[PowerPC] Implement the 128-bit vec_[all|any]_[eq | ne | lt | gt | le | ge] builtins in Clang/LLVM

[amyk]

Summary

This patch implements the vec_[all|any]_[eq | ne | lt | gt | le | ge] builtins for vector signed/unsigned __int128.

Depends on D87804

[amyk]

Updated identation/formatting of code in CGExprScalar.cpp and altivec.h.

[Conanap]

lgtm

clang/include/clang/Basic/BuiltinsPPC.def

 BUILTIN(__builtin_altivec_vcmpgtsd, "V2LLiV2LLiV2LLi", "")
 BUILTIN(__builtin_altivec_vcmpgtud, "V2LLiV2ULLiV2ULLi", "")
 BUILTIN(__builtin_altivec_vcmpgtfp, "V4iV4fV4f", "")
 
 // P10 Vector compare builtins.
 BUILTIN(__builtin_altivec_vcmpequq, "V1LLLiV1ULLLiV1ULLLi", "")
 BUILTIN(__builtin_altivec_vcmpgtsq, "V1LLLiV1SLLLiV1SLLLi", "")
 BUILTIN(__builtin_altivec_vcmpgtuq, "V1LLLiV1ULLLiV1ULLLi", "")
+BUILTIN(__builtin_altivec_vcmpequq_p, "iiV1ULLLiV1LLLi", "")
+BUILTIN(__builtin_altivec_vcmpgtsq_p, "iiV1SLLLiV1SLLLi", "")
+BUILTIN(__builtin_altivec_vcmpgtuq_p, "iiV1ULLLiV1ULLLi", "")
 
 BUILTIN(__builtin_altivec_vmaxsb, "V16ScV16ScV16Sc", "")
 BUILTIN(__builtin_altivec_vmaxub, "V16UcV16UcV16Uc", "")
 BUILTIN(__builtin_altivec_vmaxsh, "V8SsV8SsV8Ss", "")
 BUILTIN(__builtin_altivec_vmaxuh, "V8UsV8UsV8Us", "")
 BUILTIN(__builtin_altivec_vmaxsw, "V4SiV4SiV4Si", "")
 BUILTIN(__builtin_altivec_vmaxuw, "V4UiV4UiV4Ui", "")
 BUILTIN(__builtin_altivec_vmaxsd, "V2LLiV2LLiV2LLi", "")

clang/lib/CodeGen/CGExprScalar.cpp

case BuiltinType::LongLong:
     return (IT == VCMPEQ) ? llvm::Intrinsic::ppc_altivec_vcmpequd_p :
                             llvm::Intrinsic::ppc_altivec_vcmpgtsd_p;
   case BuiltinType::Float:
     return (IT == VCMPEQ) ? llvm::Intrinsic::ppc_altivec_vcmpeqfp_p :
                             llvm::Intrinsic::ppc_altivec_vcmpgtfp_p;
   case BuiltinType::Double:
     return (IT == VCMPEQ) ? llvm::Intrinsic::ppc_vsx_xvcmpeqdp_p :
                             llvm::Intrinsic::ppc_vsx_xvcmpgtdp_p;
+  case BuiltinType::UInt128:
+    return (IT == VCMPEQ) ? llvm::Intrinsic::ppc_altivec_vcmpequq_p
+                          : llvm::Intrinsic::ppc_altivec_vcmpgtuq_p;
+  case BuiltinType::Int128:
+    return (IT == VCMPEQ) ? llvm::Intrinsic::ppc_altivec_vcmpequq_p
+                          : llvm::Intrinsic::ppc_altivec_vcmpgtsq_p;
   }
 }
 
 Value *ScalarExprEmitter::EmitCompare(const BinaryOperator *E,
                                       llvm::CmpInst::Predicate UICmpOpc,
                                       llvm::CmpInst::Predicate SICmpOpc,
                                       llvm::CmpInst::Predicate FCmpOpc,
                                       bool IsSignaling) {

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_INFINITY)
 
 #define __ATTRS_o_ai __attribute__((__overloadable__, __always_inline__))
 
 #ifdef __POWER9_VECTOR__
 #include <stddef.h>
 #endif
 
 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);

 
 #ifdef __VSX__
 static __inline__ int __ATTRS_o_ai vec_all_eq(vector double __a,
                                               vector double __b) {
   return __builtin_vsx_xvcmpeqdp_p(__CR6_LT, __a, __b);
 }
 #endif
 
