(VMOVHPSrm VR128:$src1, addr:$src2)>;
// FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
- // is during lowering, where it's not possible to recognize the load fold
+ // is during lowering, where it's not possible to recognize the load fold
// cause it has two uses through a bitcast. One use disappears at isel time
// and the fold opportunity reappears.
def : Pat<(v2f64 (X86Unpckl VR128:$src1,
let Predicates = [UseSSE2] in {
// FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
- // is during lowering, where it's not possible to recognize the load fold
+ // is during lowering, where it's not possible to recognize the load fold
// cause it has two uses through a bitcast. One use disappears at isel time
// and the fold opportunity reappears.
def : Pat<(v2f64 (X86Unpckl VR128:$src1,
(VCVTDQ2PSYrm addr:$src)>;
// Match fround and fextend for 128/256-bit conversions
+ def : Pat<(v4f32 (X86vfpround (v2f64 VR128:$src))),
+ (VCVTPD2PSrr VR128:$src)>;
+ def : Pat<(v4f32 (X86vfpround (memopv2f64 addr:$src))),
+ (VCVTPD2PSXrm addr:$src)>;
def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
(VCVTPD2PSYrr VR256:$src)>;
def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
}
let Predicates = [UseSSE2] in {
- // Match fextend for 128 conversions
+ // Match fround and fextend for 128 conversions
+ def : Pat<(v4f32 (X86vfpround (v2f64 VR128:$src))),
+ (CVTPD2PSrr VR128:$src)>;
+ def : Pat<(v4f32 (X86vfpround (memopv2f64 addr:$src))),
+ (CVTPD2PSrm addr:$src)>;
+
def : Pat<(v2f64 (X86vfpext (v4f32 VR128:$src))),
(CVTPS2PDrr VR128:$src)>;
}
// sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
- Operand CC, SDNode OpNode, ValueType VT,
+ Operand CC, SDNode OpNode, ValueType VT,
PatFrag ld_frag, string asm, string asm_alt,
OpndItins itins> {
def rr : SIi8<0xC2, MRMSrcReg,
// sse12_cmp_packed - sse 1 & 2 compare packed instructions
multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
- Operand CC, Intrinsic Int, string asm,
+ Operand CC, Intrinsic Int, string asm,
string asm_alt, Domain d> {
def rri : PIi8<0xC2, MRMSrcReg,
(outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTS>,
sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd, SSE_SQRTS>;
+/// sse1_fp_unop_s_rw - SSE1 unops where vector form has a read-write operand.
+multiclass sse1_fp_unop_rw<bits<8> opc, string OpcodeStr, SDNode OpNode,
+ Intrinsic F32Int, OpndItins itins> {
+ def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
+ !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
+ [(set FR32:$dst, (OpNode FR32:$src))]>;
+ // For scalar unary operations, fold a load into the operation
+ // only in OptForSize mode. It eliminates an instruction, but it also
+ // eliminates a whole-register clobber (the load), so it introduces a
+ // partial register update condition.
+ def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
+ !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
+ [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
+ Requires<[UseSSE1, OptForSize]>;
+ let Constraints = "$src1 = $dst" in {
+ def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
+ (ins VR128:$src1, VR128:$src2),
+ !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
+ [], itins.rr>;
+ def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
+ (ins VR128:$src1, ssmem:$src2),
+ !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
+ [], itins.rm>;
+ }
+}
+
// Reciprocal approximations. Note that these typically require refinement
// in order to obtain suitable precision.
-defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss,
- SSE_SQRTS>,
+defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss,
+ SSE_SQRTS>,
sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTS>,
sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
SSE_SQRTS>;
-defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss,
- SSE_RCPS>,
+let Predicates = [UseSSE1] in {
+ def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
+ (RSQRTSSr_Int VR128:$src, VR128:$src)>;
+}
+
+defm RCP : sse1_fp_unop_rw<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss,
+ SSE_RCPS>,
sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPS>,
sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps, SSE_RCPS>;
+let Predicates = [UseSSE1] in {
+ def : Pat<(int_x86_sse_rcp_ss VR128:$src),
+ (RCPSSr_Int VR128:$src, VR128:$src)>;
+}
// There is no f64 version of the reciprocal approximation instructions.
