- a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -926,6 +926,14 @@ PPCTargetLowering::PPCTargetLowering(const PPCTargetMachine &TM,
     if (Subtarget.hasP9Altivec()) {
       setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v8i16, Custom);
       setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v16i8, Custom);
+
+      setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v4i8,  Legal);
+      setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v4i16, Legal);
+      setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v4i32, Legal);
+      setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v2i8,  Legal);
+      setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v2i16, Legal);
+      setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v2i32, Legal);
+      setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v2i64, Legal);
     }
   }
 

diff --git a/llvm/lib/Target/PowerPC/PPCInstrAltivec.td b/llvm/lib/Target/PowerPC/PPCInstrAltivec.td
index fd3fc2af232795ca40570968c96dc3b5898b0a83..26c506ef4e28d6caed6ccdf5cf3f613961ad305f 100644 (file)
--- a/llvm/lib/Target/PowerPC/PPCInstrAltivec.td
+++ b/llvm/lib/Target/PowerPC/PPCInstrAltivec.td
@@ -1410,6 +1410,12 @@ let isCodeGenOnly = 1 in {
   def VEXTSW2Ds : VX_VT5_EO5_VB5s<1538, 26, "vextsw2d", []>;
 }
 
+def : Pat<(v4i32 (sext_inreg v4i32:$VRB, v4i8)), (v4i32 (VEXTSB2W $VRB))>;
+def : Pat<(v4i32 (sext_inreg v4i32:$VRB, v4i16)), (v4i32 (VEXTSH2W $VRB))>;
+def : Pat<(v2i64 (sext_inreg v2i64:$VRB, v2i8)), (v2i64 (VEXTSB2D $VRB))>;
+def : Pat<(v2i64 (sext_inreg v2i64:$VRB, v2i16)), (v2i64 (VEXTSH2D $VRB))>;
+def : Pat<(v2i64 (sext_inreg v2i64:$VRB, v2i32)), (v2i64 (VEXTSW2D $VRB))>;
+
 // Vector Integer Negate
 def VNEGW : VX_VT5_EO5_VB5<1538, 6, "vnegw",
                            [(set v4i32:$vD,

diff --git a/llvm/test/CodeGen/PowerPC/vector-extend-sign.ll b/llvm/test/CodeGen/PowerPC/vector-extend-sign.ll
new file mode 100644 (file)
index 0000000..7192829
--- /dev/null
+++ b/llvm/test/CodeGen/PowerPC/vector-extend-sign.ll
@@ -0,0 +1,178 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -verify-machineinstrs -mtriple=powerpc64le-unknown-linux-gnu \
+; RUN:     -mcpu=pwr9 < %s | FileCheck %s --check-prefix=CHECK-P9
+; RUN: llc -verify-machineinstrs -mtriple=powerpc64-unknown-linux-gnu \
+; RUN:     -mcpu=pwr9 < %s | FileCheck %s --check-prefix=CHECK-P9
+; RUN: llc -verify-machineinstrs -mtriple=powerpc64le-unknown-linux-gnu \
+; RUN:     -mcpu=pwr9 -mattr=-altivec < %s | FileCheck %s \
+; RUN:     --check-prefix=CHECK-P9-NOALTIVEC
+; RUN: llc -verify-machineinstrs -mtriple=powerpc64le-unknown-linux-gnu \
+; RUN:     -mcpu=pwr8 < %s | FileCheck %s --check-prefix=CHECK-P8
+
