Fix ISA disabling in case AVX512VL present

[igorb]

Summary

AVX : Fix ISA disabling in case AVX512VL , some instructions should be disabled only if AVX512BW and AVX512VL present.
Tests added.

[delena]

LGTM

llvm/trunk/lib/Target/X86/X86InstrInfo.td

def HasDQI       : Predicate<"Subtarget->hasDQI()">,
                      AssemblerPredicate<"FeatureDQI", "AVX-512 DQ ISA">;
 def NoDQI        : Predicate<"!Subtarget->hasDQI()">;
 def HasBWI       : Predicate<"Subtarget->hasBWI()">,
                      AssemblerPredicate<"FeatureBWI", "AVX-512 BW ISA">;
 def NoBWI        : Predicate<"!Subtarget->hasBWI()">;
 def HasVLX       : Predicate<"Subtarget->hasVLX()">,
                      AssemblerPredicate<"FeatureVLX", "AVX-512 VL ISA">;
 def NoVLX        : Predicate<"!Subtarget->hasVLX()">;
+def NoVLX_Or_NoBWI : Predicate<"!Subtarget->hasVLX() || !Subtarget->hasBWI()">;
 
 def HasPOPCNT    : Predicate<"Subtarget->hasPOPCNT()">;
 def HasAES       : Predicate<"Subtarget->hasAES()">;
 def HasPCLMUL    : Predicate<"Subtarget->hasPCLMUL()">;
 def HasFMA       : Predicate<"Subtarget->hasFMA()">;
 def UseFMAOnAVX  : Predicate<"Subtarget->hasFMA() && !Subtarget->hasAVX512()">;
 def HasFMA4      : Predicate<"Subtarget->hasFMA4()">;
 def HasXOP       : Predicate<"Subtarget->hasXOP()">;

llvm/trunk/lib/Target/X86/X86InstrSSE.td

def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
            !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
        [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
                                      (bitconvert (memop_frag addr:$src2)))))]>,
        Sched<[itins.Sched.Folded, ReadAfterLd]>;
 }
 } // ExeDomain = SSEPackedInt
 
 defm PADDB   : PDI_binop_all<0xFC, "paddb", add, v16i8, v32i8,
-                             SSE_INTALU_ITINS_P, 1, NoBWI>;
+                             SSE_INTALU_ITINS_P, 1, NoVLX_Or_NoBWI>;
 defm PADDW   : PDI_binop_all<0xFD, "paddw", add, v8i16, v16i16,
-                             SSE_INTALU_ITINS_P, 1, NoBWI>;
+                             SSE_INTALU_ITINS_P, 1, NoVLX_Or_NoBWI>;
 defm PADDD   : PDI_binop_all<0xFE, "paddd", add, v4i32, v8i32,
                              SSE_INTALU_ITINS_P, 1, NoVLX>;
 defm PADDQ   : PDI_binop_all<0xD4, "paddq", add, v2i64, v4i64,
                              SSE_INTALUQ_ITINS_P, 1, NoVLX>;
 defm PMULLW  : PDI_binop_all<0xD5, "pmullw", mul, v8i16, v16i16,
-                             SSE_INTMUL_ITINS_P, 1, NoBWI>;
+                             SSE_INTMUL_ITINS_P, 1, NoVLX_Or_NoBWI>;
 defm PMULHUW : PDI_binop_all<0xE4, "pmulhuw", mulhu, v8i16, v16i16,
-                             SSE_INTMUL_ITINS_P, 1, NoBWI>;
+                             SSE_INTMUL_ITINS_P, 1, NoVLX_Or_NoBWI>;
 defm PMULHW  : PDI_binop_all<0xE5, "pmulhw", mulhs, v8i16, v16i16,
-                             SSE_INTMUL_ITINS_P, 1, NoBWI>;
+                             SSE_INTMUL_ITINS_P, 1, NoVLX_Or_NoBWI>;
 defm PSUBB   : PDI_binop_all<0xF8, "psubb", sub, v16i8, v32i8,
-                             SSE_INTALU_ITINS_P, 0, NoBWI>;
+                             SSE_INTALU_ITINS_P, 0, NoVLX_Or_NoBWI>;
 defm PSUBW   : PDI_binop_all<0xF9, "psubw", sub, v8i16, v16i16,
-                             SSE_INTALU_ITINS_P, 0, NoBWI>;
+                             SSE_INTALU_ITINS_P, 0, NoVLX_Or_NoBWI>;
 defm PSUBD   : PDI_binop_all<0xFA, "psubd", sub, v4i32, v8i32,
                              SSE_INTALU_ITINS_P, 0, NoVLX>;
 defm PSUBQ   : PDI_binop_all<0xFB, "psubq", sub, v2i64, v4i64,
                              SSE_INTALUQ_ITINS_P, 0, NoVLX>;
 defm PSUBUSB : PDI_binop_all<0xD8, "psubusb", X86subus, v16i8, v32i8,
-                             SSE_INTALU_ITINS_P, 0, NoBWI>;
+                             SSE_INTALU_ITINS_P, 0, NoVLX_Or_NoBWI>;
 defm PSUBUSW : PDI_binop_all<0xD9, "psubusw", X86subus, v8i16, v16i16,
-                             SSE_INTALU_ITINS_P, 0, NoBWI>;
+                             SSE_INTALU_ITINS_P, 0, NoVLX_Or_NoBWI>;
 defm PMINUB  : PDI_binop_all<0xDA, "pminub", umin, v16i8, v32i8,
-                             SSE_INTALU_ITINS_P, 1, NoBWI>;
+                             SSE_INTALU_ITINS_P, 1, NoVLX_Or_NoBWI>;
 defm PMINSW  : PDI_binop_all<0xEA, "pminsw", smin, v8i16, v16i16,
-                             SSE_INTALU_ITINS_P, 1, NoBWI>;
+                             SSE_INTALU_ITINS_P, 1, NoVLX_Or_NoBWI>;
 defm PMAXUB  : PDI_binop_all<0xDE, "pmaxub", umax, v16i8, v32i8,
-                             SSE_INTALU_ITINS_P, 1, NoBWI>;
+                             SSE_INTALU_ITINS_P, 1, NoVLX_Or_NoBWI>;
 defm PMAXSW  : PDI_binop_all<0xEE, "pmaxsw", smax, v8i16, v16i16,
-                             SSE_INTALU_ITINS_P, 1, NoBWI>;
+                             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_psubs_b, SSE_INTALU_ITINS_P, 0>;
 defm PSUBSW  : PDI_binop_all_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
                                  int_x86_avx2_psubs_w, SSE_INTALU_ITINS_P, 0>;
 defm PADDSB  : PDI_binop_all_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
                                  int_x86_avx2_padds_b, SSE_INTALU_ITINS_P, 1>;

