AVX512: Implemented encoding and intrinsics for VGETEXPSS/D instructions

Added tests for intrinsics and encoding.

Differential Revision: http://reviews.llvm.org/D11528

llvm-svn: 243390

6 files changed, 150 insertions, 1 deletions

diff --git a/llvm/include/llvm/IR/IntrinsicsX86.td b/llvm/include/llvm/IR/IntrinsicsX86.td
index b822f47ce03..0d60aefed15 100644
--- a/llvm/include/llvm/IR/IntrinsicsX86.td
+++ b/llvm/include/llvm/IR/IntrinsicsX86.td
@@ -4860,6 +4860,13 @@ let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
         Intrinsic<[llvm_v16f32_ty], [llvm_v16f32_ty, llvm_v16f32_ty,
                                      llvm_i16_ty, llvm_i32_ty], [IntrNoMem]>;
 
+  def int_x86_avx512_mask_getexp_ss : GCCBuiltin<"__builtin_ia32_getexpss_mask">,
+        Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty,
+                                    llvm_i8_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_getexp_sd : GCCBuiltin<"__builtin_ia32_getexpsd_mask">,
+        Intrinsic<[llvm_v2f64_ty], [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty,
+                                    llvm_i8_ty, llvm_i32_ty], [IntrNoMem]>;
+
   def int_x86_avx512_rsqrt14_ss : GCCBuiltin<"__builtin_ia32_rsqrt14ss_mask">,
         Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty,
                                     llvm_i8_ty], [IntrNoMem]>;
diff --git a/llvm/lib/Target/X86/X86InstrAVX512.td b/llvm/lib/Target/X86/X86InstrAVX512.td
index bb60cbc5238..ef33993f6f6 100644
--- a/llvm/lib/Target/X86/X86InstrAVX512.td
+++ b/llvm/lib/Target/X86/X86InstrAVX512.td
@@ -5210,6 +5210,8 @@ let hasSideEffects = 0, Predicates = [HasERI] in {
   defm VRCP28   : avx512_eri_s<0xCB, "vrcp28",   X86rcp28s>,   T8PD, EVEX_4V;
   defm VRSQRT28 : avx512_eri_s<0xCD, "vrsqrt28", X86rsqrt28s>, T8PD, EVEX_4V;
 }
+
+defm VGETEXP   : avx512_eri_s<0x43, "vgetexp", X86fgetexpRnds>, T8PD, EVEX_4V;
 /// avx512_fp28_p rcp28ps, rcp28pd, rsqrt28ps, rsqrt28pd
 
 multiclass avx512_fp28_p<bits<8> opc, string OpcodeStr, X86VectorVTInfo _,
diff --git a/llvm/lib/Target/X86/X86InstrFragmentsSIMD.td b/llvm/lib/Target/X86/X86InstrFragmentsSIMD.td
index a41ba518cd4..361e1966ae7 100644
--- a/llvm/lib/Target/X86/X86InstrFragmentsSIMD.td
+++ b/llvm/lib/Target/X86/X86InstrFragmentsSIMD.td
@@ -326,7 +326,8 @@ def X86fmaxRnd   : SDNode<"X86ISD::FMAX_RND",      SDTFPBinOpRound>;
 def X86scalef    : SDNode<"X86ISD::SCALEF",    SDTFPBinOpRound>;
 def X86fminRnd   : SDNode<"X86ISD::FMIN_RND",      SDTFPBinOpRound>;
 def X86fsqrtRnd     : SDNode<"X86ISD::FSQRT_RND",  SDTFPUnaryOpRound>;
-def X86fgetexpRnd   : SDNode<"X86ISD::FGETEXP_RND",  SDTFPUnaryOpRound>;
+def X86fgetexpRnd   : SDNode<"X86ISD::FGETEXP_RND", SDTFPUnaryOpRound>;
+def X86fgetexpRnds  : SDNode<"X86ISD::FGETEXP_RND", STDFp2SrcRm>;
 
