aboutsummaryrefslogtreecommitdiffstats
path: root/lib/Target/X86/X86InstrAVX512.td
diff options
context:
space:
mode:
Diffstat (limited to 'lib/Target/X86/X86InstrAVX512.td')
-rw-r--r--lib/Target/X86/X86InstrAVX512.td155
1 files changed, 63 insertions, 92 deletions
diff --git a/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td
index 509602f..0959162 100644
--- a/lib/Target/X86/X86InstrAVX512.td
+++ b/lib/Target/X86/X86InstrAVX512.td
@@ -2971,60 +2971,36 @@ multiclass avx512_binop_rm_vl_all<bits<8> opc_b, bits<8> opc_w,
itins, HasBWI, IsCommutable>;
}
-multiclass avx512_binop_rm2<bits<8> opc, string OpcodeStr, ValueType DstVT,
- ValueType SrcVT, RegisterClass KRC, RegisterClass RC,
- PatFrag memop_frag, X86MemOperand x86memop,
- PatFrag scalar_mfrag, X86MemOperand x86scalar_mop,
- string BrdcstStr, OpndItins itins, bit IsCommutable = 0> {
- let isCommutable = IsCommutable in
- {
- def rr : AVX512BI<opc, MRMSrcReg, (outs RC:$dst),
- (ins RC:$src1, RC:$src2),
- !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- []>, EVEX_4V;
- def rrk : AVX512BI<opc, MRMSrcReg, (outs RC:$dst),
- (ins KRC:$mask, RC:$src1, RC:$src2),
- !strconcat(OpcodeStr,
- "\t{$src2, $src1, $dst {${mask}}|$dst {${mask}}, $src1, $src2}"),
- [], itins.rr>, EVEX_4V, EVEX_K;
- def rrkz : AVX512BI<opc, MRMSrcReg, (outs RC:$dst),
- (ins KRC:$mask, RC:$src1, RC:$src2),
- !strconcat(OpcodeStr, "\t{$src2, $src1, $dst {${mask}} {z}" ,
- "|$dst {${mask}} {z}, $src1, $src2}"),
- [], itins.rr>, EVEX_4V, EVEX_KZ;
- }
+multiclass avx512_binop_rm2<bits<8> opc, string OpcodeStr, OpndItins itins,
+ SDNode OpNode,X86VectorVTInfo _Src,
+ X86VectorVTInfo _Dst, bit IsCommutable = 0> {
+ defm rr : AVX512_maskable<opc, MRMSrcReg, _Dst, (outs _Dst.RC:$dst),
+ (ins _Src.RC:$src1, _Src.RC:$src2), OpcodeStr,
+ "$src2, $src1","$src1, $src2",
+ (_Dst.VT (OpNode
+ (_Src.VT _Src.RC:$src1),
+ (_Src.VT _Src.RC:$src2))),
+ "",itins.rr, IsCommutable>,
+ AVX512BIBase, EVEX_4V;
let mayLoad = 1 in {
- def rm : AVX512BI<opc, MRMSrcMem, (outs RC:$dst),
- (ins RC:$src1, x86memop:$src2),
- !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- []>, EVEX_4V;
- def rmk : AVX512BI<opc, MRMSrcMem, (outs RC:$dst),
- (ins KRC:$mask, RC:$src1, x86memop:$src2),
- !strconcat(OpcodeStr,
- "\t{$src2, $src1, $dst {${mask}}|$dst {${mask}}, $src1, $src2}"),
- [], itins.rm>, EVEX_4V, EVEX_K;
- def rmkz : AVX512BI<opc, MRMSrcMem, (outs RC:$dst),
- (ins KRC:$mask, RC:$src1, x86memop:$src2),
- !strconcat(OpcodeStr,
- "\t{$src2, $src1, $dst {${mask}} {z}|$dst {${mask}} {z}, $src1, $src2}"),
- [], itins.rm>, EVEX_4V, EVEX_KZ;
- def rmb : AVX512BI<opc, MRMSrcMem, (outs RC:$dst),