+define <4 x i32> @test_vextsh2w(<4 x i32> %m) {
+; CHECK-P9-LABEL: test_vextsh2w:
+; CHECK-P9:       # %bb.0: # %entry
+; CHECK-P9-NEXT:    vextsh2w 2, 2
+; CHECK-P9-NEXT:    blr
+;
+; CHECK-P9-NOALTIVEC-LABEL: test_vextsh2w:
+; CHECK-P9-NOALTIVEC:       # %bb.0: # %entry
+; CHECK-P9-NOALTIVEC-NEXT:    extsh 6, 6
+; CHECK-P9-NOALTIVEC-NEXT:    extsh 5, 5
+; CHECK-P9-NOALTIVEC-NEXT:    extsh 4, 4
+; CHECK-P9-NOALTIVEC-NEXT:    extsh 3, 3
+; CHECK-P9-NOALTIVEC-NEXT:    blr
+;
+; CHECK-P8-LABEL: test_vextsh2w:
+; CHECK-P8:       # %bb.0: # %entry
+; CHECK-P8-NEXT:    vspltisw 3, 8
+; CHECK-P8-NEXT:    vadduwm 3, 3, 3
+; CHECK-P8-NEXT:    vslw 2, 2, 3
+; CHECK-P8-NEXT:    vsraw 2, 2, 3
+; CHECK-P8-NEXT:    blr
+entry:
+  %shl = shl <4 x i32> %m, <i32 16, i32 16, i32 16, i32 16>
+  %shr = ashr exact <4 x i32> %shl, <i32 16, i32 16, i32 16, i32 16>
+  ret <4 x i32> %shr
+}
+
+define <4 x i32> @test_vextsb2w(<4 x i32> %m) {
+; CHECK-P9-LABEL: test_vextsb2w:
+; CHECK-P9:       # %bb.0: # %entry
+; CHECK-P9-NEXT:    vextsb2w 2, 2
+; CHECK-P9-NEXT:    blr
+;
+; CHECK-P9-NOALTIVEC-LABEL: test_vextsb2w:
+; CHECK-P9-NOALTIVEC:       # %bb.0: # %entry
+; CHECK-P9-NOALTIVEC-NEXT:    extsb 6, 6
+; CHECK-P9-NOALTIVEC-NEXT:    extsb 5, 5
+; CHECK-P9-NOALTIVEC-NEXT:    extsb 4, 4
+; CHECK-P9-NOALTIVEC-NEXT:    extsb 3, 3
+; CHECK-P9-NOALTIVEC-NEXT:    blr
+;
+; CHECK-P8-LABEL: test_vextsb2w:
+; CHECK-P8:       # %bb.0: # %entry
+; CHECK-P8-NEXT:    vspltisw 3, 12
+; CHECK-P8-NEXT:    vadduwm 3, 3, 3
+; CHECK-P8-NEXT:    vslw 2, 2, 3
+; CHECK-P8-NEXT:    vsraw 2, 2, 3
+; CHECK-P8-NEXT:    blr
+entry:
+  %shl = shl <4 x i32> %m, <i32 24, i32 24, i32 24, i32 24>
+  %shr = ashr exact <4 x i32> %shl, <i32 24, i32 24, i32 24, i32 24>
+  ret <4 x i32> %shr
+}
+
+define <2 x i64> @test_vextsb2d(<2 x i64> %m) {
+; CHECK-P9-LABEL: test_vextsb2d:
+; CHECK-P9:       # %bb.0: # %entry
+; CHECK-P9-NEXT:    vextsb2d 2, 2
+; CHECK-P9-NEXT:    blr
+;
+; CHECK-P9-NOALTIVEC-LABEL: test_vextsb2d:
+; CHECK-P9-NOALTIVEC:       # %bb.0: # %entry
+; CHECK-P9-NOALTIVEC-NEXT:    extsb 3, 3
+; CHECK-P9-NOALTIVEC-NEXT:    extsb 4, 4
+; CHECK-P9-NOALTIVEC-NEXT:    blr
+;
+; CHECK-P8-LABEL: test_vextsb2d:
+; CHECK-P8:       # %bb.0: # %entry
+; CHECK-P8-NEXT:    addis 3, 2, .LCPI2_0@toc@ha
+; CHECK-P8-NEXT:    addi 3, 3, .LCPI2_0@toc@l
+; CHECK-P8-NEXT:    lxvd2x 0, 0, 3
+; CHECK-P8-NEXT:    xxswapd 35, 0
+; CHECK-P8-NEXT:    vsld 2, 2, 3
+; CHECK-P8-NEXT:    vsrad 2, 2, 3
+; CHECK-P8-NEXT:    blr
+entry:
+  %shl = shl <2 x i64> %m, <i64 56, i64 56>
+  %shr = ashr exact <2 x i64> %shl, <i64 56, i64 56>
+  ret <2 x i64> %shr
+}
+
+define <2 x i64> @test_vextsh2d(<2 x i64> %m) {
+; CHECK-P9-LABEL: test_vextsh2d:
+; CHECK-P9:       # %bb.0: # %entry
