[clang] ef90657 - [PowerPC] Fix vec_add for 64-bit on pre-Power7 subtargets

[Nemanja Ivanovic via cfe-commits]

Author: Nemanja Ivanovic
Date: 2021-06-24T18:42:44-05:00
New Revision: ef906573a127cffef7cae75d5155c15a8a2a3a5e

URL: https://github.com/llvm/llvm-project/commit/ef906573a127cffef7cae75d5155c15a8a2a3a5e
DIFF: https://github.com/llvm/llvm-project/commit/ef906573a127cffef7cae75d5155c15a8a2a3a5e.diff

LOG: [PowerPC] Fix vec_add for 64-bit on pre-Power7 subtargets

The shift of the carry was actually incorrect.

Added: 
    

Modified: 
    clang/lib/Headers/altivec.h
    clang/test/CodeGen/builtins-ppc-vsx.c

Removed: 
    


################################################################################
diff  --git a/clang/lib/Headers/altivec.h b/clang/lib/Headers/altivec.h
index dadf6b5cf75bb..3517da798547a 100644
--- a/clang/lib/Headers/altivec.h
+++ b/clang/lib/Headers/altivec.h
@@ -312,16 +312,20 @@ vec_add_u128(vector unsigned char __a, vector unsigned char __b) {
 #elif defined(__VSX__)
 static __inline__ vector signed long long __ATTRS_o_ai
 vec_add(vector signed long long __a, vector signed long long __b) {
+#ifdef __LITTLE_ENDIAN__
+  // Little endian systems on CPU's prior to Power8 don't really exist
+  // so scalarizing is fine.
+  return __a + __b;
+#else
   vector unsigned int __res =
       (vector unsigned int)__a + (vector unsigned int)__b;
   vector unsigned int __carry = __builtin_altivec_vaddcuw(
       (vector unsigned int)__a, (vector unsigned int)__b);
-#ifdef __LITTLE_ENDIAN__
-  __carry = __builtin_shufflevector(__carry, __carry, 3, 0, 1, 2);
-#else
-  __carry = __builtin_shufflevector(__carry, __carry, 1, 2, 3, 0);
-#endif
+  __carry = __builtin_shufflevector((vector unsigned char)__carry,
+                                    (vector unsigned char)__carry, 0, 0, 0, 7,
+                                    0, 0, 0, 0, 0, 0, 0, 15, 0, 0, 0, 0);
   return (vector signed long long)(__res + __carry);
+#endif
 }
 
 static __inline__ vector unsigned long long __ATTRS_o_ai

diff  --git a/clang/test/CodeGen/builtins-ppc-vsx.c b/clang/test/CodeGen/builtins-ppc-vsx.c
index abd08d463e634..b5ddd03722ad0 100644
--- a/clang/test/CodeGen/builtins-ppc-vsx.c
+++ b/clang/test/CodeGen/builtins-ppc-vsx.c
@@ -2319,21 +2319,15 @@ void test_p8overloads_backwards_compat() {
   res_vsll = vec_add(vsll, vsll);
   // CHECK: add <4 x i32>
   // CHECK: call <4 x i32> @llvm.ppc.altivec.vaddcuw
-  // CHECK: shufflevector <4 x i32> {{%.*}}, <4 x i32> {{%.*}}, <4 x i32> <i32 1, i32 2, i32 3, i32 0>
+  // CHECK: shufflevector <16 x i8> {{%.*}}, <16 x i8> {{%.*}}, <16 x i32> <i32 0, i32 0, i32 0, i32 7, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 15, i32 0, i32 0, i32 0, i32 0>
   // CHECK: add <4 x i32>
-  // CHECK-LE: add <4 x i32>
-  // CHECK-LE: call <4 x i32> @llvm.ppc.altivec.vaddcuw
-  // CHECK-LE: shufflevector <4 x i32> {{%.*}}, <4 x i32> {{%.*}}, <4 x i32> <i32 3, i32 0, i32 1, i32 2>
-  // CHECK-LE: add <4 x i32>
+  // CHECK-LE: add <2 x i64>
   res_vull = vec_add(vull, vull);
   // CHECK: add <4 x i32>
   // CHECK: call <4 x i32> @llvm.ppc.altivec.vaddcuw
-  // CHECK: shufflevector <4 x i32> {{%.*}}, <4 x i32> {{%.*}}, <4 x i32> <i32 1, i32 2, i32 3, i32 0>
+  // CHECK: shufflevector <16 x i8> {{%.*}}, <16 x i8> {{%.*}}, <16 x i32> <i32 0, i32 0, i32 0, i32 7, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 15, i32 0, i32 0, i32 0, i32 0>
   // CHECK: add <4 x i32>
-  // CHECK-LE: add <4 x i32>
-  // CHECK-LE: call <4 x i32> @llvm.ppc.altivec.vaddcuw
-  // CHECK-LE: shufflevector <4 x i32> {{%.*}}, <4 x i32> {{%.*}}, <4 x i32> <i32 3, i32 0, i32 1, i32 2>
-  // CHECK-LE: add <4 x i32>
+  // CHECK-LE: add <2 x i64>
   dummy();
   // CHECK: call void @dummy()
   // CHECK-LE: call void @dummy()