IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
}
-let Predicates = [HasAVX] in {
-def : Pat<(v8i32 (X86vzmovl
- (insert_subvector undef, (v4i32 VR128:$src), (iPTR 0)))),
- (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
-def : Pat<(v4i64 (X86vzmovl
- (insert_subvector undef, (v2i64 VR128:$src), (iPTR 0)))),
- (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
-def : Pat<(v8f32 (X86vzmovl
- (insert_subvector undef, (v4f32 VR128:$src), (iPTR 0)))),
- (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
-def : Pat<(v4f64 (X86vzmovl
- (insert_subvector undef, (v2f64 VR128:$src), (iPTR 0)))),
- (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
-}
-
-
def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
(VMOVUPSYmr addr:$dst, VR256:$src)>;
def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
multiclass sse12_mov_hilo_packed<bits<8>opc, SDNode psnode, SDNode pdnode,
string base_opc, InstrItinClass itin> {
- defm V#NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
+ let Predicates = [UseAVX] in
+ defm V#NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}",
itin>, VEX_4V;
-let Constraints = "$src1 = $dst" in
- defm NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
+ let Constraints = "$src1 = $dst" in
+ defm NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
"\t{$src2, $dst|$dst, $src2}",
itin>;
}
}
let SchedRW = [WriteStore] in {
+let Predicates = [UseAVX] in {
def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
"movlps\t{$src, $dst|$dst, $src}",
[(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
[(store (f64 (vector_extract (v2f64 VR128:$src),
(iPTR 0))), addr:$dst)],
IIC_SSE_MOV_LH>, VEX;
+}// UseAVX
def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
"movlps\t{$src, $dst|$dst, $src}",
[(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
IIC_SSE_MOV_LH>;
} // SchedRW
-let Predicates = [HasAVX] in {
+let Predicates = [UseAVX] in {
// Shuffle with VMOVLPS
def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
(VMOVLPSrm VR128:$src1, addr:$src2)>;
let SchedRW = [WriteStore] in {
// v2f64 extract element 1 is always custom lowered to unpack high to low
// and extract element 0 so the non-store version isn't too horrible.
+let Predicates = [UseAVX] in {
def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
"movhps\t{$src, $dst|$dst, $src}",
[(store (f64 (vector_extract
[(store (f64 (vector_extract
(v2f64 (X86Unpckh VR128:$src, VR128:$src)),
(iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
+} // UseAVX
def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
"movhps\t{$src, $dst|$dst, $src}",
[(store (f64 (vector_extract
(iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
} // SchedRW
-let Predicates = [HasAVX] in {
+let Predicates = [UseAVX] in {
// VMOVHPS patterns
def : Pat<(X86Movlhps VR128:$src1,
(bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
defm V#NAME#PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
VR256, v8f32, f256mem, loadv8f32, SSEPackedSingle, itins, 0>,
PS, VEX_4V, VEX_L;
-
+
defm V#NAME#PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
VR256, v4f64, f256mem, loadv4f64, SSEPackedDouble, itins, 0>,
PD, VEX_4V, VEX_L;
SSE_INTALU_ITINS_P, 1, NoVLX_Or_NoBWI>;
defm PMAXSW : PDI_binop_all<0xEE, "pmaxsw", smax, v8i16, v16i16,
SSE_INTALU_ITINS_P, 1, NoVLX_Or_NoBWI>;
+defm PAVGB : PDI_binop_all<0xE0, "pavgb", X86avg, v16i8, v32i8,
+ SSE_INTALU_ITINS_P, 1, NoVLX_Or_NoBWI>;
+defm PAVGW : PDI_binop_all<0xE3, "pavgw", X86avg, v8i16, v16i16,
+ SSE_INTALU_ITINS_P, 1, NoVLX_Or_NoBWI>;
// Intrinsic forms
