[PowerPC] Implement builtin for vbpermd

[nemanjai]

Summary

The instruction has similar semantics to vbpermq but for doublewords. It was added in P9 and the ABI documents the builtin.

[bmahjour]

clang/lib/Headers/altivec.h

17337

This should be guarded under P8. It would also be good to add a vec_vbpermd(vector unsigned long long ...) counter part under __POWER9_VECTOR__ for consistency.

17349

vbpermq variants should be guarded under __POWER8_VECTOR__

[lei]

LGTM

clang/lib/Headers/altivec.h

17337

I think this is actually guarded under __POWER8_VECTOR__. See line 17237.
As far as I can see, there is no LLVM intrinsic corresponding to vbpermd.

17352

nit: It would be more clear if we had a comment here what this endif if for. I assume it's /* __POWER8_VECTOR__ */

[bmahjour]

clang/lib/Headers/altivec.h

17337

I think this is actually guarded under POWER8_VECTOR. See line 17237.

You are right. I missed that.

As far as I can see, there is no LLVM intrinsic corresponding to vbpermd.

__builtin_altivec_vbpermd is being called on line 17356...not sure if it corresponds to an IR intrinsic or not, but regardless we have vec_vbpermq in this header but not vec_vbpermd. That seems inconsistent to me.

[bmahjour]

clang/lib/Headers/altivec.h

17349

never mind, looks like it already is guarded under P8.

clang/include/clang/Basic/BuiltinsPPC.def

 BUILTIN(__builtin_altivec_vcmpgtsw_p, "iiV4SiV4Si", "")
 BUILTIN(__builtin_altivec_vcmpgtuw_p, "iiV4UiV4Ui", "")
 BUILTIN(__builtin_altivec_vcmpgtsd_p, "iiV2LLiV2LLi", "")
 BUILTIN(__builtin_altivec_vcmpgtud_p, "iiV2ULLiV2ULLi", "")
 BUILTIN(__builtin_altivec_vcmpgtfp_p, "iiV4fV4f", "")
 
 BUILTIN(__builtin_altivec_vgbbd, "V16UcV16Uc", "")
 BUILTIN(__builtin_altivec_vbpermq, "V2ULLiV16UcV16Uc", "")
+BUILTIN(__builtin_altivec_vbpermd, "V2ULLiV2ULLiV16Uc", "")
 
 // P8 Crypto built-ins.
 BUILTIN(__builtin_altivec_crypto_vsbox, "V2ULLiV2ULLi", "")
 BUILTIN(__builtin_altivec_crypto_vpermxor, "V16UcV16UcV16UcV16Uc", "")
 BUILTIN(__builtin_altivec_crypto_vshasigmaw, "V4UiV4UiIiIi", "")
 BUILTIN(__builtin_altivec_crypto_vshasigmad, "V2ULLiV2ULLiIiIi", "")
 BUILTIN(__builtin_altivec_crypto_vcipher, "V2ULLiV2ULLiV2ULLi", "")
 BUILTIN(__builtin_altivec_crypto_vcipherlast, "V2ULLiV2ULLiV2ULLi", "")

clang/lib/Headers/altivec.h

 /*===---- altivec.h - Standard header for type generic math ---------------===*\
  *
  * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
  * See https://llvm.org/LICENSE.txt for license information.
  * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
  *
 \*===----------------------------------------------------------------------===*/
 
 #ifndef __ALTIVEC_H
 #define __ALTIVEC_H
 
 #ifndef __ALTIVEC__
 #error "AltiVec support not enabled"
 #endif
 
 /* Constants for mapping CR6 bits to predicate result. */
 
 #define __CR6_EQ 0
 #define __CR6_EQ_REV 1
 #define __CR6_LT 2
 #define __CR6_LT_REV 3
#define __VEC_CLASS_FP_NOT_NORMAL (__VEC_CLASS_FP_NAN        | \
                                    __VEC_CLASS_FP_SUBNORMAL  | \
                                    __VEC_CLASS_FP_ZERO       | \
                                    __VEC_CLASS_FP_INFINITY)
 
 #define __ATTRS_o_ai __attribute__((__overloadable__, __always_inline__))
 