 def X86Fmadd     : SDNode<"X86ISD::FMADD",     SDTFma>;
 def X86Fnmadd    : SDNode<"X86ISD::FNMADD",    SDTFma>;
diff --git a/llvm/lib/Target/X86/X86IntrinsicsInfo.h b/llvm/lib/Target/X86/X86IntrinsicsInfo.h
index 407ae65f320..c065f3d86c5 100644
--- a/llvm/lib/Target/X86/X86IntrinsicsInfo.h
+++ b/llvm/lib/Target/X86/X86IntrinsicsInfo.h
@@ -659,6 +659,10 @@ static const IntrinsicData  IntrinsicsWithoutChain[] = {
                      X86ISD::FGETEXP_RND, 0),
   X86_INTRINSIC_DATA(avx512_mask_getexp_ps_512, INTR_TYPE_1OP_MASK_RM,
                      X86ISD::FGETEXP_RND, 0),
+  X86_INTRINSIC_DATA(avx512_mask_getexp_sd, INTR_TYPE_SCALAR_MASK_RM,
+                     X86ISD::FGETEXP_RND, 0),
+  X86_INTRINSIC_DATA(avx512_mask_getexp_ss, INTR_TYPE_SCALAR_MASK_RM,
+                    X86ISD::FGETEXP_RND, 0),
   X86_INTRINSIC_DATA(avx512_mask_max_pd_128, INTR_TYPE_2OP_MASK, X86ISD::FMAX, 0),
   X86_INTRINSIC_DATA(avx512_mask_max_pd_256, INTR_TYPE_2OP_MASK, X86ISD::FMAX, 0),
   X86_INTRINSIC_DATA(avx512_mask_max_pd_512, INTR_TYPE_2OP_MASK, X86ISD::FMAX,
diff --git a/llvm/test/CodeGen/X86/avx512-intrinsics.ll b/llvm/test/CodeGen/X86/avx512-intrinsics.ll
index 6046953cd9b..9a494ea1c3b 100644
--- a/llvm/test/CodeGen/X86/avx512-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512-intrinsics.ll
@@ -3896,3 +3896,58 @@ define <2 x double>@test_int_x86_avx512_mask_scalef_sd(<2 x double> %x0, <2 x do
   %res2 = fadd <2 x double> %res, %res1
   ret <2 x double> %res2
 }
+
+declare <4 x float> @llvm.x86.avx512.mask.getexp.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32) nounwind readnone
+
+define <4 x float> @test_getexp_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) {
+; CHECK-LABEL: test_getexp_ss:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    andl $1, %edi
+; CHECK-NEXT:    kmovw %edi, %k1
+; CHECK-NEXT:    vmovaps %zmm2, %zmm3
+; CHECK-NEXT:    vgetexpss %xmm1, %xmm0, %xmm3 {%k1}
+; CHECK-NEXT:    vgetexpss {sae}, %xmm1, %xmm0, %xmm2 {%k1}
+; CHECK-NEXT:    vgetexpss {sae}, %xmm1, %xmm0, %xmm4 {%k1} {z}
+; CHECK-NEXT:    vgetexpss {sae}, %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vaddps %xmm2, %xmm3, %xmm1
+; CHECK-NEXT:    vaddps %xmm0, %xmm4, %xmm0
+; CHECK-NEXT:    vaddps %xmm0, %xmm1, %xmm0
+; CHECK-NEXT:    retq
+  %res0 = call <4 x float> @llvm.x86.avx512.mask.getexp.ss(<4 x float>%a0, <4 x float> %a1, <4 x float> %a2, i8 %mask, i32 4)
