[X86] Add IFMA instructions to the load folding tables and make them commutable for...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314080 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td
index 1155b6dcb4953b3e01d23c745dd6966e44b356ec..a8b7c80cdab95e1a5154a1c973518661c86eb8b4 100644 (file)
--- a/lib/Target/X86/X86InstrAVX512.td
+++ b/lib/Target/X86/X86InstrAVX512.td
@@ -6484,7 +6484,7 @@ multiclass avx512_pmadd52_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
   defm r: AVX512_maskable_3src<opc, MRMSrcReg, _, (outs _.RC:$dst),
           (ins _.RC:$src2, _.RC:$src3),
           OpcodeStr, "$src3, $src2", "$src2, $src3",
-          (_.VT (OpNode _.RC:$src1, _.RC:$src2, _.RC:$src3))>,
+          (_.VT (OpNode _.RC:$src1, _.RC:$src2, _.RC:$src3)), 1, 1>,
          AVX512FMA3Base;
 
   defm m: AVX512_maskable_3src<opc, MRMSrcMem, _, (outs _.RC:$dst),

diff --git a/lib/Target/X86/X86InstrInfo.cpp b/lib/Target/X86/X86InstrInfo.cpp
index 4b56807cffcae5362703ce6e91121e9c4a2ae33f..0561bcd8d0ab77ed7a8f7edda75c929bd64622c5 100644 (file)
--- a/lib/Target/X86/X86InstrInfo.cpp
+++ b/lib/Target/X86/X86InstrInfo.cpp
@@ -2528,6 +2528,8 @@ X86InstrInfo::X86InstrInfo(X86Subtarget &STI)
     { X86::VPERMT2PDrr,           X86::VPERMT2PDrm,           0 },
     { X86::VPERMT2Qrr,            X86::VPERMT2Qrm,            0 },
     { X86::VPERMT2Wrr,            X86::VPERMT2Wrm,            0 },
+    { X86::VPMADD52HUQZr,         X86::VPMADD52HUQZm,         0 },
+    { X86::VPMADD52LUQZr,         X86::VPMADD52LUQZm,         0 },
     { X86::VPTERNLOGDZrri,        X86::VPTERNLOGDZrmi,        0 },
     { X86::VPTERNLOGQZrri,        X86::VPTERNLOGQZrmi,        0 },
 
@@ -2544,6 +2546,8 @@ X86InstrInfo::X86InstrInfo(X86Subtarget &STI)
     { X86::VPERMT2PS256rr,        X86::VPERMT2PS256rm,        0 },
     { X86::VPERMT2Q256rr,         X86::VPERMT2Q256rm,         0 },
     { X86::VPERMT2W256rr,         X86::VPERMT2W256rm,         0 },
+    { X86::VPMADD52HUQZ256r,      X86::VPMADD52HUQZ256m,      0 },
+    { X86::VPMADD52LUQZ256r,      X86::VPMADD52LUQZ256m,      0 },
     { X86::VPTERNLOGDZ256rri,     X86::VPTERNLOGDZ256rmi,     0 },
     { X86::VPTERNLOGQZ256rri,     X86::VPTERNLOGQZ256rmi,     0 },
 
@@ -2560,6 +2564,8 @@ X86InstrInfo::X86InstrInfo(X86Subtarget &STI)
     { X86::VPERMT2PS128rr,        X86::VPERMT2PS128rm,        0 },
     { X86::VPERMT2Q128rr,         X86::VPERMT2Q128rm,         0 },
     { X86::VPERMT2W128rr,         X86::VPERMT2W128rm,         0 },
+    { X86::VPMADD52HUQZ128r,      X86::VPMADD52HUQZ128m,      0 },
+    { X86::VPMADD52LUQZ128r,      X86::VPMADD52LUQZ128m,      0 },
     { X86::VPTERNLOGDZ128rri,     X86::VPTERNLOGDZ128rmi,     0 },
     { X86::VPTERNLOGQZ128rri,     X86::VPTERNLOGQZ128rmi,     0 },
 