- (ins RC:$src1, x86scalar_mop:$src2),
- !strconcat(OpcodeStr, "\t{${src2}", BrdcstStr,
- ", $src1, $dst|$dst, $src1, ${src2}", BrdcstStr, "}"),
- [], itins.rm>, EVEX_4V, EVEX_B;
- def rmbk : AVX512BI<opc, MRMSrcMem, (outs RC:$dst),
- (ins KRC:$mask, RC:$src1, x86scalar_mop:$src2),
- !strconcat(OpcodeStr, "\t{${src2}", BrdcstStr,
- ", $src1, $dst {${mask}}|$dst {${mask}}, $src1, ${src2}",
- BrdcstStr, "}"),
- [], itins.rm>, EVEX_4V, EVEX_B, EVEX_K;
- def rmbkz : AVX512BI<opc, MRMSrcMem, (outs RC:$dst),
- (ins KRC:$mask, RC:$src1, x86scalar_mop:$src2),
- !strconcat(OpcodeStr, "\t{${src2}", BrdcstStr,
- ", $src1, $dst {${mask}} {z}|$dst {${mask}} {z}, $src1, ${src2}",
- BrdcstStr, "}"),
- [], itins.rm>, EVEX_4V, EVEX_B, EVEX_KZ;
+ defm rm : AVX512_maskable<opc, MRMSrcMem, _Dst, (outs _Dst.RC:$dst),
+ (ins _Src.RC:$src1, _Src.MemOp:$src2), OpcodeStr,
+ "$src2, $src1", "$src1, $src2",
+ (_Dst.VT (OpNode (_Src.VT _Src.RC:$src1),
+ (bitconvert (_Src.LdFrag addr:$src2)))),
+ "", itins.rm>,
+ AVX512BIBase, EVEX_4V;
+
+ defm rmb : AVX512_maskable<opc, MRMSrcMem, _Dst, (outs _Dst.RC:$dst),
+ (ins _Src.RC:$src1, _Dst.ScalarMemOp:$src2),
+ OpcodeStr,
+ "${src2}"##_Dst.BroadcastStr##", $src1",
+ "$src1, ${src2}"##_Dst.BroadcastStr,
+ (_Dst.VT (OpNode (_Src.VT _Src.RC:$src1), (bc_v16i32
+ (_Dst.VT (X86VBroadcast
+ (_Dst.ScalarLdFrag addr:$src2)))))),
+ "", itins.rm>,
+ AVX512BIBase, EVEX_4V, EVEX_B;
}
}
@@ -3039,24 +3015,13 @@ defm VPMULLW : avx512_binop_rm_vl_w<0xD5, "vpmull", mul,
defm VPMULLQ : avx512_binop_rm_vl_q<0x40, "vpmull", mul,
SSE_INTALU_ITINS_P, HasDQI, 1>, T8PD;
-defm VPMULDQZ : avx512_binop_rm2<0x28, "vpmuldq", v8i64, v16i32, VK8WM, VR512,
- loadv8i64, i512mem, loadi64, i64mem, "{1to8}",
- SSE_INTALU_ITINS_P, 1>, T8PD, EVEX_V512,
- EVEX_CD8<64, CD8VF>, VEX_W;
-
-defm VPMULUDQZ : avx512_binop_rm2<0xF4, "vpmuludq", v8i64, v16i32, VK8WM, VR512,
- loadv8i64, i512mem, loadi64, i64mem, "{1to8}",
- SSE_INTMUL_ITINS_P, 1>, EVEX_V512, EVEX_CD8<64, CD8VF>, VEX_W;
+defm VPMULDQZ : avx512_binop_rm2<0x28, "vpmuldq", SSE_INTALU_ITINS_P,
+ X86pmuldq, v16i32_info, v8i64_info, 1>,
+ T8PD, EVEX_V512, EVEX_CD8<64, CD8VF>, VEX_W;
-def : Pat<(v8i64 (X86pmuludq (v16i32 VR512:$src1), (v16i32 VR512:$src2))),
- (VPMULUDQZrr VR512:$src1, VR512:$src2)>;
-
-def : Pat<(v8i64 (int_x86_avx512_mask_pmulu_dq_512 (v16i32 VR512:$src1),
- (v16i32 VR512:$src2), (bc_v8i64 (v16i32 immAllZerosV)), (i8 -1))),
- (VPMULUDQZrr VR512:$src1, VR512:$src2)>;
-def : Pat<(v8i64 (int_x86_avx512_mask_pmul_dq_512 (v16i32 VR512:$src1),
- (v16i32 VR512:$src2), (bc_v8i64 (v16i32 immAllZerosV)), (i8 -1))),
- (VPMULDQZrr VR512:$src1, VR512:$src2)>;
