AVX512: vpmovzxbw lowering patterns.

[igorb]

Summary

AVX512: Add vpmovzxbw/d/q ,vpmovzxw/d/q ,vpmovzxbdq lowering patterns that support 256bit inputs like AVX patterns ( that are disable in case HasVLX , see SS41I_pmovx_avx2_patterns). Bug Bug-26661

[RKSimon]

Please can you add test cases for xmm -> ymm as well? These fail in a similar manner with -mcpu=skx

define <16 x i16> @zext_16x8_to_16x16(<16 x i8> %a) nounwind readnone {
  %1 = shufflevector <16 x i8> %a, <16 x i8> zeroinitializer, <32 x i32> <i32 0, i32 16, i32 1, i32 16, i32 2, i32 16, i32 3, i32 16, i32 4, i32 16, i32 5, i32 16, i32 6, i32 16, i32 7, i32 16, i32 8, i32 16, i32 9, i32 16, i32 10, i32 16, i32 11, i32 16, i32 12, i32 16, i32 13, i32 16, i32 14, i32 16, i32 15, i32 16>
  %2 = bitcast <32 x i8> %1 to <16 x i16>
  ret <16 x i16> %2
}

[igorb]

add test case (shuffle_zext_16x8_to_16x16)

[RKSimon]

LGTM - thanks.

[delena]

LGTM

[igorb]

Thanks for the review!

llvm/trunk/lib/Target/X86/X86InstrAVX512.td

multiclass avx512_extend_common<bits<8> opc, string OpcodeStr,
   let mayLoad = 1 in {
     defm rm : AVX512_maskable<opc, MRMSrcMem, DestInfo, (outs DestInfo.RC:$dst),
                     (ins x86memop:$src), OpcodeStr ,"$src", "$src",
                     (DestInfo.VT (LdFrag addr:$src))>,
                   EVEX;
   }
 }
 
+// support full register inputs (like SSE paterns)
+multiclass avx512_extend_lowering<SDNode OpNode, X86VectorVTInfo To,
+                                  X86VectorVTInfo From, SubRegIndex SubRegIdx> {
+  def : Pat<(To.VT (OpNode (From.VT From.RC:$src))),
+            (!cast<Instruction>(NAME#To.ZSuffix#"rr") 
+              (EXTRACT_SUBREG From.RC:$src, SubRegIdx))>;
+}
+
 multiclass avx512_extend_BW<bits<8> opc, string OpcodeStr, SDNode OpNode,
           string ExtTy,PatFrag LdFrag = !cast<PatFrag>(ExtTy#"extloadvi8")> {
   let Predicates = [HasVLX, HasBWI] in {
     defm Z128:  avx512_extend_common<opc, OpcodeStr, v8i16x_info,
                     v16i8x_info, i64mem, LdFrag, OpNode>,
                      EVEX_CD8<8, CD8VH>, T8PD, EVEX_V128;
 
     defm Z256:  avx512_extend_common<opc, OpcodeStr, v16i16x_info,
                     v16i8x_info, i128mem, LdFrag, OpNode>,
+                avx512_extend_lowering<OpNode, v16i16x_info, v32i8x_info, sub_xmm>,
                      EVEX_CD8<8, CD8VH>, T8PD, EVEX_V256;
   }
   let Predicates = [HasBWI] in {
     defm Z   :  avx512_extend_common<opc, OpcodeStr, v32i16_info,
                     v32i8x_info, i256mem, LdFrag, OpNode>,
                      EVEX_CD8<8, CD8VH>, T8PD, EVEX_V512;
   }
 }
 
 multiclass avx512_extend_BD<bits<8> opc, string OpcodeStr, SDNode OpNode,
           string ExtTy,PatFrag LdFrag = !cast<PatFrag>(ExtTy#"extloadvi8")> {
   let Predicates = [HasVLX, HasAVX512] in {
     defm Z128:  avx512_extend_common<opc, OpcodeStr, v4i32x_info,
                    v16i8x_info, i32mem, LdFrag, OpNode>,
                          EVEX_CD8<8, CD8VQ>, T8PD, EVEX_V128;
 