def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
"movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
VEX, VEX_L;
-}
def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
VEX;
def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
"movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
VEX, VEX_L;
+}
// For Disassembler
let isCodeGenOnly = 1 in {
IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
}
-let canFoldAsLoad = 1, mayLoad = 1 in {
+let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
+ neverHasSideEffects = 1 in {
def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
"movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
VEX;
}
}
-let mayStore = 1 in {
+let mayStore = 1, neverHasSideEffects = 1 in {
def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
(ins i128mem:$dst, VR128:$src),
"movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
[], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
}
-let canFoldAsLoad = 1, mayLoad = 1 in {
+let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
+ neverHasSideEffects = 1 in {
def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
"movdqa\t{$src, $dst|$dst, $src}",
[/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/],
XS, Requires<[UseSSE2]>;
}
-// Intrinsic forms of MOVDQU load and store
-def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
- "vmovdqu\t{$src, $dst|$dst, $src}",
- [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)],
- IIC_SSE_MOVU_P_MR>,
- XS, VEX, Requires<[HasAVX]>;
-
-def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
- "movdqu\t{$src, $dst|$dst, $src}",
- [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)],
- IIC_SSE_MOVU_P_MR>,
- XS, Requires<[UseSSE2]>;
-
} // ExeDomain = SSEPackedInt
let Predicates = [HasAVX] in {
+ def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
+ (VMOVDQUmr addr:$dst, VR128:$src)>;
def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
(VMOVDQUYmr addr:$dst, VR256:$src)>;
}
+let Predicates = [UseSSE2] in
+def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
+ (MOVDQUmr addr:$dst, VR128:$src)>;
//===---------------------------------------------------------------------===//
// SSE2 - Packed Integer Arithmetic Instructions
i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
i128mem, SSE_INTALUQ_ITINS_P, 0, 0>, VEX_4V;
+defm VPSUBUSB : PDI_binop_rm<0xD8, "vpsubusb", X86subus, v16i8, VR128,
+ memopv2i64, i128mem, SSE_INTALU_ITINS_P, 0, 0>,
+ VEX_4V;
+defm VPSUBUSW : PDI_binop_rm<0xD9, "vpsubusw", X86subus, v8i16, VR128,
+ memopv2i64, i128mem, SSE_INTALU_ITINS_P, 0, 0>,
+ VEX_4V;
defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
VEX_4V;
+defm VPMINUB : PDI_binop_rm<0xDA, "vpminub", X86umin, v16i8, VR128, memopv2i64,
+ i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
+defm VPMINSW : PDI_binop_rm<0xEA, "vpminsw", X86smin, v8i16, VR128, memopv2i64,
+ i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
+defm VPMAXUB : PDI_binop_rm<0xDE, "vpmaxub", X86umax, v16i8, VR128, memopv2i64,
+ i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
+defm VPMAXSW : PDI_binop_rm<0xEE, "vpmaxsw", X86smax, v8i16, VR128, memopv2i64,
+ i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
// Intrinsic forms
defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
VR128, memopv2i64, i128mem,
SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
VR128, memopv2i64, i128mem,
SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
VR128, memopv2i64, i128mem,
SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
VR128, memopv2i64, i128mem,
SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L;
defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
i256mem, SSE_INTALUQ_ITINS_P, 0, 0>, VEX_4V, VEX_L;
+defm VPSUBUSBY : PDI_binop_rm<0xD8, "vpsubusb", X86subus, v32i8, VR256,
+ memopv4i64, i256mem, SSE_INTALU_ITINS_P, 0, 0>,
+ VEX_4V, VEX_L;
+defm VPSUBUSWY : PDI_binop_rm<0xD9, "vpsubusw", X86subus, v16i16, VR256,
+ memopv4i64, i256mem, SSE_INTALU_ITINS_P, 0, 0>,
+ VEX_4V, VEX_L;
defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
VR256, memopv4i64, i256mem,
SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L;
+defm VPMINUBY : PDI_binop_rm<0xDA, "vpminub", X86umin, v32i8,
+ VR256, memopv4i64, i256mem,
+ SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
+defm VPMINSWY : PDI_binop_rm<0xEA, "vpminsw", X86smin, v16i16,
+ VR256, memopv4i64, i256mem,
+ SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
+defm VPMAXUBY : PDI_binop_rm<0xDE, "vpmaxub", X86umax, v32i8,
+ VR256, memopv4i64, i256mem,
+ SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
+defm VPMAXSWY : PDI_binop_rm<0xEE, "vpmaxsw", X86smax, v16i16,
+ VR256, memopv4i64, i256mem,
+ SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
// Intrinsic forms
defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
VR256, memopv4i64, i256mem,
SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L;
-defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L;
-defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L;
defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
VR256, memopv4i64, i256mem,
SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
VR256, memopv4i64, i256mem,
SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
VR256, memopv4i64, i256mem,
SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
i128mem, SSE_INTALU_ITINS_P>;
defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
i128mem, SSE_INTALUQ_ITINS_P>;
+defm PSUBUSB : PDI_binop_rm<0xD8, "psubusb", X86subus, v16i8, VR128, memopv2i64,
+ i128mem, SSE_INTALU_ITINS_P>;
+defm PSUBUSW : PDI_binop_rm<0xD9, "psubusw", X86subus, v8i16, VR128, memopv2i64,
+ i128mem, SSE_INTALU_ITINS_P>;
defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1>;
+defm PMINUB : PDI_binop_rm<0xDA, "pminub", X86umin, v16i8, VR128, memopv2i64,
+ i128mem, SSE_INTALU_ITINS_P, 1>;
+defm PMINSW : PDI_binop_rm<0xEA, "pminsw", X86smin, v8i16, VR128, memopv2i64,
+ i128mem, SSE_INTALU_ITINS_P, 1>;
+defm PMAXUB : PDI_binop_rm<0xDE, "pmaxub", X86umax, v16i8, VR128, memopv2i64,
+ i128mem, SSE_INTALU_ITINS_P, 1>;
+defm PMAXSW : PDI_binop_rm<0xEE, "pmaxsw", X86smax, v8i16, VR128, memopv2i64,
+ i128mem, SSE_INTALU_ITINS_P, 1>;
// Intrinsic forms
defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
VR128, memopv2i64, i128mem,
SSE_INTALU_ITINS_P>;
-defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P>;
-defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P>;
defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
VR128, memopv2i64, i128mem,
SSE_INTALU_ITINS_P, 1>;
defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
VR128, memopv2i64, i128mem,
SSE_INTALU_ITINS_P, 1>;
-defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1>;
-defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1>;
-defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1>;
-defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1>;
defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
VR128, memopv2i64, i128mem,
SSE_INTALU_ITINS_P, 1>;
defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
+let Predicates = [HasAVX2] in {
+ def : Pat<(v8i32 (X86vsmovl (v8i16 (bitconvert (v2i64 (load addr:$src)))))),
+ (VPMOVSXWDYrm addr:$src)>;
+ def : Pat<(v4i64 (X86vsmovl (v4i32 (bitconvert (v2i64 (load addr:$src)))))),
+ (VPMOVSXDQYrm addr:$src)>;
+
+ def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2i64
+ (scalar_to_vector (loadi64 addr:$src))))))),
+ (VPMOVSXBDYrm addr:$src)>;
+ def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2f64
+ (scalar_to_vector (loadf64 addr:$src))))))),
+ (VPMOVSXBDYrm addr:$src)>;
+
+ def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2i64
+ (scalar_to_vector (loadi64 addr:$src))))))),
+ (VPMOVSXWQYrm addr:$src)>;
+ def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2f64
+ (scalar_to_vector (loadf64 addr:$src))))))),
+ (VPMOVSXWQYrm addr:$src)>;
+
+ def : Pat<(v4i64 (X86vsext (v16i8 (bitconvert (v4i32
+ (scalar_to_vector (loadi32 addr:$src))))))),
+ (VPMOVSXBQYrm addr:$src)>;
+}
+
let Predicates = [HasAVX] in {
// Common patterns involving scalar load
def : Pat<(int_x86_sse41_pmovsxbq
(bitconvert (v4i32 (X86vzmovl
(v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
(PMOVZXBQrm addr:$src)>;
+
+ def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
+ (scalar_to_vector (loadi64 addr:$src))))))),
+ (PMOVSXWDrm addr:$src)>;
+ def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
+ (scalar_to_vector (loadf64 addr:$src))))))),
+ (PMOVSXWDrm addr:$src)>;
+ def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
+ (scalar_to_vector (loadi32 addr:$src))))))),
+ (PMOVSXBDrm addr:$src)>;
+ def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
+ (scalar_to_vector (loadi32 addr:$src))))))),
+ (PMOVSXWQrm addr:$src)>;
+ def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
+ (scalar_to_vector (extloadi32i16 addr:$src))))))),
+ (PMOVSXBQrm addr:$src)>;
+ def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
+ (scalar_to_vector (loadi64 addr:$src))))))),
+ (PMOVSXDQrm addr:$src)>;
+ def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
+ (scalar_to_vector (loadf64 addr:$src))))))),
+ (PMOVSXDQrm addr:$src)>;
+ def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
+ (scalar_to_vector (loadi64 addr:$src))))))),
+ (PMOVSXBWrm addr:$src)>;
+ def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
+ (scalar_to_vector (loadf64 addr:$src))))))),
+ (PMOVSXBWrm addr:$src)>;
+}
+
+let Predicates = [HasAVX2] in {
+ def : Pat<(v16i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWYrr VR128:$src)>;
+ def : Pat<(v8i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDYrr VR128:$src)>;
+ def : Pat<(v4i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQYrr VR128:$src)>;
+
+ def : Pat<(v8i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDYrr VR128:$src)>;
+ def : Pat<(v4i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQYrr VR128:$src)>;
+