 #include <stddef.h>
 
 static __inline__ vector signed char __ATTRS_o_ai vec_perm(
     vector signed char __a, vector signed char __b, vector unsigned char __c);
 
 static __inline__ vector unsigned char __ATTRS_o_ai
 vec_perm(vector unsigned char __a, vector unsigned char __b,
          vector unsigned char __c);
 
 static __inline__ vector bool char __ATTRS_o_ai
 vec_perm(vector bool char __a, vector bool char __b, vector unsigned char __c);
 
 static __inline__ vector short __ATTRS_o_ai vec_perm(vector signed short __a,
                                                      vector signed short __b,
                                                      vector unsigned char __c);
 
 static __inline__ vector unsigned short __ATTRS_o_ai
 vec_perm(vector unsigned short __a, vector unsigned short __b,
          vector unsigned char __c);
 
 static __inline__ vector bool short __ATTRS_o_ai vec_perm(
     vector bool short __a, vector bool short __b, vector unsigned char __c);
 
 static __inline__ vector pixel __ATTRS_o_ai vec_perm(vector pixel __a,
                                                      vector pixel __b,
                                                      vector unsigned char __c);

 
 static __inline__ vector long long __ATTRS_o_ai
 vec_vbpermq(vector signed char __a, vector signed char __b) {
   return __builtin_altivec_vbpermq((vector unsigned char)__a,
                                    (vector unsigned char)__b);
 }
 
 static __inline__ vector long long __ATTRS_o_ai
 vec_vbpermq(vector unsigned char __a, vector unsigned char __b) {
   return __builtin_altivec_vbpermq(__a, __b);
 }
 
 #if defined(__powerpc64__) && defined(__SIZEOF_INT128__)
-static __inline__ vector unsigned long long __attribute__((__always_inline__))
+static __inline__ vector unsigned long long __ATTRS_o_ai
 vec_bperm(vector unsigned __int128 __a, vector unsigned char __b) {
   return __builtin_altivec_vbpermq((vector unsigned char)__a,
                                    (vector unsigned char)__b);
 }
 #endif
+static __inline__ vector unsigned char __ATTRS_o_ai
+vec_bperm(vector unsigned char __a, vector unsigned char __b) {
+  return __builtin_altivec_vbpermq(__a, __b);
+}
+#endif
+#ifdef __POWER9_VECTOR__
+static __inline__ vector unsigned long long __ATTRS_o_ai
+vec_bperm(vector unsigned long long __a, vector unsigned char __b) {
+  return __builtin_altivec_vbpermd(__a, __b);
+}
 #endif
 
 
 /* vec_reve */
 
 static inline __ATTRS_o_ai vector bool char vec_reve(vector bool char __a) {
   return __builtin_shufflevector(__a, __a, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6,
                                  5, 4, 3, 2, 1, 0);

clang/test/CodeGen/builtins-ppc-p8vector.c

// CHECK-PPC: warning: implicit declaration of function 'vec_gb'
   res_vsll = vec_gbb(vsll);
 // CHECK: llvm.ppc.altivec.vgbbd
 // CHECK-LE: llvm.ppc.altivec.vgbbd
 
   res_vull = vec_gbb(vull);
 // CHECK: llvm.ppc.altivec.vgbbd
 // CHECK-LE: llvm.ppc.altivec.vgbbd
 
-  res_vull = vec_bperm(vux, vux);
+  res_vull = vec_bperm(vux, vuc);
+  // CHECK: llvm.ppc.altivec.vbpermq
+  // CHECK-LE: llvm.ppc.altivec.vbpermq
+
+  res_vull = vec_bperm(vuc, vuc);
   // CHECK: llvm.ppc.altivec.vbpermq
   // CHECK-LE: llvm.ppc.altivec.vbpermq
-// CHECK-PPC: warning: implicit declaration of function 'vec_bperm'
 
   res_vsll = vec_neg(vsll);
 // CHECK: sub <2 x i64> zeroinitializer, {{%[0-9]+}}
 // CHECK-LE: sub <2 x i64> zeroinitializer, {{%[0-9]+}}
 // CHECK_PPC: call to 'vec_neg' is ambiguous
 
 
 }

clang/test/CodeGen/builtins-ppc-p9vector.c

vector signed long long test_vec_signextll_sll_ss(void) {
     return vec_signextll(vssa);
 }
 
 vector signed long long test_vec_signextll_sll_si(void) {
     // CHECK: @llvm.ppc.altivec.vextsw2d(<4 x i32>
     // CHECK-NEXT: ret <2 x i64>
     return vec_signextll(vsia);
 }
+
+vector unsigned long long test_vbpermd(void) {
+  // CHECK: @llvm.ppc.altivec.vbpermd(<2 x i64>
+  // CHECK-BE: @llvm.ppc.altivec.vbpermd(<2 x i64>
+  return vec_bperm(vula, vuca);
+}