+#ifdef __POWER10_VECTOR__
+static __inline__ int __ATTRS_o_ai vec_all_eq(vector signed __int128 __a,
+                                              vector signed __int128 __b) {
+  return __builtin_altivec_vcmpequq_p(__CR6_LT, __a, __b);
+}
+
+static __inline__ int __ATTRS_o_ai vec_all_eq(vector unsigned __int128 __a,
+                                              vector unsigned __int128 __b) {
+  return __builtin_altivec_vcmpequq_p(__CR6_LT, __a, __b);
+}
+#endif
+
 /* vec_all_ge */
 
 static __inline__ int __ATTRS_o_ai vec_all_ge(vector signed char __a,
                                               vector signed char __b) {
   return __builtin_altivec_vcmpgtsb_p(__CR6_EQ, __b, __a);
 }
 
 static __inline__ int __ATTRS_o_ai vec_all_ge(vector signed char __a,

 
 #ifdef __VSX__
 static __inline__ int __ATTRS_o_ai vec_all_ge(vector double __a,
                                               vector double __b) {
   return __builtin_vsx_xvcmpgedp_p(__CR6_LT, __a, __b);
 }
 #endif
 
+#ifdef __POWER10_VECTOR__
+static __inline__ int __ATTRS_o_ai vec_all_ge(vector signed __int128 __a,
+                                              vector signed __int128 __b) {
+  return __builtin_altivec_vcmpgtsq_p(__CR6_EQ, __b, __a);
+}
+
+static __inline__ int __ATTRS_o_ai vec_all_ge(vector unsigned __int128 __a,
+                                              vector unsigned __int128 __b) {
+  return __builtin_altivec_vcmpgtuq_p(__CR6_EQ, __b, __a);
+}
+#endif
+
 /* vec_all_gt */
 
 static __inline__ int __ATTRS_o_ai vec_all_gt(vector signed char __a,
                                               vector signed char __b) {
   return __builtin_altivec_vcmpgtsb_p(__CR6_LT, __a, __b);
 }
 
 static __inline__ int __ATTRS_o_ai vec_all_gt(vector signed char __a,

 
 #ifdef __VSX__
 static __inline__ int __ATTRS_o_ai vec_all_gt(vector double __a,
                                               vector double __b) {
   return __builtin_vsx_xvcmpgtdp_p(__CR6_LT, __a, __b);
 }
 #endif
 
+#ifdef __POWER10_VECTOR__
+static __inline__ int __ATTRS_o_ai vec_all_gt(vector signed __int128 __a,
+                                              vector signed __int128 __b) {
+  return __builtin_altivec_vcmpgtsq_p(__CR6_LT, __a, __b);
+}
+
+static __inline__ int __ATTRS_o_ai vec_all_gt(vector unsigned __int128 __a,
+                                              vector unsigned __int128 __b) {
+  return __builtin_altivec_vcmpgtuq_p(__CR6_LT, __a, __b);
+}
+#endif
+
 /* vec_all_in */
 
 static __inline__ int __attribute__((__always_inline__))
 vec_all_in(vector float __a, vector float __b) {
   return __builtin_altivec_vcmpbfp_p(__CR6_EQ, __a, __b);
 }
 
 /* vec_all_le */

 
 #ifdef __VSX__
 static __inline__ int __ATTRS_o_ai vec_all_le(vector double __a,
                                               vector double __b) {
   return __builtin_vsx_xvcmpgedp_p(__CR6_LT, __b, __a);
 }
 #endif
 
+#ifdef __POWER10_VECTOR__
+static __inline__ int __ATTRS_o_ai vec_all_le(vector signed __int128 __a,
+                                              vector signed __int128 __b) {
+  return __builtin_altivec_vcmpgtsq_p(__CR6_EQ, __a, __b);
+}
+
+static __inline__ int __ATTRS_o_ai vec_all_le(vector unsigned __int128 __a,
+                                              vector unsigned __int128 __b) {
+  return __builtin_altivec_vcmpgtuq_p(__CR6_EQ, __a, __b);
+}
+#endif
+
 /* vec_all_lt */
 
 static __inline__ int __ATTRS_o_ai vec_all_lt(vector signed char __a,
                                               vector signed char __b) {
   return __builtin_altivec_vcmpgtsb_p(__CR6_LT, __b, __a);
 }
 
 static __inline__ int __ATTRS_o_ai vec_all_lt(vector signed char __a,

 
 #ifdef __VSX__
 static __inline__ int __ATTRS_o_ai vec_all_lt(vector double __a,
                                               vector double __b) {
   return __builtin_vsx_xvcmpgtdp_p(__CR6_LT, __b, __a);
 }
 #endif
 