+ def : Pat<(v4i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQYrr VR128:$src)>;
+
+ def : Pat<(v16i16 (X86vzext (v32i8 VR256:$src))),
+ (VPMOVZXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+ def : Pat<(v8i32 (X86vzext (v32i8 VR256:$src))),
+ (VPMOVZXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+ def : Pat<(v4i64 (X86vzext (v32i8 VR256:$src))),
+ (VPMOVZXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+
+ def : Pat<(v8i32 (X86vzext (v16i16 VR256:$src))),
+ (VPMOVZXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+ def : Pat<(v4i64 (X86vzext (v16i16 VR256:$src))),
+ (VPMOVZXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+
+ def : Pat<(v4i64 (X86vzext (v8i32 VR256:$src))),
+ (VPMOVZXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+}
+
+let Predicates = [HasAVX] in {
+ def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWrr VR128:$src)>;
+ def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDrr VR128:$src)>;
+ def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQrr VR128:$src)>;
+
+ def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDrr VR128:$src)>;
+ def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQrr VR128:$src)>;
+
+ def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQrr VR128:$src)>;
+
+ def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
+ (VPMOVZXBWrm addr:$src)>;
+ def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
+ (VPMOVZXBWrm addr:$src)>;
+ def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
+ (VPMOVZXBDrm addr:$src)>;
+ def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
+ (VPMOVZXBQrm addr:$src)>;
+
+ def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
+ (VPMOVZXWDrm addr:$src)>;
+ def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
+ (VPMOVZXWDrm addr:$src)>;
+ def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
+ (VPMOVZXWQrm addr:$src)>;
+
+ def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
+ (VPMOVZXDQrm addr:$src)>;
+ def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
+ (VPMOVZXDQrm addr:$src)>;
+ def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
+ (VPMOVZXDQrm addr:$src)>;
+
+ def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
+ (scalar_to_vector (loadi64 addr:$src))))))),
+ (VPMOVSXWDrm addr:$src)>;
+ def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
+ (scalar_to_vector (loadi64 addr:$src))))))),
+ (VPMOVSXDQrm addr:$src)>;
+ def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
+ (scalar_to_vector (loadf64 addr:$src))))))),
+ (VPMOVSXWDrm addr:$src)>;
+ def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
+ (scalar_to_vector (loadf64 addr:$src))))))),
+ (VPMOVSXDQrm addr:$src)>;
+ def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
+ (scalar_to_vector (loadi64 addr:$src))))))),
+ (VPMOVSXBWrm addr:$src)>;
+ def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
+ (scalar_to_vector (loadf64 addr:$src))))))),
+ (VPMOVSXBWrm addr:$src)>;
+
+ def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
+ (scalar_to_vector (loadi32 addr:$src))))))),
+ (VPMOVSXBDrm addr:$src)>;
+ def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
+ (scalar_to_vector (loadi32 addr:$src))))))),
+ (VPMOVSXWQrm addr:$src)>;
+ def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
+ (scalar_to_vector (extloadi32i16 addr:$src))))))),
+ (VPMOVSXBQrm addr:$src)>;
+}
+
+let Predicates = [UseSSE41] in {
+ def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (PMOVZXBWrr VR128:$src)>;
+ def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (PMOVZXBDrr VR128:$src)>;
+ def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (PMOVZXBQrr VR128:$src)>;
+
+ def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (PMOVZXWDrr VR128:$src)>;
+ def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (PMOVZXWQrr VR128:$src)>;
+
+ def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (PMOVZXDQrr VR128:$src)>;
+
+ def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
+ (PMOVZXBWrm addr:$src)>;
+ def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
+ (PMOVZXBWrm addr:$src)>;
+ def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
+ (PMOVZXBDrm addr:$src)>;
+ def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
+ (PMOVZXBQrm addr:$src)>;
+
+ def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
+ (PMOVZXWDrm addr:$src)>;
+ def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
+ (PMOVZXWDrm addr:$src)>;
+ def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
+ (PMOVZXWQrm addr:$src)>;
+
+ def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
+ (PMOVZXDQrm addr:$src)>;
+ def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
+ (PMOVZXDQrm addr:$src)>;
+ def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
+ (PMOVZXDQrm addr:$src)>;
}
//===----------------------------------------------------------------------===//
def : Pat<(v4f32 (ffloor VR128:$src)),
(VROUNDPSr VR128:$src, (i32 0x1))>;
+ def : Pat<(v4f32 (fnearbyint VR128:$src)),
+ (VROUNDPSr VR128:$src, (i32 0xC))>;