defm PSUBSB : PDI_binop_all_int<0xE8, "psubsb", int_x86_sse2_psubs_b,
int_x86_avx2_paddus_w, SSE_INTALU_ITINS_P, 1>;
defm PMADDWD : PDI_binop_all_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
int_x86_avx2_pmadd_wd, SSE_PMADD, 1>;
-defm PAVGB : PDI_binop_all_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
- int_x86_avx2_pavg_b, SSE_INTALU_ITINS_P, 1>;
-defm PAVGW : PDI_binop_all_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
- int_x86_avx2_pavg_w, SSE_INTALU_ITINS_P, 1>;
-defm PSADBW : PDI_binop_all_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
- int_x86_avx2_psad_bw, SSE_PMADD, 1>;
-
-let Predicates = [HasAVX2] in
- def : Pat<(v32i8 (X86psadbw (v32i8 VR256:$src1),
- (v32i8 VR256:$src2))),
- (VPSADBWYrr VR256:$src2, VR256:$src1)>;
let Predicates = [HasAVX] in
- def : Pat<(v16i8 (X86psadbw (v16i8 VR128:$src1),
- (v16i8 VR128:$src2))),
- (VPSADBWrr VR128:$src2, VR128:$src1)>;
-
-def : Pat<(v16i8 (X86psadbw (v16i8 VR128:$src1),
- (v16i8 VR128:$src2))),
- (PSADBWrr VR128:$src2, VR128:$src1)>;
+defm VPSADBW : PDI_binop_rm2<0xF6, "vpsadbw", X86psadbw, v2i64, v16i8, VR128,
+ loadv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
+ VEX_4V;
+let Predicates = [HasAVX2] in
+defm VPSADBWY : PDI_binop_rm2<0xF6, "vpsadbw", X86psadbw, v4i64, v32i8, VR256,
+ loadv4i64, i256mem, SSE_INTMUL_ITINS_P, 1, 0>,
+ VEX_4V, VEX_L;
+let Constraints = "$src1 = $dst" in
+defm PSADBW : PDI_binop_rm2<0xF6, "psadbw", X86psadbw, v2i64, v16i8, VR128,
+ memopv2i64, i128mem, SSE_INTALU_ITINS_P, 1>;
let Predicates = [HasAVX] in
defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
}// Predicates = [HasAVX2]
-let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift], hasSideEffects = 0 ,
+let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift], hasSideEffects = 0 ,
Predicates = [HasAVX2, NoVLX_Or_NoBWI] in {
// 256-bit logical shifts.
def VPSLLDQYri : PDIi8<0x73, MRM7r,
} // ExeDomain = SSEPackedInt
//===---------------------------------------------------------------------===//
-// SSE2 - Move Doubleword
+// SSE2 - Move Doubleword/Quadword
//===---------------------------------------------------------------------===//
//===---------------------------------------------------------------------===//
IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
}
-//===---------------------------------------------------------------------===//
-// Patterns and instructions to describe movd/movq to XMM register zero-extends
-//
-let isCodeGenOnly = 1, SchedRW = [WriteMove] in {
-let AddedComplexity = 15 in {
-def VMOVZQI2PQIrr : VS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
- "movq\t{$src, $dst|$dst, $src}", // X86-64 only
- [(set VR128:$dst, (v2i64 (X86vzmovl
- (v2i64 (scalar_to_vector GR64:$src)))))],
- IIC_SSE_MOVDQ>,
- VEX, VEX_W;
-def MOVZQI2PQIrr : RS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
- "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
- [(set VR128:$dst, (v2i64 (X86vzmovl
- (v2i64 (scalar_to_vector GR64:$src)))))],
- IIC_SSE_MOVDQ>;
-}
-} // isCodeGenOnly, SchedRW
-
let Predicates = [UseAVX] in {
- let AddedComplexity = 15 in
+ let AddedComplexity = 15 in {
def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector GR32:$src)))),
(VMOVDI2PDIrr GR32:$src)>;
+ def : Pat<(v2i64 (X86vzmovl (v2i64 (scalar_to_vector GR64:$src)))),
+ (VMOV64toPQIrr GR64:$src)>;
+
+ def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
+ (v2i64 (scalar_to_vector GR64:$src)),(iPTR 0)))),
+ (SUBREG_TO_REG (i64 0), (VMOV64toPQIrr GR64:$src), sub_xmm)>;
+ }
// AVX 128-bit movd/movq instructions write zeros in the high 128-bit part.