+#ifdef __POWER10_VECTOR__
+static __inline__ int __ATTRS_o_ai vec_all_lt(vector signed __int128 __a,
+                                              vector signed __int128 __b) {
+  return __builtin_altivec_vcmpgtsq_p(__CR6_LT, __b, __a);
+}
+
+static __inline__ int __ATTRS_o_ai vec_all_lt(vector unsigned __int128 __a,
+                                              vector unsigned __int128 __b) {
+  return __builtin_altivec_vcmpgtuq_p(__CR6_LT, __b, __a);
+}
+#endif
+
 /* vec_all_nan */
 
 static __inline__ int __ATTRS_o_ai vec_all_nan(vector float __a) {
 #ifdef __VSX__
   return __builtin_vsx_xvcmpeqsp_p(__CR6_EQ, __a, __a);
 #else
   return __builtin_altivec_vcmpeqfp_p(__CR6_EQ, __a, __a);
 #endif

 
 #ifdef __VSX__
 static __inline__ int __ATTRS_o_ai vec_all_ne(vector double __a,
                                               vector double __b) {
   return __builtin_vsx_xvcmpeqdp_p(__CR6_EQ, __a, __b);
 }
 #endif
 
+#ifdef __POWER10_VECTOR__
+static __inline__ int __ATTRS_o_ai vec_all_ne(vector signed __int128 __a,
+                                              vector signed __int128 __b) {
+  return __builtin_altivec_vcmpequq_p(__CR6_EQ, __a, __b);
+}
+
+static __inline__ int __ATTRS_o_ai vec_all_ne(vector unsigned __int128 __a,
+                                              vector unsigned __int128 __b) {
+  return __builtin_altivec_vcmpequq_p(__CR6_EQ, __a, __b);
+}
+#endif
+
 /* vec_all_nge */
 
 static __inline__ int __ATTRS_o_ai vec_all_nge(vector float __a,
                                                vector float __b) {
 #ifdef __VSX__
   return __builtin_vsx_xvcmpgesp_p(__CR6_EQ, __a, __b);
 #else
   return __builtin_altivec_vcmpgefp_p(__CR6_EQ, __a, __b);

 
 #ifdef __VSX__
 static __inline__ int __ATTRS_o_ai vec_any_eq(vector double __a,
                                               vector double __b) {
   return __builtin_vsx_xvcmpeqdp_p(__CR6_EQ_REV, __a, __b);
 }
 #endif
 
+#ifdef __POWER10_VECTOR__
+static __inline__ int __ATTRS_o_ai vec_any_eq(vector signed __int128 __a,
+                                              vector signed __int128 __b) {
+  return __builtin_altivec_vcmpequq_p(__CR6_EQ_REV, __a, __b);
+}
+
+static __inline__ int __ATTRS_o_ai vec_any_eq(vector unsigned __int128 __a,
+                                              vector unsigned __int128 __b) {
+  return __builtin_altivec_vcmpequq_p(__CR6_EQ_REV, __a, __b);
+}
+#endif
+
 /* vec_any_ge */
 
 static __inline__ int __ATTRS_o_ai vec_any_ge(vector signed char __a,
                                               vector signed char __b) {
   return __builtin_altivec_vcmpgtsb_p(__CR6_LT_REV, __b, __a);
 }
 
 static __inline__ int __ATTRS_o_ai vec_any_ge(vector signed char __a,

 
 #ifdef __VSX__
 static __inline__ int __ATTRS_o_ai vec_any_ge(vector double __a,
                                               vector double __b) {
   return __builtin_vsx_xvcmpgedp_p(__CR6_EQ_REV, __a, __b);
 }
 #endif
 