+ def : Pat<(v4f32 (fceil VR128:$src)),
+ (VROUNDPSr VR128:$src, (i32 0x2))>;
+ def : Pat<(v4f32 (frint VR128:$src)),
+ (VROUNDPSr VR128:$src, (i32 0x4))>;
+ def : Pat<(v4f32 (ftrunc VR128:$src)),
+ (VROUNDPSr VR128:$src, (i32 0x3))>;
+
def : Pat<(v2f64 (ffloor VR128:$src)),
(VROUNDPDr VR128:$src, (i32 0x1))>;
+ def : Pat<(v2f64 (fnearbyint VR128:$src)),
+ (VROUNDPDr VR128:$src, (i32 0xC))>;
+ def : Pat<(v2f64 (fceil VR128:$src)),
+ (VROUNDPDr VR128:$src, (i32 0x2))>;
+ def : Pat<(v2f64 (frint VR128:$src)),
+ (VROUNDPDr VR128:$src, (i32 0x4))>;
+ def : Pat<(v2f64 (ftrunc VR128:$src)),
+ (VROUNDPDr VR128:$src, (i32 0x3))>;
+
def : Pat<(v8f32 (ffloor VR256:$src)),
(VROUNDYPSr VR256:$src, (i32 0x1))>;
+ def : Pat<(v8f32 (fnearbyint VR256:$src)),
+ (VROUNDYPSr VR256:$src, (i32 0xC))>;
+ def : Pat<(v8f32 (fceil VR256:$src)),
+ (VROUNDYPSr VR256:$src, (i32 0x2))>;
+ def : Pat<(v8f32 (frint VR256:$src)),
+ (VROUNDYPSr VR256:$src, (i32 0x4))>;
+ def : Pat<(v8f32 (ftrunc VR256:$src)),
+ (VROUNDYPSr VR256:$src, (i32 0x3))>;
+
def : Pat<(v4f64 (ffloor VR256:$src)),
(VROUNDYPDr VR256:$src, (i32 0x1))>;
+ def : Pat<(v4f64 (fnearbyint VR256:$src)),
+ (VROUNDYPDr VR256:$src, (i32 0xC))>;
+ def : Pat<(v4f64 (fceil VR256:$src)),
+ (VROUNDYPDr VR256:$src, (i32 0x2))>;
+ def : Pat<(v4f64 (frint VR256:$src)),
+ (VROUNDYPDr VR256:$src, (i32 0x4))>;
+ def : Pat<(v4f64 (ftrunc VR256:$src)),
+ (VROUNDYPDr VR256:$src, (i32 0x3))>;
}
defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
def : Pat<(v4f32 (ffloor VR128:$src)),
(ROUNDPSr VR128:$src, (i32 0x1))>;
+ def : Pat<(v4f32 (fnearbyint VR128:$src)),
+ (ROUNDPSr VR128:$src, (i32 0xC))>;
+ def : Pat<(v4f32 (fceil VR128:$src)),
+ (ROUNDPSr VR128:$src, (i32 0x2))>;
+ def : Pat<(v4f32 (frint VR128:$src)),
+ (ROUNDPSr VR128:$src, (i32 0x4))>;
+ def : Pat<(v4f32 (ftrunc VR128:$src)),
+ (ROUNDPSr VR128:$src, (i32 0x3))>;
+
def : Pat<(v2f64 (ffloor VR128:$src)),
(ROUNDPDr VR128:$src, (i32 0x1))>;
+ def : Pat<(v2f64 (fnearbyint VR128:$src)),
+ (ROUNDPDr VR128:$src, (i32 0xC))>;
+ def : Pat<(v2f64 (fceil VR128:$src)),
+ (ROUNDPDr VR128:$src, (i32 0x2))>;
+ def : Pat<(v2f64 (frint VR128:$src)),
+ (ROUNDPDr VR128:$src, (i32 0x4))>;
+ def : Pat<(v2f64 (ftrunc VR128:$src)),
+ (ROUNDPDr VR128:$src, (i32 0x3))>;
}
//===----------------------------------------------------------------------===//
(bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
}
-let Predicates = [HasAVX] in {
- let isCommutable = 0 in
- defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
- 0>, VEX_4V;
- defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
- 0>, VEX_4V;
- defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
- 0>, VEX_4V;
- defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
- 0>, VEX_4V;
- defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
- 0>, VEX_4V;
- defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
- 0>, VEX_4V;
- defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
- 0>, VEX_4V;
- defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
- 0>, VEX_4V;
- defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
- 0>, VEX_4V;
- defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
- 0>, VEX_4V;
-}
-
-let Predicates = [HasAVX2] in {
- let isCommutable = 0 in
- defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
- int_x86_avx2_packusdw>, VEX_4V, VEX_L;
- defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
- int_x86_avx2_pmins_b>, VEX_4V, VEX_L;
- defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
- int_x86_avx2_pmins_d>, VEX_4V, VEX_L;
- defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
- int_x86_avx2_pminu_d>, VEX_4V, VEX_L;
- defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
- int_x86_avx2_pminu_w>, VEX_4V, VEX_L;
- defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
- int_x86_avx2_pmaxs_b>, VEX_4V, VEX_L;
- defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
- int_x86_avx2_pmaxs_d>, VEX_4V, VEX_L;
- defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
- int_x86_avx2_pmaxu_d>, VEX_4V, VEX_L;
- defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
- int_x86_avx2_pmaxu_w>, VEX_4V, VEX_L;
- defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
- int_x86_avx2_pmul_dq>, VEX_4V, VEX_L;
-}
-
-let Constraints = "$src1 = $dst" in {
- let isCommutable = 0 in
- defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
- defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
- defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
- defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
- defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
- defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
- defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
- defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
- defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
- defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
-}
/// SS48I_binop_rm - Simple SSE41 binary operator.
multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
(bitconvert (memop_frag addr:$src2)))))]>, OpSize;
}
+let Predicates = [HasAVX] in {
+ let isCommutable = 0 in
+ defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
+ 0>, VEX_4V;
+ defm VPMINSB : SS48I_binop_rm<0x38, "vpminsb", X86smin, v16i8, VR128,
+ memopv2i64, i128mem, 0>, VEX_4V;
+ defm VPMINSD : SS48I_binop_rm<0x39, "vpminsd", X86smin, v4i32, VR128,
+ memopv2i64, i128mem, 0>, VEX_4V;
+ defm VPMINUD : SS48I_binop_rm<0x3B, "vpminud", X86umin, v4i32, VR128,
+ memopv2i64, i128mem, 0>, VEX_4V;
+ defm VPMINUW : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v8i16, VR128,
+ memopv2i64, i128mem, 0>, VEX_4V;
+ defm VPMAXSB : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v16i8, VR128,
+ memopv2i64, i128mem, 0>, VEX_4V;
+ defm VPMAXSD : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v4i32, VR128,
+ memopv2i64, i128mem, 0>, VEX_4V;
+ defm VPMAXUD : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v4i32, VR128,
+ memopv2i64, i128mem, 0>, VEX_4V;
+ defm VPMAXUW : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v8i16, VR128,
+ memopv2i64, i128mem, 0>, VEX_4V;
+ defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
+ 0>, VEX_4V;
+}
+
+let Predicates = [HasAVX2] in {
+ let isCommutable = 0 in
+ defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
+ int_x86_avx2_packusdw>, VEX_4V, VEX_L;
+ defm VPMINSBY : SS48I_binop_rm<0x38, "vpminsb", X86smin, v32i8, VR256,
+ memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
+ defm VPMINSDY : SS48I_binop_rm<0x39, "vpminsd", X86smin, v8i32, VR256,
+ memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
+ defm VPMINUDY : SS48I_binop_rm<0x3B, "vpminud", X86umin, v8i32, VR256,
+ memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
+ defm VPMINUWY : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v16i16, VR256,
+ memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
+ defm VPMAXSBY : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v32i8, VR256,
+ memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
+ defm VPMAXSDY : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v8i32, VR256,
+ memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
+ defm VPMAXUDY : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v8i32, VR256,
+ memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
+ defm VPMAXUWY : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v16i16, VR256,
+ memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
+ defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
+ int_x86_avx2_pmul_dq>, VEX_4V, VEX_L;
+}
+
+let Constraints = "$src1 = $dst" in {
+ let isCommutable = 0 in
+ defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
+ defm PMINSB : SS48I_binop_rm<0x38, "pminsb", X86smin, v16i8, VR128,
+ memopv2i64, i128mem>;
+ defm PMINSD : SS48I_binop_rm<0x39, "pminsd", X86smin, v4i32, VR128,
+ memopv2i64, i128mem>;
+ defm PMINUD : SS48I_binop_rm<0x3B, "pminud", X86umin, v4i32, VR128,
+ memopv2i64, i128mem>;
+ defm PMINUW : SS48I_binop_rm<0x3A, "pminuw", X86umin, v8i16, VR128,
+ memopv2i64, i128mem>;
+ defm PMAXSB : SS48I_binop_rm<0x3C, "pmaxsb", X86smax, v16i8, VR128,
+ memopv2i64, i128mem>;
+ defm PMAXSD : SS48I_binop_rm<0x3D, "pmaxsd", X86smax, v4i32, VR128,
+ memopv2i64, i128mem>;
+ defm PMAXUD : SS48I_binop_rm<0x3F, "pmaxud", X86umax, v4i32, VR128,
+ memopv2i64, i128mem>;
+ defm PMAXUW : SS48I_binop_rm<0x3E, "pmaxuw", X86umax, v8i16, VR128,
+ memopv2i64, i128mem>;
+ defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
+}
+
let Predicates = [HasAVX] in {
defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
memopv2i64, i128mem, 0>, VEX_4V;
(v4f64 VR256:$src2))),
(VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
- def : Pat<(v8f32 (X86Blendps (v8f32 VR256:$src1), (v8f32 VR256:$src2),
+ def : Pat<(v8f32 (X86Blendi (v8f32 VR256:$src1), (v8f32 VR256:$src2),
(imm:$mask))),
- (VBLENDPSYrri VR256:$src2, VR256:$src1, imm:$mask)>;
- def : Pat<(v4f64 (X86Blendpd (v4f64 VR256:$src1), (v4f64 VR256:$src2),
+ (VBLENDPSYrri VR256:$src1, VR256:$src2, imm:$mask)>;
+ def : Pat<(v4f64 (X86Blendi (v4f64 VR256:$src1), (v4f64 VR256:$src2),
(imm:$mask))),
- (VBLENDPDYrri VR256:$src2, VR256:$src1, imm:$mask)>;
+ (VBLENDPDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
- def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2),
+ def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
(imm:$mask))),
- (VPBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>;
- def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2),
+ (VPBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
(imm:$mask))),
- (VBLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>;
- def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2),
+ (VBLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