llvm/include/llvm/IR/IntrinsicsPowerPC.td

let TargetPrefix = "ppc" in {  // All PPC intrinsics start with "llvm.ppc.".
   def int_ppc_altivec_vsel : GCCBuiltin<"__builtin_altivec_vsel_4si">,
               Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty,
                          llvm_v4i32_ty, llvm_v4i32_ty], [IntrNoMem]>;
   def int_ppc_altivec_vgbbd : GCCBuiltin<"__builtin_altivec_vgbbd">,
               Intrinsic<[llvm_v16i8_ty], [llvm_v16i8_ty], [IntrNoMem]>;
   def int_ppc_altivec_vbpermq : GCCBuiltin<"__builtin_altivec_vbpermq">,
               Intrinsic<[llvm_v2i64_ty], [llvm_v16i8_ty, llvm_v16i8_ty],
                         [IntrNoMem]>;
+  def int_ppc_altivec_vbpermd : GCCBuiltin<"__builtin_altivec_vbpermd">,
+              Intrinsic<[llvm_v2i64_ty], [llvm_v2i64_ty, llvm_v16i8_ty],
+                        [IntrNoMem]>;
 }
 
 def int_ppc_altivec_vexptefp  : PowerPC_Vec_FF_Intrinsic<"vexptefp">;
 def int_ppc_altivec_vlogefp   : PowerPC_Vec_FF_Intrinsic<"vlogefp">;
 def int_ppc_altivec_vrefp     : PowerPC_Vec_FF_Intrinsic<"vrefp">;
 def int_ppc_altivec_vrsqrtefp : PowerPC_Vec_FF_Intrinsic<"vrsqrtefp">;
 
 // Power8 Intrinsics

llvm/lib/Target/PowerPC/PPCInstrAltivec.td

 def VPRTYBW : VX_VT5_EO5_VB5<1538, 8, "vprtybw", [(set v4i32:$vD,
                             (int_ppc_altivec_vprtybw v4i32:$vB))]>;
 def VPRTYBD : VX_VT5_EO5_VB5<1538,  9, "vprtybd", [(set v2i64:$vD,
                             (int_ppc_altivec_vprtybd v2i64:$vB))]>;
 def VPRTYBQ : VX_VT5_EO5_VB5<1538, 10, "vprtybq", [(set v1i128:$vD,
                             (int_ppc_altivec_vprtybq v1i128:$vB))]>;
 
 // Vector (Bit) Permute (Right-indexed)
-def VBPERMD : VXForm_1<1484, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
-                       "vbpermd $vD, $vA, $vB", IIC_VecFP, []>;
+def VBPERMD : VX1_Int_Ty3<1484, "vbpermd", int_ppc_altivec_vbpermd,
+                          v2i64, v2i64, v16i8>;
 def VPERMR : VAForm_1a<59, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB, vrrc:$vC),
                        "vpermr $vD, $vA, $vB, $vC", IIC_VecFP, []>;
 
 class VX1_VT5_VA5_VB5<bits<11> xo, string opc, list<dag> pattern>
   : VXForm_1<xo, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
              !strconcat(opc, " $vD, $vA, $vB"), IIC_VecFP, pattern>;
 
 // Vector Rotate Left Mask/Mask-Insert

llvm/test/CodeGen/PowerPC/p9-vbpermd.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -verify-machineinstrs -mtriple=powerpc64-unknown-linux-gnu \
+; RUN:  -mcpu=pwr9 < %s | FileCheck %s
+; RUN: llc -verify-machineinstrs -mtriple=powerpc64le-unknown-linux-gnu \
+; RUN:  -mcpu=pwr9 < %s | FileCheck %s
+
+@vull = common global <2 x i64> zeroinitializer, align 16
+@vuc = common global <16 x i8> zeroinitializer, align 16
+@res_vull = common global <2 x i64> zeroinitializer, align 16
+
+define void @test1() {
+; CHECK-LABEL: test1:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    addis 3, 2, .LC0@toc@ha
+; CHECK-NEXT:    ld 3, .LC0@toc@l(3)
+; CHECK-NEXT:    lxv 34, 0(3)
+; CHECK-NEXT:    addis 3, 2, .LC1@toc@ha
+; CHECK-NEXT:    ld 3, .LC1@toc@l(3)
+; CHECK-NEXT:    lxv 35, 0(3)
+; CHECK-NEXT:    addis 3, 2, .LC2@toc@ha
+; CHECK-NEXT:    ld 3, .LC2@toc@l(3)
+; CHECK-NEXT:    vbpermd 2, 2, 3
+; CHECK-NEXT:    stxv 34, 0(3)
+; CHECK-NEXT:    blr
+entry:
+  %0 = load <2 x i64>, <2 x i64>* @vull, align 16
+  %1 = load <16 x i8>, <16 x i8>* @vuc, align 16
+  %2 = call <2 x i64> @llvm.ppc.altivec.vbpermd(<2 x i64> %0, <16 x i8> %1)
+  store <2 x i64> %2, <2 x i64>* @res_vull, align 16
+  ret void
+}
+declare <2 x i64> @llvm.ppc.altivec.vbpermd(<2 x i64>, <16 x i8>)