diff options
| author | tangaac <tangyan01@loongson.cn> | 2025-08-06 09:11:34 +0800 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2025-08-06 09:11:34 +0800 |
| commit | b05e26be8a487d21cf15d34ef60b39b3ca94f235 (patch) | |
| tree | 472571d1c3c54ef6723b7040d8f6b98794be8aa6 /llvm/lib | |
| parent | fe0948c9a5d4ad0255c94306b16ac977c2e84ee0 (diff) | |
| download | llvm-b05e26be8a487d21cf15d34ef60b39b3ca94f235.zip llvm-b05e26be8a487d21cf15d34ef60b39b3ca94f235.tar.gz llvm-b05e26be8a487d21cf15d34ef60b39b3ca94f235.tar.bz2 | |
[LoongArch] Optimize extracting f32/f64 from 256-bit vector by using XVPICKVE. (#151914)
Diffstat (limited to 'llvm/lib')
| -rw-r--r-- | llvm/lib/Target/LoongArch/LoongArchLASXInstrInfo.td | 27 |
1 files changed, 13 insertions, 14 deletions
diff --git a/llvm/lib/Target/LoongArch/LoongArchLASXInstrInfo.td b/llvm/lib/Target/LoongArch/LoongArchLASXInstrInfo.td index 5096a8f..d8bb16f 100644 --- a/llvm/lib/Target/LoongArch/LoongArchLASXInstrInfo.td +++ b/llvm/lib/Target/LoongArch/LoongArchLASXInstrInfo.td @@ -1651,20 +1651,19 @@ def : Pat<(vector_insert v8i32:$xd, GRLenVT:$rj, uimm3:$imm), (XVINSGR2VR_W v8i32:$xd, GRLenVT:$rj, uimm3:$imm)>; def : Pat<(vector_insert v4i64:$xd, GRLenVT:$rj, uimm2:$imm), (XVINSGR2VR_D v4i64:$xd, GRLenVT:$rj, uimm2:$imm)>; -def : Pat<(vector_insert v8f32:$xd, (loongarch_movgr2fr_w_la64 GPR:$rj), uimm3:$imm), - (XVINSGR2VR_W $xd, $rj, uimm3:$imm)>; -def : Pat<(vector_insert v4f64:$xd, (f64 (bitconvert i64:$rj)), uimm2:$imm), - (XVINSGR2VR_D $xd, $rj, uimm2:$imm)>; -def : Pat<(vector_insert v8f32:$xd, (f32 (vector_extract v8f32:$xj, uimm3:$imm1)), uimm3:$imm2), - (XVINSGR2VR_W $xd, (XVPICKVE2GR_W v8f32:$xj, uimm3:$imm1), uimm3:$imm2)>; -def : Pat<(vector_insert v4f64:$xd, (f64 (vector_extract v4f64:$xj, uimm2:$imm1)), uimm2:$imm2), - (XVINSGR2VR_D $xd, (XVPICKVE2GR_D v4f64:$xj, uimm2:$imm1), uimm2:$imm2)>; +def : Pat<(vector_insert v8f32:$xd, (loongarch_movgr2fr_w_la64 GPR:$rj), + uimm3:$imm), + (XVINSGR2VR_W v8f32:$xd, GPR:$rj, uimm3:$imm)>; +def : Pat<(vector_insert v4f64:$xd, (f64(bitconvert i64:$rj)), uimm2:$imm), + (XVINSGR2VR_D v4f64:$xd, GPR:$rj, uimm2:$imm)>; // XVINSVE0_{W/D} def : Pat<(vector_insert v8f32:$xd, FPR32:$fj, uimm3:$imm), - (XVINSVE0_W $xd, (SUBREG_TO_REG (i64 0), FPR32:$fj, sub_32), uimm3:$imm)>; + (XVINSVE0_W v8f32:$xd, (SUBREG_TO_REG(i64 0), FPR32:$fj, sub_32), + uimm3:$imm)>; def : Pat<(vector_insert v4f64:$xd, FPR64:$fj, uimm2:$imm), - (XVINSVE0_D $xd, (SUBREG_TO_REG (i64 0), FPR64:$fj, sub_64), uimm2:$imm)>; + (XVINSVE0_D v4f64:$xd, (SUBREG_TO_REG(i64 0), FPR64:$fj, sub_64), + uimm2:$imm)>; // scalar_to_vector def : Pat<(v8f32 (scalar_to_vector FPR32:$fj)), @@ -1884,10 +1883,10 @@ def : Pat<(i64 (vector_extract v8i32:$xj, uimm3:$imm)), (XVPICKVE2GR_W v8i32:$xj, uimm3:$imm)>; def : Pat<(i64 (vector_extract v4i64:$xj, uimm2:$imm)), (XVPICKVE2GR_D v4i64:$xj, uimm2:$imm)>; -def : Pat<(f32 (vector_extract v8f32:$xj, uimm3:$imm)), - (MOVGR2FR_W (XVPICKVE2GR_W v8f32:$xj, uimm3:$imm))>; -def : Pat<(f64 (vector_extract v4f64:$xj, uimm2:$imm)), - (MOVGR2FR_D (XVPICKVE2GR_D v4f64:$xj, uimm2:$imm))>; +def : Pat<(f32(vector_extract v8f32:$xj, uimm3:$imm)), + (EXTRACT_SUBREG(XVPICKVE_W v8f32:$xj, uimm3:$imm), sub_32)>; +def : Pat<(f64(vector_extract v4f64:$xj, uimm2:$imm)), + (EXTRACT_SUBREG(XVPICKVE_D v4f64:$xj, uimm2:$imm), sub_64)>; // vselect def : Pat<(v32i8 (vselect LASX256:$xd, (v32i8 (SplatPat_uimm8 uimm8:$imm)), |