(imm:$mask))),
- (VBLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>;
+ (VBLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
}
let Predicates = [HasAVX2] in {
def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
(v32i8 VR256:$src2))),
- (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
- def : Pat<(v16i16 (X86Blendpw (v16i16 VR256:$src1), (v16i16 VR256:$src2),
+ (VPBLENDVBYrr VR256:$src1, VR256:$src2, VR256:$mask)>;
+ def : Pat<(v16i16 (X86Blendi (v16i16 VR256:$src1), (v16i16 VR256:$src2),
(imm:$mask))),
- (VPBLENDWYrri VR256:$src2, VR256:$src1, imm:$mask)>;
+ (VPBLENDWYrri VR256:$src1, VR256:$src2, imm:$mask)>;
}
/// SS41I_ternary_int - SSE 4.1 ternary operator
(v2f64 VR128:$src2))),
(BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
- def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2),
+ def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
(imm:$mask))),
- (PBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>;
- def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2),
+ (PBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
(imm:$mask))),
- (BLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>;
- def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2),
+ (BLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
(imm:$mask))),
- (BLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>;
+ (BLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
}
imm:$src3))]>;
def MEM : PseudoI<(outs VR128:$dst),
(ins VR128:$src1, i128mem:$src2, i8imm:$src3),
- [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
- VR128:$src1, (load addr:$src2), imm:$src3))]>;
+ [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1,
+ (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
}
let Defs = [EFLAGS], usesCustomInserter = 1 in {
defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[UseSSE42]>;
}
-let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
- def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
- (ins VR128:$src1, VR128:$src2, i8imm:$src3),
- "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
+multiclass pcmpistrm_SS42AI<string asm> {
+ def rr : SS42AI<0x62, MRMSrcReg, (outs),
+ (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+ !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
+ []>, OpSize;
let mayLoad = 1 in
- def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
- (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
- "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
+ def rm :SS42AI<0x62, MRMSrcMem, (outs),
+ (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
+ !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
+ []>, OpSize;
}
let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
- def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
- (ins VR128:$src1, VR128:$src2, i8imm:$src3),
- "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
- let mayLoad = 1 in
- def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
- (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
- "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
+ let Predicates = [HasAVX] in
+ defm VPCMPISTRM128 : pcmpistrm_SS42AI<"vpcmpistrm">, VEX;
+ defm PCMPISTRM128 : pcmpistrm_SS42AI<"pcmpistrm"> ;
}
// Packed Compare Explicit Length Strings, Return Mask
VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
def MEM : PseudoI<(outs VR128:$dst),
(ins VR128:$src1, i128mem:$src3, i8imm:$src5),
- [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
- VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
+ [(set VR128:$dst, (int_x86_sse42_pcmpestrm128 VR128:$src1, EAX,
+ (bc_v16i8 (memopv2i64 addr:$src3)), EDX, imm:$src5))]>;
}
let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[UseSSE42]>;
}
-let Predicates = [HasAVX],
- Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
- def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
- (ins VR128:$src1, VR128:$src3, i8imm:$src5),
- "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
+multiclass SS42AI_pcmpestrm<string asm> {
+ def rr : SS42AI<0x60, MRMSrcReg, (outs),
+ (ins VR128:$src1, VR128:$src3, i8imm:$src5),
+ !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
+ []>, OpSize;
let mayLoad = 1 in
- def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
- (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
- "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
+ def rm : SS42AI<0x60, MRMSrcMem, (outs),
+ (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
+ !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
+ []>, OpSize;
}
let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
- def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
- (ins VR128:$src1, VR128:$src3, i8imm:$src5),
- "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
- let mayLoad = 1 in
