[X86] AVX512: Only allow k1-k7 as predicates to vpcmp*

[oota-llvm.git] / lib / Target / X86 / X86InstrAVX512.td

diff --git a/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td
index dd1a52facf0117e16b3f0e742b9d33dbec7f3a35..41e900ed11a497fa180aa0d4c10925670bc20964 100644 (file)
--- a/lib/Target/X86/X86InstrAVX512.td
+++ b/lib/Target/X86/X86InstrAVX512.td
@@ -693,22 +693,32 @@ defm VPERMI2PD : avx512_perm_3src<0x77, "vpermi2pd",  VR512, memopv8f64,
 
 multiclass avx512_perm_table_3src<bits<8> opc, string Suffix, RegisterClass RC,
                           PatFrag mem_frag, X86MemOperand x86memop,
-                          SDNode OpNode, ValueType OpVT, RegisterClass KRC> :
+                          SDNode OpNode, ValueType OpVT, RegisterClass KRC,
+                          ValueType MaskVT, RegisterClass MRC> :
         avx512_perm_3src<opc, "vpermt2"##Suffix, RC, mem_frag, x86memop, OpNode,
                          OpVT, KRC> {
   def : Pat<(OpVT (!cast<Intrinsic>("int_x86_avx512_mask_vpermt_"##Suffix##"_512")
                      VR512:$idx, VR512:$src1, VR512:$src2, -1)),
             (!cast<Instruction>(NAME#rr) VR512:$src1, VR512:$idx, VR512:$src2)>;
+
+  def : Pat<(OpVT (!cast<Intrinsic>("int_x86_avx512_mask_vpermt_"##Suffix##"_512")
+                     VR512:$idx, VR512:$src1, VR512:$src2, MRC:$mask)),
+            (!cast<Instruction>(NAME#rrk) VR512:$src1,
+              (MaskVT (COPY_TO_REGCLASS MRC:$mask, KRC)), VR512:$idx, VR512:$src2)>;
 }
 
 defm VPERMT2D  : avx512_perm_table_3src<0x7E, "d",  VR512, memopv16i32, i512mem,
-                               X86VPermv3, v16i32, VK16WM>, EVEX_V512, EVEX_CD8<32, CD8VF>;
+                               X86VPermv3, v16i32, VK16WM, v16i1, GR16>,
+                 EVEX_V512, EVEX_CD8<32, CD8VF>;
 defm VPERMT2Q  : avx512_perm_table_3src<0x7E, "q",  VR512, memopv8i64, i512mem,
-                               X86VPermv3, v8i64, VK8WM>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
+                               X86VPermv3, v8i64, VK8WM, v8i1, GR8>,
+                 EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
 defm VPERMT2PS : avx512_perm_table_3src<0x7F, "ps",  VR512, memopv16f32, i512mem,
-                               X86VPermv3, v16f32, VK16WM>, EVEX_V512, EVEX_CD8<32, CD8VF>;
+                               X86VPermv3, v16f32, VK16WM, v16i1, GR16>,
+                 EVEX_V512, EVEX_CD8<32, CD8VF>;
 defm VPERMT2PD : avx512_perm_table_3src<0x7F, "pd",  VR512, memopv8f64, i512mem,
-                               X86VPermv3, v8f64, VK8WM>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
+                               X86VPermv3, v8f64, VK8WM, v8i1, GR8>,
+                 EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
 
 //===----------------------------------------------------------------------===//
 // AVX-512 - BLEND using mask
@@ -863,7 +873,7 @@ def : Pat<(v8i1 (X86pcmpeqm (v8i32 VR256X:$src1), (v8i32 VR256X:$src2))),
             (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)),
             (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src2, sub_ymm))), VK8)>;
 
-multiclass avx512_icmp_cc<bits<8> opc, RegisterClass KRC,
+multiclass avx512_icmp_cc<bits<8> opc, RegisterClass WMRC, RegisterClass KRC,
               RegisterClass RC, X86MemOperand x86memop, PatFrag memop_frag, 
               SDNode OpNode, ValueType vt, Operand CC, string Suffix> {
   def rri : AVX512AIi8<opc, MRMSrcReg,
@@ -886,7 +896,7 @@ multiclass avx512_icmp_cc<bits<8> opc, RegisterClass KRC,
                   "\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}"),
                [], IIC_SSE_ALU_F32P_RR>, EVEX_4V;
     def rrik_alt : AVX512AIi8<opc, MRMSrcReg,
-               (outs KRC:$dst), (ins KRC:$mask, RC:$src1, RC:$src2, i8imm:$cc),
+               (outs KRC:$dst), (ins WMRC:$mask, RC:$src1, RC:$src2, i8imm:$cc),
                !strconcat("vpcmp", Suffix,
                   "\t{$cc, $src2, $src1, $dst {${mask}}|$dst {${mask}}, $src1, $src2, $cc}"),
                [], IIC_SSE_ALU_F32P_RR>, EVEX_4V, EVEX_K;
@@ -896,24 +906,24 @@ multiclass avx512_icmp_cc<bits<8> opc, RegisterClass KRC,
                   "\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}"),
                [], IIC_SSE_ALU_F32P_RM>, EVEX_4V;
     def rmik_alt : AVX512AIi8<opc, MRMSrcMem,
-               (outs KRC:$dst), (ins KRC:$mask, RC:$src1, x86memop:$src2, i8imm:$cc),
+               (outs KRC:$dst), (ins WMRC:$mask, RC:$src1, x86memop:$src2, i8imm:$cc),
                !strconcat("vpcmp", Suffix,
                   "\t{$cc, $src2, $src1, $dst {${mask}}|$dst {${mask}}, $src1, $src2, $cc}"),
                [], IIC_SSE_ALU_F32P_RM>, EVEX_4V, EVEX_K;
   }
 }
 
-defm VPCMPDZ :  avx512_icmp_cc<0x1F, VK16, VR512, i512mem, memopv16i32,
+defm VPCMPDZ :  avx512_icmp_cc<0x1F, VK16WM, VK16, VR512, i512mem, memopv16i32,
                                X86cmpm, v16i32, AVXCC, "d">,
                 EVEX_V512, EVEX_CD8<32, CD8VF>;
-defm VPCMPUDZ : avx512_icmp_cc<0x1E, VK16, VR512, i512mem, memopv16i32,
+defm VPCMPUDZ : avx512_icmp_cc<0x1E, VK16WM, VK16, VR512, i512mem, memopv16i32,
                                X86cmpmu, v16i32, AVXCC, "ud">,
                 EVEX_V512, EVEX_CD8<32, CD8VF>;
 
-defm VPCMPQZ :  avx512_icmp_cc<0x1F, VK8, VR512, i512mem, memopv8i64,
+defm VPCMPQZ :  avx512_icmp_cc<0x1F, VK8WM, VK8, VR512, i512mem, memopv8i64,
                                X86cmpm, v8i64, AVXCC, "q">,
                 VEX_W, EVEX_V512, EVEX_CD8<64, CD8VF>;
-defm VPCMPUQZ : avx512_icmp_cc<0x1E, VK8, VR512, i512mem, memopv8i64,
+defm VPCMPUQZ : avx512_icmp_cc<0x1E, VK8WM, VK8, VR512, i512mem, memopv8i64,
                                X86cmpmu, v8i64, AVXCC, "uq">,
                 VEX_W, EVEX_V512, EVEX_CD8<64, CD8VF>;