(a) Slight rethink on i64 zero/sign/any extend code - use a shuffle to
    directly zero-extend i32 to i64, but use rotates and shifts for
    sign extension. Also ensure unified register consistency.
(b) Add new test harness for i64 operations: i64ops.ll


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59970 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Scott Michel 2008-11-24 18:20:46 +00:00
parent 942b8c8533
commit 045a14503f
2 changed files with 42 additions and 4 deletions

View File

@ -2363,16 +2363,27 @@ static SDValue LowerI64Math(SDValue Op, SelectionDAG &DAG, unsigned Opc)
SDValue PromoteScalar =
DAG.getNode(SPUISD::PROMOTE_SCALAR, Op0VecVT, Op0);
SDValue RotQuad =
DAG.getNode(SPUISD::ROTQUAD_RZ_BYTES, Op0VecVT,
PromoteScalar, DAG.getConstant(4, MVT::i32));
if (Opc != ISD::SIGN_EXTEND) {
// Use a shuffle to zero extend the i32 to i64 directly:
SDValue shufMask =
DAG.getNode(ISD::BUILD_VECTOR, Op0VecVT,
DAG.getConstant(0x80808080, MVT::i32),
DAG.getConstant(0x00010203, MVT::i32),
DAG.getConstant(0x80808080, MVT::i32),
DAG.getConstant(0x08090a0b, MVT::i32));
SDValue zextShuffle =
DAG.getNode(SPUISD::SHUFB, Op0VecVT,
PromoteScalar, PromoteScalar, shufMask);
return DAG.getNode(SPUISD::VEC2PREFSLOT, VT,
DAG.getNode(ISD::BIT_CONVERT, VecVT, RotQuad));
DAG.getNode(ISD::BIT_CONVERT, VecVT, zextShuffle));
} else {
// SPU has no "rotate quadword and replicate bit 0" (i.e. rotate/shift
// right and propagate the sign bit) instruction.
SDValue RotQuad =
DAG.getNode(SPUISD::ROTQUAD_RZ_BYTES, Op0VecVT,
PromoteScalar, DAG.getConstant(4, MVT::i32));
SDValue SignQuad =
DAG.getNode(SPUISD::VEC_SRA, Op0VecVT,
PromoteScalar, DAG.getConstant(32, MVT::i32));

View File

@ -0,0 +1,27 @@
; RUN: llvm-as -o - %s | llc -march=cellspu > %t1.s
; RUN: grep {fsmbi.*61680} %t1.s | count 1
; RUN: grep rotqmbyi %t1.s | count 1
; RUN: grep rotmai %t1.s | count 1
; RUN: grep selb %t1.s | count 1
; RUN: grep shufb %t1.s | count 2
; RUN: grep cg %t1.s | count 1
; RUN: grep addx %t1.s | count 1
; ModuleID = 'stores.bc'
target datalayout = "E-p:32:32:128-f64:64:128-f32:32:128-i64:32:128-i32:32:128-i16:16:128-i8:8:128-i1:8:128-a0:0:128-v128:128:128-s0:128:128"
target triple = "spu"
define i64 @sext_i64_i32(i32 %a) nounwind {
%1 = sext i32 %a to i64
ret i64 %1
}
define i64 @zext_i64_i32(i32 %a) nounwind {
%1 = zext i32 %a to i64
ret i64 %1
}
define i64 @add_i64(i64 %a, i64 %b) nounwind {
%1 = add i64 %a, %b
ret i64 %1
}