AVX512: Fix pcmpeqw predicate of AVX pattern.

[igorb]

Summary

AVX512: Fix predicate of AVX pcmpeqw/b , pcmpgtb/w/d instructions . AVX512 version of this instructions return result in kmask register, so AVX patterns should not be disabled.

[igorb]

change test case

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

 
 def tls64baseaddr : ComplexPattern<i64, 5, "selectTLSADDRAddr",
                                [tglobaltlsaddr], []>;
 
 def vectoraddr : ComplexPattern<iPTR, 5, "selectVectorAddr", [],[SDNPWantParent]>;
 
 //===----------------------------------------------------------------------===//
 // X86 Instruction Predicate Definitions.
+def TruePredicate : Predicate<"true">;
+
 def HasCMov      : Predicate<"Subtarget->hasCMov()">;
 def NoCMov       : Predicate<"!Subtarget->hasCMov()">;
 
 def HasMMX       : Predicate<"Subtarget->hasMMX()">;
 def Has3DNow     : Predicate<"Subtarget->has3DNow()">;
 def Has3DNowA    : Predicate<"Subtarget->has3DNowA()">;
 def HasSSE1      : Predicate<"Subtarget->hasSSE1()">;
 def UseSSE1      : Predicate<"Subtarget->hasSSE1() && !Subtarget->hasAVX()">;

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

 }
 } // Constraints = "$src1 = $dst"
 
 //===---------------------------------------------------------------------===//
 // SSE2 - Packed Integer Comparison Instructions
 //===---------------------------------------------------------------------===//
 
 defm PCMPEQB : PDI_binop_all<0x74, "pcmpeqb", X86pcmpeq, v16i8, v32i8,
-                             SSE_INTALU_ITINS_P, 1, NoVLX_Or_NoBWI>;
+                             SSE_INTALU_ITINS_P, 1, TruePredicate>;
 defm PCMPEQW : PDI_binop_all<0x75, "pcmpeqw", X86pcmpeq, v8i16, v16i16,
-                             SSE_INTALU_ITINS_P, 1, NoVLX_Or_NoBWI>;
+                             SSE_INTALU_ITINS_P, 1, TruePredicate>;
 defm PCMPEQD : PDI_binop_all<0x76, "pcmpeqd", X86pcmpeq, v4i32, v8i32,
-                             SSE_INTALU_ITINS_P, 1, NoVLX>;
+                             SSE_INTALU_ITINS_P, 1, TruePredicate>;
 defm PCMPGTB : PDI_binop_all<0x64, "pcmpgtb", X86pcmpgt, v16i8, v32i8,
-                             SSE_INTALU_ITINS_P, 0, NoVLX_Or_NoBWI>;
+                             SSE_INTALU_ITINS_P, 0, TruePredicate>;
 defm PCMPGTW : PDI_binop_all<0x65, "pcmpgtw", X86pcmpgt, v8i16, v16i16,
-                             SSE_INTALU_ITINS_P, 0, NoVLX_Or_NoBWI>;
+                             SSE_INTALU_ITINS_P, 0, TruePredicate>;
 defm PCMPGTD : PDI_binop_all<0x66, "pcmpgtd", X86pcmpgt, v4i32, v8i32,
-                             SSE_INTALU_ITINS_P, 0, NoVLX>;
+                             SSE_INTALU_ITINS_P, 0, TruePredicate>;
 
 //===---------------------------------------------------------------------===//
 // SSE2 - Packed Integer Shuffle Instructions
 //===---------------------------------------------------------------------===//
 
 let ExeDomain = SSEPackedInt in {
 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt128, ValueType vt256,
                          SDNode OpNode> {

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

define   <2 x double> @_broadcast2xdouble(double %a) {
   ret <2 x double> %c
 }
 
 define <4 x float> @test_x86_fmsub_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) {
   %x = fmul <4 x float> %a0, %a1
   %res = fsub <4 x float> %x, %a2
   ret <4 x float> %res
 }
+
+define <32 x i8> @test_cmpgtb(<32 x i8> %A) {
+; generate the follow code
+;  vpxor   %ymm1, %ymm1, %ymm1
+;  vpcmpgtb %ymm0, %ymm1, %ymm0
+  %B = ashr <32 x i8> %A, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+  ret <32 x i8> %B
+}
+