@@ -3234,6 +3240,8 @@ X86InstrInfo::X86InstrInfo(X86Subtarget &STI)
     { X86::VPERMT2Qrrk,        X86::VPERMT2Qrmk,          0 },
     { X86::VPERMT2Wrrk,        X86::VPERMT2Wrmk,          0 },
     { X86::VPERMWZrrk,         X86::VPERMWZrmk,           0 },
+    { X86::VPMADD52HUQZrk,     X86::VPMADD52HUQZmk,       0 },
+    { X86::VPMADD52LUQZrk,     X86::VPMADD52LUQZmk,       0 },
     { X86::VPMADDUBSWZrrk,     X86::VPMADDUBSWZrmk,       0 },
     { X86::VPMADDWDZrrk,       X86::VPMADDWDZrmk,         0 },
     { X86::VPMAXSBZrrk,        X86::VPMAXSBZrmk,          0 },
@@ -3376,6 +3384,8 @@ X86InstrInfo::X86InstrInfo(X86Subtarget &STI)
     { X86::VPERMT2Q256rrk,     X86::VPERMT2Q256rmk,       0 },
     { X86::VPERMT2W256rrk,     X86::VPERMT2W256rmk,       0 },
     { X86::VPERMWZ256rrk,      X86::VPERMWZ256rmk,        0 },
+    { X86::VPMADD52HUQZ256rk,  X86::VPMADD52HUQZ256mk,    0 },
+    { X86::VPMADD52LUQZ256rk,  X86::VPMADD52LUQZ256mk,    0 },
     { X86::VPMADDUBSWZ256rrk,  X86::VPMADDUBSWZ256rmk,    0 },
     { X86::VPMADDWDZ256rrk,    X86::VPMADDWDZ256rmk,      0 },
     { X86::VPMAXSBZ256rrk,     X86::VPMAXSBZ256rmk,       0 },
@@ -3509,6 +3519,8 @@ X86InstrInfo::X86InstrInfo(X86Subtarget &STI)
     { X86::VPERMT2Q128rrk,     X86::VPERMT2Q128rmk,       0 },
     { X86::VPERMT2W128rrk,     X86::VPERMT2W128rmk,       0 },
     { X86::VPERMWZ128rrk,      X86::VPERMWZ128rmk,        0 },
+    { X86::VPMADD52HUQZ128rk,  X86::VPMADD52HUQZ128mk,    0 },
+    { X86::VPMADD52LUQZ128rk,  X86::VPMADD52LUQZ128mk,    0 },
     { X86::VPMADDUBSWZ128rrk,  X86::VPMADDUBSWZ128rmk,    0 },
     { X86::VPMADDWDZ128rrk,    X86::VPMADDWDZ128rmk,      0 },
     { X86::VPMAXSBZ128rrk,     X86::VPMAXSBZ128rmk,       0 },
@@ -3597,6 +3609,8 @@ X86InstrInfo::X86InstrInfo(X86Subtarget &STI)
     { X86::VPERMT2PDrrkz,      X86::VPERMT2PDrmkz,        0 },
     { X86::VPERMT2Qrrkz,       X86::VPERMT2Qrmkz,         0 },
     { X86::VPERMT2Wrrkz,       X86::VPERMT2Wrmkz,         0 },
+    { X86::VPMADD52HUQZrkz,    X86::VPMADD52HUQZmkz,      0 },
+    { X86::VPMADD52LUQZrkz,    X86::VPMADD52LUQZmkz,      0 },
     { X86::VPTERNLOGDZrrikz,   X86::VPTERNLOGDZrmikz,     0 },
     { X86::VPTERNLOGQZrrikz,   X86::VPTERNLOGQZrmikz,     0 },
 
@@ -3613,6 +3627,8 @@ X86InstrInfo::X86InstrInfo(X86Subtarget &STI)
     { X86::VPERMT2PS256rrkz,   X86::VPERMT2PS256rmkz,     0 },
     { X86::VPERMT2Q256rrkz,    X86::VPERMT2Q256rmkz,      0 },
     { X86::VPERMT2W256rrkz,    X86::VPERMT2W256rmkz,      0 },
+    { X86::VPMADD52HUQZ256rkz, X86::VPMADD52HUQZ256mkz,   0 },
+    { X86::VPMADD52LUQZ256rkz, X86::VPMADD52LUQZ256mkz,   0 },
     { X86::VPTERNLOGDZ256rrikz,X86::VPTERNLOGDZ256rmikz,  0 },
     { X86::VPTERNLOGQZ256rrikz,X86::VPTERNLOGQZ256rmikz,  0 },
 