     defm Z256:  avx512_extend_common<opc, OpcodeStr, v8i32x_info,
                    v16i8x_info, i64mem, LdFrag, OpNode>,
+                avx512_extend_lowering<OpNode, v8i32x_info, v32i8x_info, sub_xmm>,
                          EVEX_CD8<8, CD8VQ>, T8PD, EVEX_V256;
   }
   let Predicates = [HasAVX512] in {
     defm Z   :  avx512_extend_common<opc, OpcodeStr, v16i32_info,
                    v16i8x_info, i128mem, LdFrag, OpNode>,
                          EVEX_CD8<8, CD8VQ>, T8PD, EVEX_V512;
   }
 }
 
 multiclass avx512_extend_BQ<bits<8> opc, string OpcodeStr, SDNode OpNode,
           string ExtTy,PatFrag LdFrag = !cast<PatFrag>(ExtTy#"extloadvi8")> {
   let Predicates = [HasVLX, HasAVX512] in {
     defm Z128:  avx512_extend_common<opc, OpcodeStr, v2i64x_info,
                    v16i8x_info, i16mem, LdFrag, OpNode>,
                      EVEX_CD8<8, CD8VO>, T8PD, EVEX_V128;
 
     defm Z256:  avx512_extend_common<opc, OpcodeStr, v4i64x_info,
                    v16i8x_info, i32mem, LdFrag, OpNode>,
+                avx512_extend_lowering<OpNode, v4i64x_info, v32i8x_info, sub_xmm>,
                      EVEX_CD8<8, CD8VO>, T8PD, EVEX_V256;
   }
   let Predicates = [HasAVX512] in {
     defm Z   :  avx512_extend_common<opc, OpcodeStr, v8i64_info,
                    v16i8x_info, i64mem, LdFrag, OpNode>,
                      EVEX_CD8<8, CD8VO>, T8PD, EVEX_V512;
   }
 }
 
 multiclass avx512_extend_WD<bits<8> opc, string OpcodeStr, SDNode OpNode,
          string ExtTy,PatFrag LdFrag = !cast<PatFrag>(ExtTy#"extloadvi16")> {
   let Predicates = [HasVLX, HasAVX512] in {
     defm Z128:  avx512_extend_common<opc, OpcodeStr, v4i32x_info,
                    v8i16x_info, i64mem, LdFrag, OpNode>,
                      EVEX_CD8<16, CD8VH>, T8PD, EVEX_V128;
 
     defm Z256:  avx512_extend_common<opc, OpcodeStr, v8i32x_info,
                    v8i16x_info, i128mem, LdFrag, OpNode>,
+                avx512_extend_lowering<OpNode, v8i32x_info, v16i16x_info, sub_xmm>,
                      EVEX_CD8<16, CD8VH>, T8PD, EVEX_V256;
   }
   let Predicates = [HasAVX512] in {
     defm Z   :  avx512_extend_common<opc, OpcodeStr, v16i32_info,
                    v16i16x_info, i256mem, LdFrag, OpNode>,
                      EVEX_CD8<16, CD8VH>, T8PD, EVEX_V512;
   }
 }
 
 multiclass avx512_extend_WQ<bits<8> opc, string OpcodeStr, SDNode OpNode,
          string ExtTy,PatFrag LdFrag = !cast<PatFrag>(ExtTy#"extloadvi16")> {
   let Predicates = [HasVLX, HasAVX512] in {
     defm Z128:  avx512_extend_common<opc, OpcodeStr, v2i64x_info,
                    v8i16x_info, i32mem, LdFrag, OpNode>,
                      EVEX_CD8<16, CD8VQ>, T8PD, EVEX_V128;
 