+; CHECK-P9-NEXT:    vextsh2d 2, 2
+; CHECK-P9-NEXT:    blr
+;
+; CHECK-P9-NOALTIVEC-LABEL: test_vextsh2d:
+; CHECK-P9-NOALTIVEC:       # %bb.0: # %entry
+; CHECK-P9-NOALTIVEC-NEXT:    extsh 3, 3
+; CHECK-P9-NOALTIVEC-NEXT:    extsh 4, 4
+; CHECK-P9-NOALTIVEC-NEXT:    blr
+;
+; CHECK-P8-LABEL: test_vextsh2d:
+; CHECK-P8:       # %bb.0: # %entry
+; CHECK-P8-NEXT:    addis 3, 2, .LCPI3_0@toc@ha
+; CHECK-P8-NEXT:    addi 3, 3, .LCPI3_0@toc@l
+; CHECK-P8-NEXT:    lxvd2x 0, 0, 3
+; CHECK-P8-NEXT:    xxswapd 35, 0
+; CHECK-P8-NEXT:    vsld 2, 2, 3
+; CHECK-P8-NEXT:    vsrad 2, 2, 3
+; CHECK-P8-NEXT:    blr
+entry:
+  %shl = shl <2 x i64> %m, <i64 48, i64 48>
+  %shr = ashr exact <2 x i64> %shl, <i64 48, i64 48>
+  ret <2 x i64> %shr
+}
+
+define <2 x i64> @test_vextsw2d(<2 x i64> %m) {
+; CHECK-P9-LABEL: test_vextsw2d:
+; CHECK-P9:       # %bb.0: # %entry
+; CHECK-P9-NEXT:    vextsw2d 2, 2
+; CHECK-P9-NEXT:    blr
+;
+; CHECK-P9-NOALTIVEC-LABEL: test_vextsw2d:
+; CHECK-P9-NOALTIVEC:       # %bb.0: # %entry
+; CHECK-P9-NOALTIVEC-NEXT:    extsw 3, 3
+; CHECK-P9-NOALTIVEC-NEXT:    extsw 4, 4
+; CHECK-P9-NOALTIVEC-NEXT:    blr
+;
+; CHECK-P8-LABEL: test_vextsw2d:
+; CHECK-P8:       # %bb.0: # %entry
+; CHECK-P8-NEXT:    addis 3, 2, .LCPI4_0@toc@ha
+; CHECK-P8-NEXT:    addi 3, 3, .LCPI4_0@toc@l
+; CHECK-P8-NEXT:    lxvd2x 0, 0, 3
+; CHECK-P8-NEXT:    xxswapd 35, 0
+; CHECK-P8-NEXT:    vsld 2, 2, 3
+; CHECK-P8-NEXT:    vsrad 2, 2, 3
+; CHECK-P8-NEXT:    blr
+entry:
+  %shl = shl <2 x i64> %m, <i64 32, i64 32>
+  %shr = ashr exact <2 x i64> %shl, <i64 32, i64 32>
+  ret <2 x i64> %shr
+}
+
+define <2 x i64> @test_none(<2 x i64> %m) {
+; CHECK-P9-LABEL: test_none:
+; CHECK-P9:       # %bb.0: # %entry
+; CHECK-P9-NEXT:    addis 3, 2, .LCPI5_0@toc@ha
+; CHECK-P9-NEXT:    addi 3, 3, .LCPI5_0@toc@l
+; CHECK-P9-NEXT:    lxvx 35, 0, 3
+; CHECK-P9-NEXT:    vsld 2, 2, 3
+; CHECK-P9-NEXT:    vsrad 2, 2, 3
+; CHECK-P9-NEXT:    blr
+;
+; CHECK-P9-NOALTIVEC-LABEL: test_none:
+; CHECK-P9-NOALTIVEC:       # %bb.0: # %entry
+; CHECK-P9-NOALTIVEC-NEXT:    sldi 3, 3, 16
+; CHECK-P9-NOALTIVEC-NEXT:    sldi 4, 4, 16
+; CHECK-P9-NOALTIVEC-NEXT:    sradi 3, 3, 16
+; CHECK-P9-NOALTIVEC-NEXT:    sradi 4, 4, 16
+; CHECK-P9-NOALTIVEC-NEXT:    blr
+;
+; CHECK-P8-LABEL: test_none:
+; CHECK-P8:       # %bb.0: # %entry
+; CHECK-P8-NEXT:    addis 3, 2, .LCPI5_0@toc@ha
+; CHECK-P8-NEXT:    addi 3, 3, .LCPI5_0@toc@l
+; CHECK-P8-NEXT:    lxvd2x 0, 0, 3
+; CHECK-P8-NEXT:    xxswapd 35, 0
+; CHECK-P8-NEXT:    vsld 2, 2, 3
+; CHECK-P8-NEXT:    vsrad 2, 2, 3
+; CHECK-P8-NEXT:    blr
+entry:
+  %shl = shl <2 x i64> %m, <i64 16, i64 16>
+  %shr = ashr exact <2 x i64> %shl, <i64 16, i64 16>
+  ret <2 x i64> %shr
+}