+defm VPMULUDQZ : avx512_binop_rm2<0xF4, "vpmuludq", SSE_INTMUL_ITINS_P,
+ X86pmuludq, v16i32_info, v8i64_info, 1>,
+ EVEX_V512, EVEX_CD8<64, CD8VF>, VEX_W;
defm VPMAXSB : avx512_binop_rm_vl_b<0x3C, "vpmaxs", X86smax,
SSE_INTALU_ITINS_P, HasBWI, 1>, T8PD;
@@ -3208,7 +3173,7 @@ defm VPOR : avx512_binop_rm_vl_dq<0xEB, 0xEB, "vpor", or,
defm VPXOR : avx512_binop_rm_vl_dq<0xEF, 0xEF, "vpxor", xor,
SSE_INTALU_ITINS_P, HasAVX512, 1>;
defm VPANDN : avx512_binop_rm_vl_dq<0xDF, 0xDF, "vpandn", X86andnp,
- SSE_INTALU_ITINS_P, HasAVX512, 1>;
+ SSE_INTALU_ITINS_P, HasAVX512, 0>;
//===----------------------------------------------------------------------===//
// AVX-512 FP arithmetic
@@ -3743,16 +3708,19 @@ multiclass avx512_fma3p_rm<bits<8> opc, string OpcodeStr, X86VectorVTInfo _,
defm mb: AVX512_maskable_3src<opc, MRMSrcMem, _, (outs _.RC:$dst),
(ins _.RC:$src2, _.ScalarMemOp:$src3),
- OpcodeStr, !strconcat("${src3}", _.BroadcastStr,", $src2"), !strconcat("$src2, ${src3}", _.BroadcastStr ),
- (OpNode _.RC:$src1, _.RC:$src2,(_.VT (X86VBroadcast (_.ScalarLdFrag addr:$src3))))>,
+ OpcodeStr, !strconcat("${src3}", _.BroadcastStr,", $src2"),
+ !strconcat("$src2, ${src3}", _.BroadcastStr ),
+ (OpNode _.RC:$src1,
+ _.RC:$src2,(_.VT (X86VBroadcast (_.ScalarLdFrag addr:$src3))))>,
AVX512FMA3Base, EVEX_B;
}
} // Constraints = "$src1 = $dst"
let Constraints = "$src1 = $dst" in {
// Omitting the parameter OpNode (= null_frag) disables ISel pattern matching.
-multiclass avx512_fma3_round_rrb<bits<8> opc, string OpcodeStr, X86VectorVTInfo _,
- SDPatternOperator OpNode> {
+multiclass avx512_fma3_round_rrb<bits<8> opc, string OpcodeStr,
+ X86VectorVTInfo _,
+ SDPatternOperator OpNode> {
defm rb: AVX512_maskable_3src<opc, MRMSrcReg, _, (outs _.RC:$dst),
(ins _.RC:$src2, _.RC:$src3, AVX512RC:$rc),
OpcodeStr, "$rc, $src3, $src2", "$src2, $src3, $rc",
@@ -3772,7 +3740,6 @@ multiclass avx512_fma3p_forms<bits<8> opc213, bits<8> opc231,
SDPatternOperator OpNode> {
defm v213r : avx512_fma3p_rm<opc213, !strconcat(OpcodeStr, "213", VTI.Suffix),
VTI, OpNode>, EVEX_CD8<VTI.EltSize, CD8VF>;
-
defm v231r : avx512_fma3p_rm<opc231, !strconcat(OpcodeStr, "231", VTI.Suffix),
VTI>, EVEX_CD8<VTI.EltSize, CD8VF>;
}
@@ -3794,12 +3761,14 @@ let ExeDomain = SSEPackedSingle in {
let ExeDomain = SSEPackedDouble in {
defm NAME##PDZ : avx512_fma3p_forms<opc213, opc231, OpcodeStr,
v8f64_info, OpNode>,
- avx512_fma3_round_forms<opc213, OpcodeStr,
- v8f64_info, OpNodeRnd>, EVEX_V512, VEX_W;
+ avx512_fma3_round_forms<opc213, OpcodeStr, v8f64_info,
+ OpNodeRnd>, EVEX_V512, VEX_W;
defm NAME##PDZ256 : avx512_fma3p_forms<opc213, opc231, OpcodeStr,