     defm Z256:  avx512_extend_common<opc, OpcodeStr, v4i64x_info,
                    v8i16x_info, i64mem, LdFrag, OpNode>,
+                avx512_extend_lowering<OpNode, v4i64x_info, v16i16x_info, sub_xmm>,
                      EVEX_CD8<16, CD8VQ>, T8PD, EVEX_V256;
   }
   let Predicates = [HasAVX512] in {
     defm Z   :  avx512_extend_common<opc, OpcodeStr, v8i64_info,
                    v8i16x_info, i128mem, LdFrag, OpNode>,
                      EVEX_CD8<16, CD8VQ>, T8PD, EVEX_V512;
   }
 }
 
 multiclass avx512_extend_DQ<bits<8> opc, string OpcodeStr, SDNode OpNode,
          string ExtTy,PatFrag LdFrag = !cast<PatFrag>(ExtTy#"extloadvi32")> {
 
   let Predicates = [HasVLX, HasAVX512] in {
     defm Z128:  avx512_extend_common<opc, OpcodeStr, v2i64x_info,
                    v4i32x_info, i64mem, LdFrag, OpNode>,
                      EVEX_CD8<32, CD8VH>, T8PD, EVEX_V128;
 
     defm Z256:  avx512_extend_common<opc, OpcodeStr, v4i64x_info,
                    v4i32x_info, i128mem, LdFrag, OpNode>,
+                avx512_extend_lowering<OpNode, v4i64x_info, v8i32x_info, sub_xmm>,
                      EVEX_CD8<32, CD8VH>, T8PD, EVEX_V256;
   }
   let Predicates = [HasAVX512] in {
     defm Z   :  avx512_extend_common<opc, OpcodeStr, v8i64_info,
                    v8i32x_info, i256mem, LdFrag, OpNode>,
                      EVEX_CD8<32, CD8VH>, T8PD, EVEX_V512;
   }
 }

llvm/trunk/test/CodeGen/X86/avx512-ext.ll

 ; SKX-NEXT:    vmovdqu16 %zmm0, %zmm0 {%k1} {z}
 ; SKX-NEXT:    kshiftrq $32, %k1, %k1
 ; SKX-NEXT:    vmovdqu16 %zmm1, %zmm1 {%k1} {z}
 ; SKX-NEXT:    retq
   %ret = select <64 x i1> %mask, <64 x i16> %x, <64 x i16> zeroinitializer
   ret <64 x i16> %ret
 }
 