+#ifdef __POWER10_VECTOR__
+static __inline__ int __ATTRS_o_ai vec_any_ge(vector signed __int128 __a,
+                                              vector signed __int128 __b) {
+  return __builtin_altivec_vcmpgtsq_p(__CR6_LT_REV, __b, __a);
+}
+
+static __inline__ int __ATTRS_o_ai vec_any_ge(vector unsigned __int128 __a,
+                                              vector unsigned __int128 __b) {
+  return __builtin_altivec_vcmpgtuq_p(__CR6_LT_REV, __b, __a);
+}
+#endif
+
 /* vec_any_gt */
 
 static __inline__ int __ATTRS_o_ai vec_any_gt(vector signed char __a,
                                               vector signed char __b) {
   return __builtin_altivec_vcmpgtsb_p(__CR6_EQ_REV, __a, __b);
 }
 
 static __inline__ int __ATTRS_o_ai vec_any_gt(vector signed char __a,

 
 #ifdef __VSX__
 static __inline__ int __ATTRS_o_ai vec_any_gt(vector double __a,
                                               vector double __b) {
   return __builtin_vsx_xvcmpgtdp_p(__CR6_EQ_REV, __a, __b);
 }
 #endif
 
+#ifdef __POWER10_VECTOR__
+static __inline__ int __ATTRS_o_ai vec_any_gt(vector signed __int128 __a,
+                                              vector signed __int128 __b) {
+  return __builtin_altivec_vcmpgtsq_p(__CR6_EQ_REV, __a, __b);
+}
+
+static __inline__ int __ATTRS_o_ai vec_any_gt(vector unsigned __int128 __a,
+                                              vector unsigned __int128 __b) {
+  return __builtin_altivec_vcmpgtuq_p(__CR6_EQ_REV, __a, __b);
+}
+#endif
+
 /* vec_any_le */
 
 static __inline__ int __ATTRS_o_ai vec_any_le(vector signed char __a,
                                               vector signed char __b) {
   return __builtin_altivec_vcmpgtsb_p(__CR6_LT_REV, __a, __b);
 }
 
 static __inline__ int __ATTRS_o_ai vec_any_le(vector signed char __a,

 
 #ifdef __VSX__
 static __inline__ int __ATTRS_o_ai vec_any_le(vector double __a,
                                               vector double __b) {
   return __builtin_vsx_xvcmpgedp_p(__CR6_EQ_REV, __b, __a);
 }
 #endif
 
+#ifdef __POWER10_VECTOR__
+static __inline__ int __ATTRS_o_ai vec_any_le(vector signed __int128 __a,
+                                              vector signed __int128 __b) {
+  return __builtin_altivec_vcmpgtsq_p(__CR6_LT_REV, __a, __b);
+}
+
+static __inline__ int __ATTRS_o_ai vec_any_le(vector unsigned __int128 __a,
+                                              vector unsigned __int128 __b) {
+  return __builtin_altivec_vcmpgtuq_p(__CR6_LT_REV, __a, __b);
+}
+#endif
+
 /* vec_any_lt */
 
 static __inline__ int __ATTRS_o_ai vec_any_lt(vector signed char __a,
                                               vector signed char __b) {
   return __builtin_altivec_vcmpgtsb_p(__CR6_EQ_REV, __b, __a);
 }
 
 static __inline__ int __ATTRS_o_ai vec_any_lt(vector signed char __a,

 
 #ifdef __VSX__
 static __inline__ int __ATTRS_o_ai vec_any_lt(vector double __a,
                                               vector double __b) {
   return __builtin_vsx_xvcmpgtdp_p(__CR6_EQ_REV, __b, __a);
 }
 #endif
 
+#ifdef __POWER10_VECTOR__
+static __inline__ int __ATTRS_o_ai vec_any_lt(vector signed __int128 __a,
+                                              vector signed __int128 __b) {
+  return __builtin_altivec_vcmpgtsq_p(__CR6_EQ_REV, __b, __a);
+}
+
+static __inline__ int __ATTRS_o_ai vec_any_lt(vector unsigned __int128 __a,
+                                              vector unsigned __int128 __b) {
+  return __builtin_altivec_vcmpgtuq_p(__CR6_EQ_REV, __b, __a);
+}
+#endif
+
 /* vec_any_nan */
 
 static __inline__ int __attribute__((__always_inline__))
 vec_any_nan(vector float __a) {
   return __builtin_altivec_vcmpeqfp_p(__CR6_LT_REV, __a, __a);
 }
 
