[PowerPC] Implement remaining vector comparison builtins

[jtony]

Summary

Signatures:
vector bool char vec_cmpeq (vector bool char, vector bool char);
vector bool int vec_cmpeq (vector bool int, vector bool int);
vector bool long long vec_cmpeq (vector bool long long, vector bool long long);
vector bool short vec_cmpeq (vector bool short, vector bool short);

[jtony]

Fix some existing problems in the original source (not introduced by this patch). Calling vec_cmpeq function with two different type parameters will cause ambiguous errors. Have to explicitly convert the type of one parameter to the type of another parameter. Make the two input parameters consistent so the compiler knows exactly which overloads to invoke.

[nemanjai]

I think that other than the few minor nits, this patch is fine.

lib/Headers/altivec.h

1556

When we have all three overloads, the order in the rest of the file is typically:

1. signed <TYPE>
2. unsigned <TYPE>
3. bool <TYPE>

Let's keep this consistent please.

2379

I assume that this change was made by clang-format. However, I think this is one of those examples where not formatting according to its rules makes the code more readable. So I personally prefer the old formatting.

[jtony]

Fix a few format and layout issues.

Format: clang-format is not always correct, sometimes separate parameters in different lines is better.
Layout: put the overloaded builtin functions in such order: signed version, unsigned version and bool version.

[jtony]

I migrated the commit from the old anonymous repository to the new jtony repository to prepare for committing upstream (use git svn dcommit). Should not have any difference with the previous patches, update it just in case.

[kbarton]

LGTM

lib/Headers/altivec.h

                                                       (vector char)__b);
 }
 
+static __inline__ vector bool char __ATTRS_o_ai
+vec_cmpeq(vector bool char __a, vector bool char __b) {
+  return (vector bool char)__builtin_altivec_vcmpequb((vector char)__a,
+                                                      (vector char)__b);
+}
+
 static __inline__ vector bool short __ATTRS_o_ai vec_cmpeq(vector short __a,
                                                            vector short __b) {
   return (vector bool short)__builtin_altivec_vcmpequh(__a, __b);

                                                        (vector short)__b);
 }
 
+static __inline__ vector bool short __ATTRS_o_ai
+vec_cmpeq(vector bool short __a, vector bool short __b) {
+  return (vector bool short)__builtin_altivec_vcmpequh((vector short)__a,
+                                                       (vector short)__b);
+}
+
 static __inline__ vector bool int __ATTRS_o_ai vec_cmpeq(vector int __a,
                                                          vector int __b) {
   return (vector bool int)__builtin_altivec_vcmpequw(__a, __b);

                                                      (vector int)__b);
 }
 
+static __inline__ vector bool int __ATTRS_o_ai vec_cmpeq(vector bool int __a,
+                                                         vector bool int __b) {
+  return (vector bool int)__builtin_altivec_vcmpequw((vector int)__a,
+                                                     (vector int)__b);
+}
+
 #ifdef __POWER8_VECTOR__
 static __inline__ vector bool long long __ATTRS_o_ai
 vec_cmpeq(vector signed long long __a, vector signed long long __b) {

   return (vector bool long long)__builtin_altivec_vcmpequd(
       (vector long long)__a, (vector long long)__b);
 }
+
+static __inline__ vector bool long long __ATTRS_o_ai
+vec_cmpeq(vector bool long long __a, vector bool long long __b) {
+  return (vector bool long long)__builtin_altivec_vcmpequd(
+      (vector long long)__a, (vector long long)__b);
+}
+
 #endif
 
 static __inline__ vector bool int __ATTRS_o_ai vec_cmpeq(vector float __a,

      result if either is zero.
   */
   vector bool char __tmp1 = vec_cmpeq(__a, __b);
-  vector bool char __tmp2 = __tmp1 | vec_cmpeq(__tmp1, __a) |
-    vec_cmpeq(__tmp1, __b);
+  vector bool char __tmp2 = __tmp1 |
+                            vec_cmpeq((vector signed char)__tmp1, __a) |
+                            vec_cmpeq((vector signed char)__tmp1, __b);
 
   vector unsigned long long __res =
 #ifdef __LITTLE_ENDIAN__
       vec_cnttz((vector unsigned long long)__tmp2);
 #else
       vec_cntlz((vector unsigned long long)__tmp2);
 #endif
   if (__res[0] == 64) {
     return (__res[1] + 64) >> 3;

 vec_first_match_or_eos_index(vector unsigned char __a,
                              vector unsigned char __b) {
   vector bool char __tmp1 = vec_cmpeq(__a, __b);
-  vector bool char __tmp2 = __tmp1 | vec_cmpeq(__tmp1, __a) |
-    vec_cmpeq(__tmp1, __b);
+  vector bool char __tmp2 = __tmp1 |
+                            vec_cmpeq((vector unsigned char)__tmp1, __a) |
+                            vec_cmpeq((vector unsigned char)__tmp1, __b);
 
   vector unsigned long long __res =
 #ifdef __LITTLE_ENDIAN__
       vec_cnttz((vector unsigned long long)__tmp2);
 #else
       vec_cntlz((vector unsigned long long)__tmp2);
 #endif
   if (__res[0] == 64) {
     return (__res[1] + 64) >> 3;

