Use PMULDQ for v2i64 multiplies when SSE4.1 is available. And add
load-folding table entries for PMULDQ and PMULLD.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51489 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index caa81c4..2a9cef1 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -700,6 +700,7 @@
if (Subtarget->hasSSE41()) {
// FIXME: Do we need to handle scalar-to-vector here?
setOperationAction(ISD::MUL, MVT::v4i32, Legal);
+ setOperationAction(ISD::MUL, MVT::v2i64, Legal);
// i8 and i16 vectors are custom , because the source register and source
// source memory operand types are not the same width. f32 vectors are
diff --git a/lib/Target/X86/X86InstrInfo.cpp b/lib/Target/X86/X86InstrInfo.cpp
index 4d4867f..1a5fd47 100644
--- a/lib/Target/X86/X86InstrInfo.cpp
+++ b/lib/Target/X86/X86InstrInfo.cpp
@@ -569,8 +569,12 @@
{ X86::PMAXUBrr, X86::PMAXUBrm },
{ X86::PMINSWrr, X86::PMINSWrm },
{ X86::PMINUBrr, X86::PMINUBrm },
+ { X86::PMULDQrr, X86::PMULDQrm },
+ { X86::PMULDQrr_int, X86::PMULDQrm_int },
{ X86::PMULHUWrr, X86::PMULHUWrm },
{ X86::PMULHWrr, X86::PMULHWrm },
+ { X86::PMULLDrr, X86::PMULLDrm },
+ { X86::PMULLDrr_int, X86::PMULLDrm_int },
{ X86::PMULLWrr, X86::PMULLWrm },
{ X86::PMULUDQrr, X86::PMULUDQrm },
{ X86::PORrr, X86::PORrm },
diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 5931635..f41f8cf 100644
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -3242,19 +3242,18 @@
int_x86_sse41_pmaxud, 1>;
defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw",
int_x86_sse41_pmaxuw, 1>;
-defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq",
- int_x86_sse41_pmuldq, 1>;
/// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
let Constraints = "$src1 = $dst" in {
- multiclass SS41I_binop_patint<bits<8> opc, string OpcodeStr, SDNode OpNode,
- Intrinsic IntId128, bit Commutable = 0> {
+ multiclass SS41I_binop_patint<bits<8> opc, string OpcodeStr, ValueType OpVT,
+ SDNode OpNode, Intrinsic IntId128,
+ bit Commutable = 0> {
def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
- [(set VR128:$dst, (OpNode (v4i32 VR128:$src1),
- VR128:$src2))]>, OpSize {
+ [(set VR128:$dst, (OpNode (OpVT VR128:$src1),
+ VR128:$src2))]>, OpSize {
let isCommutable = Commutable;
}
def rr_int : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
@@ -3277,8 +3276,10 @@
OpSize;
}
}
-defm PMULLD : SS41I_binop_patint<0x40, "pmulld", mul,
+defm PMULLD : SS41I_binop_patint<0x40, "pmulld", v4i32, mul,
int_x86_sse41_pmulld, 1>;
+defm PMULDQ : SS41I_binop_patint<0x28, "pmuldq", v2i64, mul,
+ int_x86_sse41_pmuldq, 1>;
/// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate