[x86] small fix for ptestm intrinsics

[AsafBadouh]

Summary

move ptestm{q|d} intrinsics from patterns form (in td file) to the intrinsics table

[delena]

LGTM

E:/Work/LLVM/trunk/include/llvm/IR/IntrinsicsX86.td

def int_x86_avx_ptestz_256 : GCCBuiltin<"__builtin_ia32_ptestz256">,
         Intrinsic<[llvm_i32_ty], [llvm_v4i64_ty,
                   llvm_v4i64_ty], [IntrNoMem]>;
   def int_x86_avx_ptestc_256 : GCCBuiltin<"__builtin_ia32_ptestc256">,
         Intrinsic<[llvm_i32_ty], [llvm_v4i64_ty,
                   llvm_v4i64_ty], [IntrNoMem]>;
   def int_x86_avx_ptestnzc_256 : GCCBuiltin<"__builtin_ia32_ptestnzc256">,
         Intrinsic<[llvm_i32_ty], [llvm_v4i64_ty,
                   llvm_v4i64_ty], [IntrNoMem]>;
-  def int_x86_avx512_mask_ptestm_d_512 : GCCBuiltin<"__builtin_ia32_ptestmd512">,
+  def int_x86_avx512_ptestm_d_512 : GCCBuiltin<"__builtin_ia32_ptestmd512">,
         Intrinsic<[llvm_i16_ty], [llvm_v16i32_ty, llvm_v16i32_ty,
                   llvm_i16_ty], [IntrNoMem]>;
-  def int_x86_avx512_mask_ptestm_q_512 : GCCBuiltin<"__builtin_ia32_ptestmq512">,
+  def int_x86_avx512_ptestm_q_512 : GCCBuiltin<"__builtin_ia32_ptestmq512">,
         Intrinsic<[llvm_i8_ty], [llvm_v8i64_ty, llvm_v8i64_ty,
                   llvm_i8_ty], [IntrNoMem]>;
 
   def int_x86_avx512_ptestm_b_128 : GCCBuiltin<"__builtin_ia32_ptestmb128">,
         Intrinsic<[llvm_i16_ty], [llvm_v16i8_ty,
                    llvm_v16i8_ty, llvm_i16_ty], [IntrNoMem]>;
   def int_x86_avx512_ptestm_b_256 : GCCBuiltin<"__builtin_ia32_ptestmb256">,
         Intrinsic<[llvm_i32_ty], [llvm_v32i8_ty,

E:/Work/LLVM/trunk/lib/Target/X86/X86InstrAVX512.td

 multiclass avx512_vptest_all_forms<bits<8> opc_wb, bits<8> opc_dq, string OpcodeStr,
                                    SDNode OpNode> :
   avx512_vptest_wb <opc_wb, OpcodeStr, OpNode>,
   avx512_vptest_dq<opc_dq, OpcodeStr, OpNode>;
 
 defm VPTESTM   : avx512_vptest_all_forms<0x26, 0x27, "vptestm", X86testm>, T8PD;
 defm VPTESTNM  : avx512_vptest_all_forms<0x26, 0x27, "vptestnm", X86testnm>, T8XS;
 
-def : Pat <(i16 (int_x86_avx512_mask_ptestm_d_512 (v16i32 VR512:$src1),
-                 (v16i32 VR512:$src2), (i16 -1))),
-                 (COPY_TO_REGCLASS (VPTESTMDZrr VR512:$src1, VR512:$src2), GR16)>;
-
-def : Pat <(i8 (int_x86_avx512_mask_ptestm_q_512 (v8i64 VR512:$src1),
-                 (v8i64 VR512:$src2), (i8 -1))),
-                 (COPY_TO_REGCLASS (VPTESTMQZrr VR512:$src1, VR512:$src2), GR8)>;
 
 //===----------------------------------------------------------------------===//
 // AVX-512  Shift instructions
 //===----------------------------------------------------------------------===//
 multiclass avx512_shift_rmi<bits<8> opc, Format ImmFormR, Format ImmFormM,
                          string OpcodeStr, SDNode OpNode, X86VectorVTInfo _> {
   defm ri : AVX512_maskable<opc, ImmFormR, _, (outs _.RC:$dst),
                    (ins _.RC:$src1, u8imm:$src2), OpcodeStr,