 }
 } // Constraints = "$src1 = $dst"
 
 //===---------------------------------------------------------------------===//
 // SSE2 - Packed Integer Comparison Instructions
 //===---------------------------------------------------------------------===//
 
 defm PCMPEQB : PDI_binop_all<0x74, "pcmpeqb", X86pcmpeq, v16i8, v32i8,
-                             SSE_INTALU_ITINS_P, 1, NoBWI>;
+                             SSE_INTALU_ITINS_P, 1, NoVLX_Or_NoBWI>;
 defm PCMPEQW : PDI_binop_all<0x75, "pcmpeqw", X86pcmpeq, v8i16, v16i16,
-                             SSE_INTALU_ITINS_P, 1, NoBWI>;
+                             SSE_INTALU_ITINS_P, 1, NoVLX_Or_NoBWI>;
 defm PCMPEQD : PDI_binop_all<0x76, "pcmpeqd", X86pcmpeq, v4i32, v8i32,
                              SSE_INTALU_ITINS_P, 1, NoVLX>;
 defm PCMPGTB : PDI_binop_all<0x64, "pcmpgtb", X86pcmpgt, v16i8, v32i8,
-                             SSE_INTALU_ITINS_P, 0, NoBWI>;
+                             SSE_INTALU_ITINS_P, 0, NoVLX_Or_NoBWI>;
 defm PCMPGTW : PDI_binop_all<0x65, "pcmpgtw", X86pcmpgt, v8i16, v16i16,
-                             SSE_INTALU_ITINS_P, 0, NoBWI>;
+                             SSE_INTALU_ITINS_P, 0, NoVLX_Or_NoBWI>;
 defm PCMPGTD : PDI_binop_all<0x66, "pcmpgtd", X86pcmpgt, v4i32, v8i32,
                              SSE_INTALU_ITINS_P, 0, NoVLX>;
 
 //===---------------------------------------------------------------------===//
 // SSE2 - Packed Integer Shuffle Instructions
 //===---------------------------------------------------------------------===//
 
 let ExeDomain = SSEPackedInt in {

llvm/trunk/test/CodeGen/X86/avx-isa-check.ll

 ; check AVX2 instructions that are disabled in case avx512VL/avx512BW present
    
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -show-mc-encoding -mcpu=core-avx2 -mattr=+avx2                 -o /dev/null
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -show-mc-encoding -mcpu=knl                                    -o /dev/null
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -show-mc-encoding -mcpu=knl  -mattr=+avx512vl                  -o /dev/null
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -show-mc-encoding -mcpu=knl  -mattr=+avx512bw                  -o /dev/null
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -show-mc-encoding -mcpu=knl  -mattr=+avx512vl -mattr=+avx512bw -o /dev/null
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -show-mc-encoding -mcpu=skx                                    -o /dev/null
 
 define <4 x i64> @vpand_256(<4 x i64> %a, <4 x i64> %b) nounwind uwtable readnone ssp {
   ; Force the execution domain with an add.
   %a2 = add <4 x i64> %a, <i64 1, i64 1, i64 1, i64 1>
   %x = and <4 x i64> %a2, %b
   ret <4 x i64> %x