 /* vec_any_ne */

 
 #ifdef __VSX__
 static __inline__ int __ATTRS_o_ai vec_any_ne(vector double __a,
                                               vector double __b) {
   return __builtin_vsx_xvcmpeqdp_p(__CR6_LT_REV, __a, __b);
 }
 #endif
 
+#ifdef __POWER10_VECTOR__
+static __inline__ int __ATTRS_o_ai vec_any_ne(vector signed __int128 __a,
+                                              vector signed __int128 __b) {
+  return __builtin_altivec_vcmpequq_p(__CR6_LT_REV, __a, __b);
+}
+
+static __inline__ int __ATTRS_o_ai vec_any_ne(vector unsigned __int128 __a,
+                                              vector unsigned __int128 __b) {
+  return __builtin_altivec_vcmpequq_p(__CR6_LT_REV, __a, __b);
+}
+#endif
+
 /* vec_any_nge */
 
 static __inline__ int __attribute__((__always_inline__))
 vec_any_nge(vector float __a, vector float __b) {
   return __builtin_altivec_vcmpgefp_p(__CR6_LT_REV, __a, __b);
 }
 
 /* vec_any_ngt */

clang/test/CodeGen/builtins-ppc-p10vector.c

 
 vector bool __int128 test_vec_cmple_u128(void) {
   // CHECK-LABEL: @test_vec_cmple_u128(
   // CHECK: call <1 x i128> @llvm.ppc.altivec.vcmpgtuq(<1 x i128>
   // CHECK-NEXT: %neg.i.i = xor <1 x i128> %8, <i128 -1>
   // CHECK-NEXT: ret <1 x i128>
   return vec_cmple(vui128a, vui128b);
 }
+
+int test_vec_any_eq_u128(void) {
+  // CHECK-LABEL: @test_vec_any_eq_u128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpequq.p(i32 1, <1 x i128> %2, <1 x i128> %3)
+  // CHECK-NEXT: ret i32
+  return vec_any_eq(vui128a, vui128b);
+}
+
+int test_vec_any_eq_s128(void) {
+  // CHECK-LABEL: @test_vec_any_eq_s128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpequq.p(i32 1, <1 x i128> %2, <1 x i128> %3)
+  // CHECK-NEXT: ret i32
+  return vec_any_eq(vsi128a, vsi128b);
+}
+
+int test_vec_any_ne_s128(void) {
+  // CHECK-LABEL: @test_vec_any_ne_s128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpequq.p(i32 3, <1 x i128> %2, <1 x i128> %3)
+  // CHECK-NEXT: ret i32
+  return vec_any_ne(vsi128a, vsi128b);
+}
+
+int test_vec_any_ne_u128(void) {
+  // CHECK-LABEL: @test_vec_any_ne_u128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpequq.p(i32 3, <1 x i128> %2, <1 x i128> %3)
+  // CHECK-NEXT: ret i32
+  return vec_any_ne(vui128a, vui128b);
+}
+
+int test_vec_any_lt_s128(void) {
+  // CHECK-LABEL: @test_vec_any_lt_s128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpgtsq.p(i32 1, <1 x i128> %2, <1 x i128> %3)
+  // CHECK-NEXT: ret i32
+  return vec_any_lt(vsi128a, vsi128b);
+}
+
+int test_vec_any_lt_u128(void) {
+  // CHECK-LABEL: @test_vec_any_lt_u128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpgtuq.p(i32 1, <1 x i128> %2, <1 x i128> %3)
+  // CHECK-NEXT: ret i32
+  return vec_any_lt(vui128a, vui128b);
+}
+
+int test_vec_any_gt_s128(void) {
+  // CHECK-LABEL: @test_vec_any_gt_s128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpgtsq.p(i32 1, <1 x i128> %2, <1 x i128> %3)
+  // CHECK-NEXT: ret i32
+  return vec_any_gt(vsi128a, vsi128b);
+}
+
+int test_vec_any_gt_u128(void) {
+  // CHECK-LABEL: @test_vec_any_gt_u128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpgtuq.p(i32 1, <1 x i128> %2, <1 x i128> %3)
+  // CHECK-NEXT: ret i32
+  return vec_any_gt(vui128a, vui128b);
+}
+
+int test_vec_any_le_s128(void) {
+  // CHECK-LABEL: @test_vec_any_le_s128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpgtsq.p(i32 3, <1 x i128> %2, <1 x i128> %3)
+  // CHECK-NEXT: ret i32
+  return vec_any_le(vsi128a, vsi128b);
+}
+
+int test_vec_any_le_u128(void) {