@@ -3629,6 +3645,8 @@ X86InstrInfo::X86InstrInfo(X86Subtarget &STI)
     { X86::VPERMT2PS128rrkz,   X86::VPERMT2PS128rmkz,     0 },
     { X86::VPERMT2Q128rrkz,    X86::VPERMT2Q128rmkz,      0 },
     { X86::VPERMT2W128rrkz,    X86::VPERMT2W128rmkz,      0 },
+    { X86::VPMADD52HUQZ128rkz, X86::VPMADD52HUQZ128mkz,   0 },
+    { X86::VPMADD52LUQZ128rkz, X86::VPMADD52LUQZ128mkz,   0 },
     { X86::VPTERNLOGDZ128rrikz,X86::VPTERNLOGDZ128rmikz,  0 },
     { X86::VPTERNLOGQZ128rrikz,X86::VPTERNLOGQZ128rmikz,  0 },
   };
@@ -5631,6 +5649,41 @@ bool X86InstrInfo::findCommutedOpIndices(MachineInstr &MI, unsigned &SrcOpIdx1,
   case X86::VPTERNLOGQZ256rmbikz:
   case X86::VPTERNLOGQZrmbikz:
     return findThreeSrcCommutedOpIndices(MI, SrcOpIdx1, SrcOpIdx2);
+  case X86::VPMADD52HUQZ128r:
+  case X86::VPMADD52HUQZ128rk:
+  case X86::VPMADD52HUQZ128rkz:
+  case X86::VPMADD52HUQZ256r:
+  case X86::VPMADD52HUQZ256rk:
+  case X86::VPMADD52HUQZ256rkz:
+  case X86::VPMADD52HUQZr:
+  case X86::VPMADD52HUQZrk:
+  case X86::VPMADD52HUQZrkz:
+  case X86::VPMADD52LUQZ128r:
+  case X86::VPMADD52LUQZ128rk:
+  case X86::VPMADD52LUQZ128rkz:
+  case X86::VPMADD52LUQZ256r:
+  case X86::VPMADD52LUQZ256rk:
+  case X86::VPMADD52LUQZ256rkz:
+  case X86::VPMADD52LUQZr:
+  case X86::VPMADD52LUQZrk:
+  case X86::VPMADD52LUQZrkz: {
+    unsigned CommutableOpIdx1 = 2;
+    unsigned CommutableOpIdx2 = 3;
+    if (Desc.TSFlags & X86II::EVEX_K) {
+      // Skip the mask register.
+      ++CommutableOpIdx1;
+      ++CommutableOpIdx2;
+    }
+    if (!fixCommutedOpIndices(SrcOpIdx1, SrcOpIdx2,
+                              CommutableOpIdx1, CommutableOpIdx2))
+      return false;
+    if (!MI.getOperand(SrcOpIdx1).isReg() ||
+        !MI.getOperand(SrcOpIdx2).isReg())
+      // No idea.
+      return false;
+    return true;
+  }
+
   default:
     const X86InstrFMA3Group *FMA3Group =
         X86InstrFMA3Info::getFMA3Group(MI.getOpcode());