+define <16 x i16> @shuffle_zext_16x8_to_16x16(<16 x i8> %a) nounwind readnone {
+; ALL-LABEL: shuffle_zext_16x8_to_16x16:
+; ALL:       ## BB#0:
+; ALL-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; ALL-NEXT:    retq
+  %1 = shufflevector <16 x i8> %a, <16 x i8> zeroinitializer, <32 x i32> <i32 0, i32 16, i32 1, i32 16, i32 2, i32 16, i32 3, i32 16, i32 4, i32 16, i32 5, i32 16, i32 6, i32 16, i32 7, i32 16, i32 8, i32 16, i32 9, i32 16, i32 10, i32 16, i32 11, i32 16, i32 12, i32 16, i32 13, i32 16, i32 14, i32 16, i32 15, i32 16>
+  %2 = bitcast <32 x i8> %1 to <16 x i16>
+  ret <16 x i16> %2
+}
+
+define <16 x i16> @zext_32x8_to_16x16(<32 x i8> %a) {
+; ALL-LABEL: zext_32x8_to_16x16:
+; ALL:       ## BB#0:
+; ALL-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; ALL-NEXT:    retq
+  %1 = shufflevector <32 x i8> %a, <32 x i8> zeroinitializer, <32 x i32> <i32 0, i32 32, i32 1, i32 32, i32 2, i32 32, i32 3, i32 32, i32 4, i32 32, i32 5, i32 32, i32 6, i32 32, i32 7, i32 32, i32 8, i32 32, i32 9, i32 32, i32 10, i32 32, i32 11, i32 32, i32 12, i32 32, i32 13, i32 32, i32 14, i32 32, i32 15, i32 32>
+  %2 = bitcast <32 x i8> %1 to <16 x i16>
+  ret <16 x i16> %2
+}
+
+define <8 x i32> @zext_32x8_to_8x32(<32 x i8> %a) {
+; ALL-LABEL: zext_32x8_to_8x32:
+; ALL:       ## BB#0:
+; ALL-NEXT:    vpmovzxbd {{.*#+}} ymm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
+; ALL-NEXT:    retq
+  %1 = shufflevector <32 x i8> %a, <32 x i8> zeroinitializer, <32 x i32> <i32 0, i32 32, i32 32, i32 32, i32 1, i32 32, i32 32, i32 32, i32 2, i32 32, i32 32, i32 32, i32 3, i32 32, i32 32, i32 32, i32 4, i32 32, i32 32, i32 32, i32 5, i32 32, i32 32, i32 32, i32 6, i32 32, i32 32, i32 32, i32 7, i32 32, i32 32, i32 32>
+  %2 = bitcast <32 x i8> %1 to <8 x i32>
+  ret <8 x i32> %2
+}
+
+define <4 x i64> @zext_32x8_to_4x64(<32 x i8> %a) {
+; ALL-LABEL: zext_32x8_to_4x64:
+; ALL:       ## BB#0:
+; ALL-NEXT:    vpmovzxbq {{.*#+}} ymm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero,xmm0[2],zero,zero,zero,zero,zero,zero,zero,xmm0[3],zero,zero,zero,zero,zero,zero,zero
+; ALL-NEXT:    retq
+  %1 = shufflevector <32 x i8> %a, <32 x i8> zeroinitializer, <32 x i32> <i32 0, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 1, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 2, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 3, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32>
+  %2 = bitcast <32 x i8> %1 to <4 x i64>
+  ret <4 x i64> %2
+}
+
+define <8 x i32> @zext_16x16_to_8x32(<16 x i16> %a) {
+; ALL-LABEL: zext_16x16_to_8x32:
+; ALL:       ## BB#0:
+; ALL-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; ALL-NEXT:    retq
+  %1 = shufflevector <16 x i16> %a, <16 x i16> zeroinitializer, <16 x i32> <i32 0, i32 16, i32 1, i32 16, i32 2, i32 16, i32 3, i32 16, i32 4, i32 16, i32 5, i32 16, i32 6, i32 16, i32 7, i32 16>
+  %2 = bitcast <16 x i16> %1 to <8 x i32>
+  ret <8 x i32> %2
+}
+
+define <4 x i64> @zext_16x16_to_4x64(<16 x i16> %a) {
+; ALL-LABEL: zext_16x16_to_4x64:
+; ALL:       ## BB#0:
+; ALL-NEXT:    vpmovzxwq {{.*#+}} ymm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
+; ALL-NEXT:    retq
+  %1 = shufflevector <16 x i16> %a, <16 x i16> zeroinitializer, <16 x i32> <i32 0, i32 16, i32 16, i32 16, i32 1, i32 16, i32 16, i32 16, i32 2, i32 16, i32 16, i32 16, i32 3, i32 16, i32 16, i32 16>
+  %2 = bitcast <16 x i16> %1 to <4 x i64>
+  ret <4 x i64> %2
+}
+
+define <4 x i64> @zext_8x32_to_4x64(<8 x i32> %a) {
+; ALL-LABEL: zext_8x32_to_4x64:
+; ALL:       ## BB#0:
+; ALL-NEXT:    vpmovzxdq {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
+; ALL-NEXT:    retq
+  %1 = shufflevector <8 x i32> %a, <8 x i32> zeroinitializer, <8 x i32> <i32 0, i32 8, i32 1, i32 8, i32 2, i32 8, i32 3, i32 8>
+  %2 = bitcast <8 x i32> %1 to <4 x i64>
+  ret <4 x i64> %2
+}