+  // CHECK-LABEL: @test_vec_any_le_u128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpgtuq.p(i32 3, <1 x i128> %2, <1 x i128> %3)
+  // CHECK-NEXT: ret i32
+  return vec_any_le(vui128a, vui128b);
+}
+
+int test_vec_any_ge_s128(void) {
+  // CHECK-LABEL: @test_vec_any_ge_s128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpgtsq.p(i32 3, <1 x i128> %2, <1 x i128> %3)
+  // CHECK-NEXT: ret i32
+  return vec_any_ge(vsi128a, vsi128b);
+}
+
+int test_vec_any_ge_u128(void) {
+  // CHECK-LABEL: @test_vec_any_ge_u128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpgtuq.p(i32 3, <1 x i128> %2, <1 x i128> %3)
+  // CHECK-NEXT: ret i32
+  return vec_any_ge(vui128a, vui128b);
+}
+
+int test_vec_all_eq_s128(void) {
+  // CHECK-LABEL: @test_vec_all_eq_s128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpequq.p(i32 2, <1 x i128> %2, <1 x i128> %3)
+  // CHECK-NEXT: ret i32
+  return vec_all_eq(vsi128a, vsi128b);
+}
+
+int test_vec_all_eq_u128(void) {
+  // CHECK-LABEL: @test_vec_all_eq_u128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpequq.p(i32 2, <1 x i128> %2, <1 x i128> %3)
+  // CHECK-NEXT: ret i32
+  return vec_all_eq(vui128a, vui128b);
+}
+
+int test_vec_all_ne_s128(void) {
+  // CHECK-LABEL: @test_vec_all_ne_s128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpequq.p(i32 0, <1 x i128> %2, <1 x i128> %3)
+  // CHECK-NEXT: ret i32
+  return vec_all_ne(vsi128a, vsi128b);
+}
+
+int test_vec_all_ne_u128(void) {
+  // CHECK-LABEL: @test_vec_all_ne_u128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpequq.p(i32 0, <1 x i128> %2, <1 x i128> %3)
+  // CHECK-NEXT: ret i32
+  return vec_all_ne(vui128a, vui128b);
+}
+
+int test_vec_all_lt_s128(void) {
+  // CHECK-LABEL: @test_vec_all_lt_s128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpgtsq.p(i32 2, <1 x i128> %2, <1 x i128> %3)
+  // CHECK-NEXT: ret i32
+  return vec_all_lt(vsi128a, vsi128b);
+}
+
+int test_vec_all_lt_u128(void) {
+  // CHECK-LABEL: @test_vec_all_lt_u128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpgtuq.p(i32 2, <1 x i128> %2, <1 x i128> %3)
+  // CHECK: ret i32
+  return vec_all_lt(vui128a, vui128b);
+}
+
+int test_vec_all_gt_s128(void) {
+  // CHECK-LABEL: @test_vec_all_gt_s128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpgtsq.p(i32 2, <1 x i128> %2, <1 x i128> %3)
+  // CHECK-NEXT: ret i32
+  return vec_all_gt(vsi128a, vsi128b);
+}
+
+int test_vec_all_gt_u128(void) {
+  // CHECK-LABEL: @test_vec_all_gt_u128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpgtuq.p(i32 2, <1 x i128> %2, <1 x i128> %3)
+  // CHECK-NEXT: ret i32
+  return vec_all_gt(vui128a, vui128b);
+}
+
+int test_vec_all_le_s128(void) {
+  // CHECK-LABEL: @test_vec_all_le_s128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpgtsq.p(i32 0, <1 x i128> %2, <1 x i128> %3)
+  // CHECK-NEXT: ret i32
+  return vec_all_le(vsi128a, vsi128b);
+}
+
+int test_vec_all_le_u128(void) {
+  // CHECK-LABEL: @test_vec_all_le_u128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpgtuq.p(i32 0, <1 x i128> %2, <1 x i128> %3)
+  // CHECK-NEXT: ret i32
+  return vec_all_le(vui128a, vui128b);
+}
+
+int test_vec_all_ge_s128(void) {
+  // CHECK-LABEL: @test_vec_all_ge_s128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpgtsq.p(i32 0, <1 x i128> %2, <1 x i128> %3)
+  // CHECK-NEXT: ret i32
+  return vec_all_ge(vsi128a, vsi128b);
+}
+
+int test_vec_all_ge_u128(void) {
+  // CHECK-LABEL: @test_vec_all_ge_u128(
+  // CHECK: call i32 @llvm.ppc.altivec.vcmpgtuq.p(i32 0, <1 x i128> %2, <1 x i128> %3)
+  // CHECK-NEXT: ret i32
+  return vec_all_ge(vui128a, vui128b);
+}