diff --git a/test/CodeGen/X86/avx512ifma-intrinsics.ll b/test/CodeGen/X86/avx512ifma-intrinsics.ll
index 2df2d528dca1d109a896034953f8bd9297045ba1..3a27d9dbf00ba7b288ce367d3939c1ee45ecd33e 100644 (file)
--- a/test/CodeGen/X86/avx512ifma-intrinsics.ll
+++ b/test/CodeGen/X86/avx512ifma-intrinsics.ll
@@ -112,3 +112,73 @@ define <8 x i64>@test_int_x86_avx512_maskz_vpmadd52l_uq_512(<8 x i64> %x0, <8 x
   %res6 = add <8 x i64> %res5, %res4
   ret <8 x i64> %res6
 }
+
+define <8 x i64>@test_int_x86_avx512_vpmadd52h_uq_512_load(<8 x i64> %x0, <8 x i64> %x1, <8 x i64>* %x2ptr) {
+; CHECK-LABEL: test_int_x86_avx512_vpmadd52h_uq_512_load:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vpmadd52huq (%rdi), %zmm1, %zmm0
+; CHECK-NEXT:    retq
+
+  %x2 = load <8 x i64>, <8 x i64>* %x2ptr
+  %res = call <8 x i64> @llvm.x86.avx512.mask.vpmadd52h.uq.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 -1)
+  ret <8 x i64> %res
+}
+
+define <8 x i64>@test_int_x86_avx512_vpmadd52h_uq_512_load_commute(<8 x i64> %x0, <8 x i64>* %x1ptr, <8 x i64> %x2) {
+; CHECK-LABEL: test_int_x86_avx512_vpmadd52h_uq_512_load_commute:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vpmadd52huq (%rdi), %zmm1, %zmm0
+; CHECK-NEXT:    retq
+
+  %x1 = load <8 x i64>, <8 x i64>* %x1ptr
+  %res = call <8 x i64> @llvm.x86.avx512.mask.vpmadd52h.uq.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 -1)
+  ret <8 x i64> %res
+}
+
+define <8 x i64>@test_int_x86_avx512_mask_vpmadd52h_uq_512_load(<8 x i64> %x0, <8 x i64> %x1, <8 x i64>* %x2ptr, i8 %x3) {
+; CHECK-LABEL: test_int_x86_avx512_mask_vpmadd52h_uq_512_load:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %esi, %k1
+; CHECK-NEXT:    vpmadd52huq (%rdi), %zmm1, %zmm0 {%k1}
+; CHECK-NEXT:    retq
+
+  %x2 = load <8 x i64>, <8 x i64>* %x2ptr
+  %res = call <8 x i64> @llvm.x86.avx512.mask.vpmadd52h.uq.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3)
+  ret <8 x i64> %res
+}
+
+define <8 x i64>@test_int_x86_avx512_mask_vpmadd52h_uq_512_load_commute(<8 x i64> %x0, <8 x i64>* %x1ptr, <8 x i64> %x2, i8 %x3) {
+; CHECK-LABEL: test_int_x86_avx512_mask_vpmadd52h_uq_512_load_commute:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %esi, %k1
+; CHECK-NEXT:    vpmadd52huq (%rdi), %zmm1, %zmm0 {%k1}
+; CHECK-NEXT:    retq
+
+  %x1 = load <8 x i64>, <8 x i64>* %x1ptr
+  %res = call <8 x i64> @llvm.x86.avx512.mask.vpmadd52h.uq.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3)
+  ret <8 x i64> %res
+}
+
+define <8 x i64>@test_int_x86_avx512_maskz_vpmadd52h_uq_512_load(<8 x i64> %x0, <8 x i64> %x1, <8 x i64>* %x2ptr, i8 %x3) {
+; CHECK-LABEL: test_int_x86_avx512_maskz_vpmadd52h_uq_512_load:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %esi, %k1
+; CHECK-NEXT:    vpmadd52huq (%rdi), %zmm1, %zmm0 {%k1} {z}
+; CHECK-NEXT:    retq
+
+  %x2 = load <8 x i64>, <8 x i64>* %x2ptr
+  %res = call <8 x i64> @llvm.x86.avx512.maskz.vpmadd52h.uq.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3)
+  ret <8 x i64> %res
+}
+
+define <8 x i64>@test_int_x86_avx512_maskz_vpmadd52h_uq_512_load_commute(<8 x i64> %x0, <8 x i64>* %x1ptr, <8 x i64> %x2, i8 %x3) {
+; CHECK-LABEL: test_int_x86_avx512_maskz_vpmadd52h_uq_512_load_commute:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %esi, %k1
+; CHECK-NEXT:    vpmadd52huq (%rdi), %zmm1, %zmm0 {%k1} {z}
+; CHECK-NEXT:    retq
+
+  %x1 = load <8 x i64>, <8 x i64>* %x1ptr
+  %res = call <8 x i64> @llvm.x86.avx512.maskz.vpmadd52h.uq.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3)
+  ret <8 x i64> %res
+}