- v4f64x_info, OpNode>, EVEX_V256, VEX_W;
+ v4f64x_info, OpNode>,
+ EVEX_V256, VEX_W;
defm NAME##PDZ128 : avx512_fma3p_forms<opc213, opc231, OpcodeStr,
- v2f64x_info, OpNode>, EVEX_V128, VEX_W;
+ v2f64x_info, OpNode>,
+ EVEX_V128, VEX_W;
}
}
@@ -3830,26 +3799,29 @@ multiclass avx512_fma3p_m132<bits<8> opc, string OpcodeStr, SDNode OpNode,
}
} // Constraints = "$src1 = $dst"
-
-multiclass avx512_fma3p_m132_f<bits<8> opc,
- string OpcodeStr,
- SDNode OpNode> {
+multiclass avx512_fma3p_m132_f<bits<8> opc, string OpcodeStr, SDNode OpNode> {
let ExeDomain = SSEPackedSingle in {
defm NAME##PSZ : avx512_fma3p_m132<opc, OpcodeStr##ps,
- OpNode,v16f32_info>, EVEX_V512, EVEX_CD8<32, CD8VF>;
+ OpNode,v16f32_info>, EVEX_V512,
+ EVEX_CD8<32, CD8VF>;
defm NAME##PSZ256 : avx512_fma3p_m132<opc, OpcodeStr##ps,
- OpNode, v8f32x_info>, EVEX_V256, EVEX_CD8<32, CD8VF>;
+ OpNode, v8f32x_info>, EVEX_V256,
+ EVEX_CD8<32, CD8VF>;
defm NAME##PSZ128 : avx512_fma3p_m132<opc, OpcodeStr##ps,
- OpNode, v4f32x_info>, EVEX_V128, EVEX_CD8<32, CD8VF>;
+ OpNode, v4f32x_info>, EVEX_V128,
+ EVEX_CD8<32, CD8VF>;
}
let ExeDomain = SSEPackedDouble in {
defm NAME##PDZ : avx512_fma3p_m132<opc, OpcodeStr##pd,
- OpNode, v8f64_info>, EVEX_V512, VEX_W, EVEX_CD8<32, CD8VF>;
+ OpNode, v8f64_info>, EVEX_V512,
+ VEX_W, EVEX_CD8<32, CD8VF>;
defm NAME##PDZ256 : avx512_fma3p_m132<opc, OpcodeStr##pd,
- OpNode, v4f64x_info>, EVEX_V256, VEX_W, EVEX_CD8<32, CD8VF>;
+ OpNode, v4f64x_info>, EVEX_V256,
+ VEX_W, EVEX_CD8<32, CD8VF>;
defm NAME##PDZ128 : avx512_fma3p_m132<opc, OpcodeStr##pd,
- OpNode, v2f64x_info>, EVEX_V128, VEX_W, EVEX_CD8<32, CD8VF>;
+ OpNode, v2f64x_info>, EVEX_V128,
+ VEX_W, EVEX_CD8<32, CD8VF>;
}
}
@@ -3860,7 +3832,6 @@ defm VFMSUBADD132 : avx512_fma3p_m132_f<0x97, "vfmsubadd132", X86Fmsubadd>;
defm VFNMADD132 : avx512_fma3p_m132_f<0x9C, "vfnmadd132", X86Fnmadd>;
defm VFNMSUB132 : avx512_fma3p_m132_f<0x9E, "vfnmsub132", X86Fnmsub>;
-
// Scalar FMA
let Constraints = "$src1 = $dst" in {
multiclass avx512_fma3s_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
@@ -3883,7 +3854,6 @@ multiclass avx512_fma3s_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
(OpVT (OpNode RC:$src2, RC:$src1,
(mem_frag addr:$src3))))]>;
}
-
} // Constraints = "$src1 = $dst"
defm VFMADDSSZ : avx512_fma3s_rm<0xA9, "vfmadd213ss", X86Fmadd, FR32X,
@@ -3920,6 +3890,7 @@ let hasSideEffects = 0 in {
EVEX_4V;
} // hasSideEffects = 0
}
+
let Predicates = [HasAVX512] in {
defm VCVTSI2SSZ : avx512_vcvtsi<0x2A, GR32, FR32X, i32mem, "cvtsi2ss{l}">,
XS, VEX_LIG, EVEX_CD8<32, CD8VT1>;
generated by cgit v1.1 at 2024-04-27 23:23:13 +0000