llvm/include/llvm/IR/IntrinsicsPowerPC.td

def int_ppc_altivec_vcmpequq : GCCBuiltin<"__builtin_altivec_vcmpequq">,
                Intrinsic<[llvm_v1i128_ty], [llvm_v1i128_ty, llvm_v1i128_ty],
                          [IntrNoMem]>;
   def int_ppc_altivec_vcmpgtsq : GCCBuiltin<"__builtin_altivec_vcmpgtsq">,
               Intrinsic<[llvm_v1i128_ty], [llvm_v1i128_ty, llvm_v1i128_ty],
                         [IntrNoMem]>;
   def int_ppc_altivec_vcmpgtuq : GCCBuiltin<"__builtin_altivec_vcmpgtuq">,
               Intrinsic<[llvm_v1i128_ty], [llvm_v1i128_ty, llvm_v1i128_ty],
                         [IntrNoMem]>;
+  def int_ppc_altivec_vcmpequq_p : GCCBuiltin<"__builtin_altivec_vcmpequq_p">,
+              Intrinsic<[llvm_i32_ty],
+                        [llvm_i32_ty,llvm_v1i128_ty,llvm_v1i128_ty],
+                        [IntrNoMem]>;
+  def int_ppc_altivec_vcmpgtsq_p : GCCBuiltin<"__builtin_altivec_vcmpgtsq_p">,
+              Intrinsic<[llvm_i32_ty],
+                        [llvm_i32_ty,llvm_v1i128_ty,llvm_v1i128_ty],
+                        [IntrNoMem]>;
+  def int_ppc_altivec_vcmpgtuq_p : GCCBuiltin<"__builtin_altivec_vcmpgtuq_p">,
+              Intrinsic<[llvm_i32_ty],
+                        [llvm_i32_ty,llvm_v1i128_ty,llvm_v1i128_ty],
+                        [IntrNoMem]>;
 
   // Predicate Comparisons.  The first operand specifies interpretation of CR6.
   def int_ppc_altivec_vcmpbfp_p : GCCBuiltin<"__builtin_altivec_vcmpbfp_p">,
               Intrinsic<[llvm_i32_ty],[llvm_i32_ty,llvm_v4f32_ty,llvm_v4f32_ty],
                         [IntrNoMem]>;
   def int_ppc_altivec_vcmpeqfp_p : GCCBuiltin<"__builtin_altivec_vcmpeqfp_p">,
               Intrinsic<[llvm_i32_ty],[llvm_i32_ty,llvm_v4f32_ty,llvm_v4f32_ty],
                         [IntrNoMem]>;

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

case Intrinsic::ppc_altivec_vcmpgtuq:
     case Intrinsic::ppc_altivec_vcmpgtsq:
       CompareOpc = 903;
       break;
     case Intrinsic::ppc_altivec_vcmpgtuq:
       CompareOpc = 647;
       break;
     }
     break;
+  case Intrinsic::ppc_altivec_vcmpequq_p:
+  case Intrinsic::ppc_altivec_vcmpgtsq_p:
+  case Intrinsic::ppc_altivec_vcmpgtuq_p:
+    if (!Subtarget.isISA3_1())
+      return false;
+    switch (IntrinsicID) {
+    default:
+      llvm_unreachable("Unknown comparison intrinsic.");
+    case Intrinsic::ppc_altivec_vcmpequq_p:
+      CompareOpc = 455;
+      break;
+    case Intrinsic::ppc_altivec_vcmpgtsq_p:
+      CompareOpc = 903;
+      break;
+    case Intrinsic::ppc_altivec_vcmpgtuq_p:
+      CompareOpc = 647;
+      break;
+    }
+    isDot = true;
+    break;
   }
   return true;
 }
 