+  %res1 = call <4 x float> @llvm.x86.avx512.mask.getexp.ss(<4 x float>%a0, <4 x float> %a1, <4 x float> %a2, i8 %mask, i32 8)
+  %res2 = call <4 x float> @llvm.x86.avx512.mask.getexp.ss(<4 x float>%a0, <4 x float> %a1, <4 x float> zeroinitializer, i8 %mask, i32 8)
+  %res3 = call <4 x float> @llvm.x86.avx512.mask.getexp.ss(<4 x float>%a0, <4 x float> %a1, <4 x float> zeroinitializer, i8 -1, i32 8)
+
+  %res.1 = fadd <4 x float> %res0, %res1
+  %res.2 = fadd <4 x float> %res2, %res3
+  %res   = fadd <4 x float> %res.1, %res.2
+  ret <4 x float> %res
+}
+
+declare <2 x double> @llvm.x86.avx512.mask.getexp.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32) nounwind readnone
+
+define <2 x double> @test_getexp_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) {
+; CHECK-LABEL: test_getexp_sd:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    andl $1, %edi
+; CHECK-NEXT:    kmovw %edi, %k1
+; CHECK-NEXT:    vmovaps %zmm2, %zmm3
+; CHECK-NEXT:    vgetexpsd %xmm1, %xmm0, %xmm3 {%k1}
+; CHECK-NEXT:    vgetexpsd %xmm1, %xmm0, %xmm4
+; CHECK-NEXT:    vgetexpsd {sae}, %xmm1, %xmm0, %xmm2 {%k1}
+; CHECK-NEXT:    vgetexpsd {sae}, %xmm1, %xmm0, %xmm0 {%k1} {z}
+; CHECK-NEXT:    vaddpd %xmm2, %xmm3, %xmm1
+; CHECK-NEXT:    vaddpd %xmm4, %xmm0, %xmm0
+; CHECK-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
+; CHECK-NEXT:    retq
+  %res0 = call <2 x double> @llvm.x86.avx512.mask.getexp.sd(<2 x double>%a0, <2 x double> %a1, <2 x double> %a2, i8 %mask, i32 4)
+  %res1 = call <2 x double> @llvm.x86.avx512.mask.getexp.sd(<2 x double>%a0, <2 x double> %a1, <2 x double> %a2, i8 %mask, i32 8)
+  %res2 = call <2 x double> @llvm.x86.avx512.mask.getexp.sd(<2 x double>%a0, <2 x double> %a1, <2 x double> zeroinitializer, i8 %mask, i32 8)
+  %res3 = call <2 x double> @llvm.x86.avx512.mask.getexp.sd(<2 x double>%a0, <2 x double> %a1, <2 x double> zeroinitializer, i8 -1, i32 4)
+
+  %res.1 = fadd <2 x double> %res0, %res1
+  %res.2 = fadd <2 x double> %res2, %res3
+  %res   = fadd <2 x double> %res.1, %res.2
+  ret <2 x double> %res
+}
+
diff --git a/llvm/test/MC/X86/avx512-encodings.s b/llvm/test/MC/X86/avx512-encodings.s
index 113c6baeac1..7dd8014213d 100644
--- a/llvm/test/MC/X86/avx512-encodings.s
+++ b/llvm/test/MC/X86/avx512-encodings.s
@@ -14878,3 +14878,83 @@ vpermilpd $0x23, 0x400(%rbx), %zmm2
 // CHECK:  encoding: [0x62,0x61,0x85,0x58,0x6d,0x9a,0xf8,0xfb,0xff,0xff]
           vpunpckhqdq -1032(%rdx){1to8}, %zmm15, %zmm27
 