- def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
- (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
- "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
+ let Predicates = [HasAVX] in
+ defm VPCMPESTRM128 : SS42AI_pcmpestrm<"vpcmpestrm">, VEX;
+ defm PCMPESTRM128 : SS42AI_pcmpestrm<"pcmpestrm">;
}
// Packed Compare Implicit Length Strings, Return Index
-let Defs = [ECX, EFLAGS], neverHasSideEffects = 1 in {
- multiclass SS42AI_pcmpistri<string asm> {
- def rr : SS42AI<0x63, MRMSrcReg, (outs),
- (ins VR128:$src1, VR128:$src2, i8imm:$src3),
- !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
- []>, OpSize;
- let mayLoad = 1 in
- def rm : SS42AI<0x63, MRMSrcMem, (outs),
- (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
- !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
- []>, OpSize;
- }
+multiclass pseudo_pcmpistri<string asm> {
+ def REG : PseudoI<(outs GR32:$dst),
+ (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+ [(set GR32:$dst, EFLAGS,
+ (X86pcmpistri VR128:$src1, VR128:$src2, imm:$src3))]>;
+ def MEM : PseudoI<(outs GR32:$dst),
+ (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
+ [(set GR32:$dst, EFLAGS, (X86pcmpistri VR128:$src1,
+ (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
}
-let Predicates = [HasAVX] in
-defm VPCMPISTRI : SS42AI_pcmpistri<"vpcmpistri">, VEX;
-defm PCMPISTRI : SS42AI_pcmpistri<"pcmpistri">;
+let Defs = [EFLAGS], usesCustomInserter = 1 in {
+ defm VPCMPISTRI : pseudo_pcmpistri<"#VPCMPISTRI">, Requires<[HasAVX]>;
+ defm PCMPISTRI : pseudo_pcmpistri<"#PCMPISTRI">, Requires<[UseSSE42]>;
+}
+
+multiclass SS42AI_pcmpistri<string asm> {
+ def rr : SS42AI<0x63, MRMSrcReg, (outs),
+ (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+ !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
+ []>, OpSize;
+ let mayLoad = 1 in
+ def rm : SS42AI<0x63, MRMSrcMem, (outs),
+ (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
+ !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
+ []>, OpSize;
+}
+
+let Defs = [ECX, EFLAGS], neverHasSideEffects = 1 in {
+ let Predicates = [HasAVX] in
+ defm VPCMPISTRI : SS42AI_pcmpistri<"vpcmpistri">, VEX;
+ defm PCMPISTRI : SS42AI_pcmpistri<"pcmpistri">;
+}
// Packed Compare Explicit Length Strings, Return Index
-let Defs = [ECX, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
- multiclass SS42AI_pcmpestri<string asm> {
- def rr : SS42AI<0x61, MRMSrcReg, (outs),
- (ins VR128:$src1, VR128:$src3, i8imm:$src5),
- !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
- []>, OpSize;
- let mayLoad = 1 in
- def rm : SS42AI<0x61, MRMSrcMem, (outs),
- (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
- !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
- []>, OpSize;
- }
+multiclass pseudo_pcmpestri<string asm> {
+ def REG : PseudoI<(outs GR32:$dst),
+ (ins VR128:$src1, VR128:$src3, i8imm:$src5),
+ [(set GR32:$dst, EFLAGS,
+ (X86pcmpestri VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
+ def MEM : PseudoI<(outs GR32:$dst),
+ (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
+ [(set GR32:$dst, EFLAGS,
+ (X86pcmpestri VR128:$src1, EAX, (bc_v16i8 (memopv2i64 addr:$src3)), EDX,
+ imm:$src5))]>;
}
-let Predicates = [HasAVX] in
-defm VPCMPESTRI : SS42AI_pcmpestri<"vpcmpestri">, VEX;
-defm PCMPESTRI : SS42AI_pcmpestri<"pcmpestri">;
+let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
+ defm VPCMPESTRI : pseudo_pcmpestri<"#VPCMPESTRI">, Requires<[HasAVX]>;
+ defm PCMPESTRI : pseudo_pcmpestri<"#PCMPESTRI">, Requires<[UseSSE42]>;
+}
+
+multiclass SS42AI_pcmpestri<string asm> {
+ def rr : SS42AI<0x61, MRMSrcReg, (outs),
+ (ins VR128:$src1, VR128:$src3, i8imm:$src5),
+ !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
+ []>, OpSize;
+ let mayLoad = 1 in
+ def rm : SS42AI<0x61, MRMSrcMem, (outs),
+ (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
+ !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
+ []>, OpSize;
+}
+
+let Defs = [ECX, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
+ let Predicates = [HasAVX] in
+ defm VPCMPESTRI : SS42AI_pcmpestri<"vpcmpestri">, VEX;
+ defm PCMPESTRI : SS42AI_pcmpestri<"pcmpestri">;
+}
//===----------------------------------------------------------------------===//
// SSE4.2 - CRC Instructions
VR256, memopv4i64, i256mem>, VEX_L;
}
+def : Pat<(v4i32 (X86Blendi (v4i32 VR128:$src1), (v4i32 VR128:$src2),
+ imm:$mask)),
+ (VPBLENDDrri VR128:$src1, VR128:$src2, imm:$mask)>;
+def : Pat<(v8i32 (X86Blendi (v8i32 VR256:$src1), (v8i32 VR256:$src2),
+ imm:$mask)),
+ (VPBLENDDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
+
//===----------------------------------------------------------------------===//
// VPBROADCAST - Load from memory and broadcast to all elements of the
// destination operand
projects / oota-llvm.git / blobdiff