 /// LowerINTRINSIC_WO_CHAIN - If this is an intrinsic that we want to custom
 /// lower, do it, otherwise return null.
 SDValue PPCTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
                                                    SelectionDAG &DAG) const {
case ISD::INTRINSIC_WO_CHAIN: {
     switch (cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue()) {
     default: break;
     case Intrinsic::ppc_altivec_vcmpbfp_p:
     case Intrinsic::ppc_altivec_vcmpeqfp_p:
     case Intrinsic::ppc_altivec_vcmpequb_p:
     case Intrinsic::ppc_altivec_vcmpequh_p:
     case Intrinsic::ppc_altivec_vcmpequw_p:
     case Intrinsic::ppc_altivec_vcmpequd_p:
+    case Intrinsic::ppc_altivec_vcmpequq_p:
     case Intrinsic::ppc_altivec_vcmpgefp_p:
     case Intrinsic::ppc_altivec_vcmpgtfp_p:
     case Intrinsic::ppc_altivec_vcmpgtsb_p:
     case Intrinsic::ppc_altivec_vcmpgtsh_p:
     case Intrinsic::ppc_altivec_vcmpgtsw_p:
     case Intrinsic::ppc_altivec_vcmpgtsd_p:
+    case Intrinsic::ppc_altivec_vcmpgtsq_p:
     case Intrinsic::ppc_altivec_vcmpgtub_p:
     case Intrinsic::ppc_altivec_vcmpgtuh_p:
     case Intrinsic::ppc_altivec_vcmpgtuw_p:
     case Intrinsic::ppc_altivec_vcmpgtud_p:
+    case Intrinsic::ppc_altivec_vcmpgtuq_p:
       Known.Zero = ~1U;  // All bits but the low one are known to be zero.
       break;
     }
   }
   }
 }
 
 Align PPCTargetLowering::getPrefLoopAlignment(MachineLoop *ML) const {

llvm/test/CodeGen/PowerPC/vec_cmpq.ll

 }
 
 define <1 x i128> @test_vcmpgtuq(<1 x i128> %x, <1 x i128> %y) {
        %tmp = tail call <1 x i128> @llvm.ppc.altivec.vcmpgtuq(<1 x i128> %x, <1 x i128> %y)
        ret <1 x i128> %tmp
 ; CHECK-LABEL: test_vcmpgtuq
 ; CHECK: vcmpgtuq {{[0-9]+}}, {{[0-9]+}}, {{[0-9]+}}
 }
+
+declare i32 @llvm.ppc.altivec.vcmpequq.p(i32, <1 x i128>, <1 x i128>) nounwind readnone
+declare i32 @llvm.ppc.altivec.vcmpgtsq.p(i32, <1 x i128>, <1 x i128>) nounwind readnone
+declare i32 @llvm.ppc.altivec.vcmpgtuq.p(i32, <1 x i128>, <1 x i128>) nounwind readnone
+
+define i32 @test_vcmpequq_p(<1 x i128> %x, <1 x i128> %y) {
+      %tmp = tail call i32 @llvm.ppc.altivec.vcmpequq.p(i32 2, <1 x i128> %x, <1 x i128> %y)
+      ret i32 %tmp
+; CHECK-LABEL: test_vcmpequq_p:
+; CHECK: vcmpequq. {{[0-9]+}}, {{[0-9]+}}, {{[0-9]+}}
+; CHECK: blr
+}
+
+define i32 @test_vcmpgtsq_p(<1 x i128> %x, <1 x i128> %y) {
+      %tmp = tail call i32 @llvm.ppc.altivec.vcmpgtsq.p(i32 2, <1 x i128> %x, <1 x i128> %y)
+      ret i32 %tmp
+; CHECK-LABEL: test_vcmpgtsq_p
+; CHECK: vcmpgtsq. {{[0-9]+}}, {{[0-9]+}}, {{[0-9]+}}
+; CHECK: blr
+}
+
+define i32 @test_vcmpgtuq_p(<1 x i128> %x, <1 x i128> %y) {
+      %tmp = tail call i32 @llvm.ppc.altivec.vcmpgtuq.p(i32 2, <1 x i128> %x, <1 x i128> %y)
+      ret i32 %tmp
+; CHECK-LABEL: test_vcmpgtuq_p
+; CHECK: vcmpgtuq. {{[0-9]+}}, {{[0-9]+}}, {{[0-9]+}}
+; CHECK: blr
+}