+// CHECK: vgetexpss %xmm26, %xmm1, %xmm20
+// CHECK:  encoding: [0x62,0x82,0x75,0x08,0x43,0xe2]
+          vgetexpss %xmm26, %xmm1, %xmm20
+
+// CHECK: vgetexpss %xmm26, %xmm1, %xmm20 {%k7}
+// CHECK:  encoding: [0x62,0x82,0x75,0x0f,0x43,0xe2]
+          vgetexpss %xmm26, %xmm1, %xmm20 {%k7}
+
+// CHECK: vgetexpss %xmm26, %xmm1, %xmm20 {%k7} {z}
+// CHECK:  encoding: [0x62,0x82,0x75,0x8f,0x43,0xe2]
+          vgetexpss %xmm26, %xmm1, %xmm20 {%k7} {z}
+
+// CHECK: vgetexpss {sae}, %xmm26, %xmm1, %xmm20
+// CHECK:  encoding: [0x62,0x82,0x75,0x18,0x43,0xe2]
+          vgetexpss {sae}, %xmm26, %xmm1, %xmm20
+
+// CHECK: vgetexpss (%rcx), %xmm1, %xmm20
+// CHECK:  encoding: [0x62,0xe2,0x75,0x08,0x43,0x21]
+          vgetexpss (%rcx), %xmm1, %xmm20
+
+// CHECK: vgetexpss 291(%rax,%r14,8), %xmm1, %xmm20
+// CHECK:  encoding: [0x62,0xa2,0x75,0x08,0x43,0xa4,0xf0,0x23,0x01,0x00,0x00]
+          vgetexpss 291(%rax,%r14,8), %xmm1, %xmm20
+
+// CHECK: vgetexpss 508(%rdx), %xmm1, %xmm20
+// CHECK:  encoding: [0x62,0xe2,0x75,0x08,0x43,0x62,0x7f]
+          vgetexpss 508(%rdx), %xmm1, %xmm20
+
+// CHECK: vgetexpss 512(%rdx), %xmm1, %xmm20
+// CHECK:  encoding: [0x62,0xe2,0x75,0x08,0x43,0xa2,0x00,0x02,0x00,0x00]
+          vgetexpss 512(%rdx), %xmm1, %xmm20
+
+// CHECK: vgetexpss -512(%rdx), %xmm1, %xmm20
+// CHECK:  encoding: [0x62,0xe2,0x75,0x08,0x43,0x62,0x80]
+          vgetexpss -512(%rdx), %xmm1, %xmm20
+
+// CHECK: vgetexpss -516(%rdx), %xmm1, %xmm20
+// CHECK:  encoding: [0x62,0xe2,0x75,0x08,0x43,0xa2,0xfc,0xfd,0xff,0xff]
+          vgetexpss -516(%rdx), %xmm1, %xmm20
+
+// CHECK: vgetexpsd %xmm2, %xmm7, %xmm2
+// CHECK:  encoding: [0x62,0xf2,0xc5,0x08,0x43,0xd2]
+          vgetexpsd %xmm2, %xmm7, %xmm2
+
+// CHECK: vgetexpsd %xmm2, %xmm7, %xmm2 {%k5}
+// CHECK:  encoding: [0x62,0xf2,0xc5,0x0d,0x43,0xd2]
+          vgetexpsd %xmm2, %xmm7, %xmm2 {%k5}
+
+// CHECK: vgetexpsd %xmm2, %xmm7, %xmm2 {%k5} {z}
+// CHECK:  encoding: [0x62,0xf2,0xc5,0x8d,0x43,0xd2]
+          vgetexpsd %xmm2, %xmm7, %xmm2 {%k5} {z}
+
+// CHECK: vgetexpsd {sae}, %xmm2, %xmm7, %xmm2
+// CHECK:  encoding: [0x62,0xf2,0xc5,0x18,0x43,0xd2]
+          vgetexpsd {sae}, %xmm2, %xmm7, %xmm2
+
+// CHECK: vgetexpsd (%rcx), %xmm7, %xmm2
+// CHECK:  encoding: [0x62,0xf2,0xc5,0x08,0x43,0x11]
+          vgetexpsd (%rcx), %xmm7, %xmm2
+
+// CHECK: vgetexpsd 291(%rax,%r14,8), %xmm7, %xmm2
+// CHECK:  encoding: [0x62,0xb2,0xc5,0x08,0x43,0x94,0xf0,0x23,0x01,0x00,0x00]
+          vgetexpsd 291(%rax,%r14,8), %xmm7, %xmm2
+
+// CHECK: vgetexpsd 1016(%rdx), %xmm7, %xmm2
+// CHECK:  encoding: [0x62,0xf2,0xc5,0x08,0x43,0x52,0x7f]
+          vgetexpsd 1016(%rdx), %xmm7, %xmm2
+
+// CHECK: vgetexpsd 1024(%rdx), %xmm7, %xmm2
+// CHECK:  encoding: [0x62,0xf2,0xc5,0x08,0x43,0x92,0x00,0x04,0x00,0x00]
+          vgetexpsd 1024(%rdx), %xmm7, %xmm2
+
+// CHECK: vgetexpsd -1024(%rdx), %xmm7, %xmm2
+// CHECK:  encoding: [0x62,0xf2,0xc5,0x08,0x43,0x52,0x80]
+          vgetexpsd -1024(%rdx), %xmm7, %xmm2
+
+// CHECK: vgetexpsd -1032(%rdx), %xmm7, %xmm2
+// CHECK:  encoding: [0x62,0xf2,0xc5,0x08,0x43,0x92,0xf8,0xfb,0xff,0xff]
+          vgetexpsd -1032(%rdx), %xmm7, %xmm2
+