E:/Work/LLVM/trunk/lib/Target/X86/X86IntrinsicsInfo.h

   X86_INTRINSIC_DATA(avx512_psad_bw_512, INTR_TYPE_2OP, X86ISD::PSADBW, 0),
   X86_INTRINSIC_DATA(avx512_psll_dq_512, INTR_TYPE_2OP_IMM8, X86ISD::VSHLDQ, 0),
   X86_INTRINSIC_DATA(avx512_psrl_dq_512, INTR_TYPE_2OP_IMM8, X86ISD::VSRLDQ, 0),
   X86_INTRINSIC_DATA(avx512_ptestm_b_128, CMP_MASK, X86ISD::TESTM, 0),
   X86_INTRINSIC_DATA(avx512_ptestm_b_256, CMP_MASK, X86ISD::TESTM, 0),
   X86_INTRINSIC_DATA(avx512_ptestm_b_512, CMP_MASK, X86ISD::TESTM, 0),
   X86_INTRINSIC_DATA(avx512_ptestm_d_128, CMP_MASK, X86ISD::TESTM, 0),
   X86_INTRINSIC_DATA(avx512_ptestm_d_256, CMP_MASK, X86ISD::TESTM, 0),
+  X86_INTRINSIC_DATA(avx512_ptestm_d_512, CMP_MASK, X86ISD::TESTM, 0),
   X86_INTRINSIC_DATA(avx512_ptestm_q_128, CMP_MASK, X86ISD::TESTM, 0),
   X86_INTRINSIC_DATA(avx512_ptestm_q_256, CMP_MASK, X86ISD::TESTM, 0),
+  X86_INTRINSIC_DATA(avx512_ptestm_q_512, CMP_MASK, X86ISD::TESTM, 0),
   X86_INTRINSIC_DATA(avx512_ptestm_w_128, CMP_MASK, X86ISD::TESTM, 0),
   X86_INTRINSIC_DATA(avx512_ptestm_w_256, CMP_MASK, X86ISD::TESTM, 0),
   X86_INTRINSIC_DATA(avx512_ptestm_w_512, CMP_MASK, X86ISD::TESTM, 0),
   X86_INTRINSIC_DATA(avx512_ptestnm_b_128, CMP_MASK, X86ISD::TESTNM, 0),
   X86_INTRINSIC_DATA(avx512_ptestnm_b_256, CMP_MASK, X86ISD::TESTNM, 0),
   X86_INTRINSIC_DATA(avx512_ptestnm_b_512, CMP_MASK, X86ISD::TESTNM, 0),
   X86_INTRINSIC_DATA(avx512_ptestnm_d_128, CMP_MASK, X86ISD::TESTNM, 0),
   X86_INTRINSIC_DATA(avx512_ptestnm_d_256, CMP_MASK, X86ISD::TESTNM, 0),

E:/Work/LLVM/trunk/test/CodeGen/X86/avx512-intrinsics.ll

 ; CHECK-NEXT:    vpaddq %zmm0, %zmm1, %zmm0
 ; CHECK-NEXT:    retq
   %res = call <8 x i64> @llvm.x86.avx512.mask.pabs.q.512(<8 x i64> %x0, <8 x i64> %x1, i8 %x2)
   %res1 = call <8 x i64> @llvm.x86.avx512.mask.pabs.q.512(<8 x i64> %x0, <8 x i64> %x1, i8 -1)
   %res2 = add <8 x i64> %res, %res1
   ret <8 x i64> %res2
 }
 
-define i8 @test_vptestmq(<8 x i64> %a0, <8 x i64> %a1) {
+define i8 @test_vptestmq(<8 x i64> %a0, <8 x i64> %a1, i8 %m) {
 ; CHECK-LABEL: test_vptestmq:
 ; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %edi, %k1
+; CHECK-NEXT:    vptestmq %zmm1, %zmm0, %k0 {%k1}
+; CHECK-NEXT:    kmovw %k0, %ecx
 ; CHECK-NEXT:    vptestmq %zmm1, %zmm0, %k0
 ; CHECK-NEXT:    kmovw %k0, %eax
+; CHECK-NEXT:    addb %cl, %al
 ; CHECK-NEXT:    retq
-  %res = call i8 @llvm.x86.avx512.mask.ptestm.q.512(<8 x i64> %a0, <8 x i64> %a1, i8 -1)
-  ret i8 %res
+  %res = call i8 @llvm.x86.avx512.ptestm.q.512(<8 x i64> %a0, <8 x i64> %a1, i8 -1)
+  %res1 = call i8 @llvm.x86.avx512.ptestm.q.512(<8 x i64> %a0, <8 x i64> %a1, i8 %m)
+  %res2 = add i8 %res1, %res
+  ret i8 %res2
 }
-declare i8 @llvm.x86.avx512.mask.ptestm.q.512(<8 x i64>, <8 x i64>, i8)
+declare i8 @llvm.x86.avx512.ptestm.q.512(<8 x i64>, <8 x i64>, i8)
 
-define i16 @test_vptestmd(<16 x i32> %a0, <16 x i32> %a1) {
+define i16 @test_vptestmd(<16 x i32> %a0, <16 x i32> %a1, i16 %m) {
 ; CHECK-LABEL: test_vptestmd:
 ; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %edi, %k1
+; CHECK-NEXT:    vptestmd %zmm1, %zmm0, %k0 {%k1}
+; CHECK-NEXT:    kmovw %k0, %ecx
 ; CHECK-NEXT:    vptestmd %zmm1, %zmm0, %k0
 ; CHECK-NEXT:    kmovw %k0, %eax
+; CHECK-NEXT:    addl %ecx, %eax
 ; CHECK-NEXT:    retq
-  %res = call i16 @llvm.x86.avx512.mask.ptestm.d.512(<16 x i32> %a0, <16 x i32> %a1, i16 -1)
-  ret i16 %res
+  %res = call i16 @llvm.x86.avx512.ptestm.d.512(<16 x i32> %a0, <16 x i32> %a1, i16 -1)
+  %res1 = call i16 @llvm.x86.avx512.ptestm.d.512(<16 x i32> %a0, <16 x i32> %a1, i16 %m)
+  %res2 = add i16 %res1, %res
+  ret i16 %res2
 }
-declare i16 @llvm.x86.avx512.mask.ptestm.d.512(<16 x i32>, <16 x i32>, i16)
+declare i16 @llvm.x86.avx512.ptestm.d.512(<16 x i32>, <16 x i32>, i16)
 
 define void @test_store1(<16 x float> %data, i8* %ptr, i8* %ptr2, i16 %mask) {
 ; CHECK-LABEL: test_store1:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovw %edx, %k1
 ; CHECK-NEXT:    vmovups %zmm0, (%rdi) {%k1}
 ; CHECK-NEXT:    vmovups %zmm0, (%rsi)
 ; CHECK-NEXT:    retq