// These instructions also write zeros in the high part of a 256-bit register.
let AddedComplexity = 20 in {
def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
(v4i32 (scalar_to_vector GR32:$src)),(iPTR 0)))),
(SUBREG_TO_REG (i32 0), (VMOVDI2PDIrr GR32:$src), sub_xmm)>;
- def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
- (v2i64 (scalar_to_vector GR64:$src)),(iPTR 0)))),
- (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
}
let Predicates = [UseSSE2] in {
- let AddedComplexity = 15 in
+ let AddedComplexity = 15 in {
def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector GR32:$src)))),
(MOVDI2PDIrr GR32:$src)>;
+ def : Pat<(v2i64 (X86vzmovl (v2i64 (scalar_to_vector GR64:$src)))),
+ (MOV64toPQIrr GR64:$src)>;
+ }
let AddedComplexity = 20 in {
def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector (loadi32 addr:$src))))),
(MOVDI2PDIrm addr:$src)>;
IIC_SSE_MOV_LH>, Sched<[WriteLoad]>;
}
-let Predicates = [HasAVX] in {
+let Predicates = [HasAVX, NoVLX] in {
defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
v4f32, VR128, loadv4f32, f128mem>, VEX;
defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
memopv4f32, f128mem>;
-let Predicates = [HasAVX] in {
+let Predicates = [HasAVX, NoVLX] in {
def : Pat<(v4i32 (X86Movshdup VR128:$src)),
(VMOVSHDUPrr VR128:$src)>;
def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (loadv2i64 addr:$src)))),
def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set VR256:$dst,
- (v4f64 (X86Movddup
- (scalar_to_vector (loadf64 addr:$src)))))]>,
+ (v4f64 (X86Movddup (loadv4f64 addr:$src))))]>,
Sched<[WriteLoad]>;
}
-let Predicates = [HasAVX] in {
+let Predicates = [HasAVX, NoVLX] in {
defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX, VEX_L;
}
defm MOVDDUP : sse3_replicate_dfp<"movddup">;
-let Predicates = [HasAVX] in {
+
+let Predicates = [HasAVX, NoVLX] in {
def : Pat<(X86Movddup (loadv2f64 addr:$src)),
(VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
+
+ // 256-bit version
+ def : Pat<(X86Movddup (loadv4i64 addr:$src)),
+ (VMOVDDUPYrm addr:$src)>;
+ def : Pat<(X86Movddup (v4i64 VR256:$src)),
+ (VMOVDDUPYrr VR256:$src)>;
+}
+
+let Predicates = [HasAVX] in {
def : Pat<(X86Movddup (bc_v2f64 (loadv4f32 addr:$src))),
(VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
def : Pat<(X86Movddup (bc_v2f64 (loadv2i64 addr:$src))),
def : Pat<(X86Movddup (bc_v2f64
(v2i64 (scalar_to_vector (loadi64 addr:$src))))),
(VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
-
- // 256-bit version
- def : Pat<(X86Movddup (loadv4f64 addr:$src)),
- (VMOVDDUPYrm addr:$src)>;
- def : Pat<(X86Movddup (loadv4i64 addr:$src)),
- (VMOVDDUPYrm addr:$src)>;
- def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
- (VMOVDDUPYrm addr:$src)>;
- def : Pat<(X86Movddup (v4i64 VR256:$src)),
- (VMOVDDUPYrr VR256:$src)>;
}
let Predicates = [UseAVX, OptForSize] in {
(VPBROADCASTWrm addr:$src)>;
def : Pat<(v16i16 (X86VBroadcast (i16 (trunc (i32 (load addr:$src)))))),
(VPBROADCASTWYrm addr:$src)>;
+ def : Pat<(v8i16 (X86VBroadcast
+ (i16 (trunc (i32 (zextloadi16 addr:$src)))))),
+ (VPBROADCASTWrm addr:$src)>;
+ def : Pat<(v16i16 (X86VBroadcast
+ (i16 (trunc (i32 (zextloadi16 addr:$src)))))),
+ (VPBROADCASTWYrm addr:$src)>;
// Provide aliases for broadcast from the same register class that
// automatically does the extract.