 static __inline__ unsigned __ATTRS_o_ai
 vec_first_match_or_eos_index(vector signed short __a, vector signed short __b) {
   vector bool short __tmp1 = vec_cmpeq(__a, __b);
-  vector bool short __tmp2 = __tmp1 | vec_cmpeq(__tmp1, __a) |
-    vec_cmpeq(__tmp1, __b);
+  vector bool short __tmp2 = __tmp1 |
+                             vec_cmpeq((vector signed short)__tmp1, __a) |
+                             vec_cmpeq((vector signed short)__tmp1, __b);
 
   vector unsigned long long __res =
 #ifdef __LITTLE_ENDIAN__
       vec_cnttz((vector unsigned long long)__tmp2);
 #else
       vec_cntlz((vector unsigned long long)__tmp2);
 #endif
   if (__res[0] == 64) {
     return (__res[1] + 64) >> 4;

 vec_first_match_or_eos_index(vector unsigned short __a,
                              vector unsigned short __b) {
   vector bool short __tmp1 = vec_cmpeq(__a, __b);
-  vector bool short __tmp2 = __tmp1 | vec_cmpeq(__tmp1, __a) |
-    vec_cmpeq(__tmp1, __b);
+  vector bool short __tmp2 = __tmp1 |
+                             vec_cmpeq((vector unsigned short)__tmp1, __a) |
+                             vec_cmpeq((vector unsigned short)__tmp1, __b);
 
   vector unsigned long long __res =
 #ifdef __LITTLE_ENDIAN__
       vec_cnttz((vector unsigned long long)__tmp2);
 #else
       vec_cntlz((vector unsigned long long)__tmp2);
 #endif
   if (__res[0] == 64) {
     return (__res[1] + 64) >> 4;

 static __inline__ unsigned __ATTRS_o_ai
 vec_first_match_or_eos_index(vector signed int __a, vector signed int __b) {
   vector bool int __tmp1 = vec_cmpeq(__a, __b);
-  vector bool int __tmp2 = __tmp1 | vec_cmpeq(__tmp1, __a) |
-    vec_cmpeq(__tmp1, __b);
+  vector bool int __tmp2 = __tmp1 | vec_cmpeq((vector signed int)__tmp1, __a) |
+                           vec_cmpeq((vector signed int)__tmp1, __b);
 
   vector unsigned long long __res =
 #ifdef __LITTLE_ENDIAN__
       vec_cnttz((vector unsigned long long)__tmp2);
 #else
       vec_cntlz((vector unsigned long long)__tmp2);
 #endif
   if (__res[0] == 64) {
     return (__res[1] + 64) >> 5;

 }
 
 static __inline__ unsigned __ATTRS_o_ai
-vec_first_match_or_eos_index(vector unsigned int __a,
-                             vector unsigned int __b) {
+vec_first_match_or_eos_index(vector unsigned int __a, vector unsigned int __b) {
   vector bool int __tmp1 = vec_cmpeq(__a, __b);
-  vector bool int __tmp2 = __tmp1 | vec_cmpeq(__tmp1, __a) |
-    vec_cmpeq(__tmp1, __b);
+  vector bool int __tmp2 = __tmp1 |
+                           vec_cmpeq((vector unsigned int)__tmp1, __a) |
+                           vec_cmpeq((vector unsigned int)__tmp1, __b);
 
   vector unsigned long long __res =
 #ifdef __LITTLE_ENDIAN__

test/CodeGen/builtins-ppc-altivec.c

 // CHECK: @llvm.ppc.altivec.vcmpequb
 // CHECK-LE: @llvm.ppc.altivec.vcmpequb
 
+  res_vbc = vec_cmpeq(vbc, vbc);
+// CHECK: @llvm.ppc.altivec.vcmpequb
+// CHECK-LE: @llvm.ppc.altivec.vcmpequb
+
+  res_vbc = vec_cmpeq(vbc, vbc);
+// CHECK: @llvm.ppc.altivec.vcmpequb
+// CHECK-LE: @llvm.ppc.altivec.vcmpequb
+
   res_vbs = vec_cmpeq(vs, vs);
 // CHECK: @llvm.ppc.altivec.vcmpequh
 // CHECK-LE: @llvm.ppc.altivec.vcmpequh

 // CHECK: @llvm.ppc.altivec.vcmpequh
 // CHECK-LE: @llvm.ppc.altivec.vcmpequh
 
+  res_vbs = vec_cmpeq(vbs, vbs);
+// CHECK: @llvm.ppc.altivec.vcmpequh
+// CHECK-LE: @llvm.ppc.altivec.vcmpequh
+
+  res_vbs = vec_cmpeq(vbs, vbs);
+// CHECK: @llvm.ppc.altivec.vcmpequh
+// CHECK-LE: @llvm.ppc.altivec.vcmpequh
+
   res_vbi = vec_cmpeq(vi, vi);
 // CHECK: @llvm.ppc.altivec.vcmpequw
 // CHECK-LE: @llvm.ppc.altivec.vcmpequw

 // CHECK: @llvm.ppc.altivec.vcmpequw
 // CHECK-LE: @llvm.ppc.altivec.vcmpequw
 
+  res_vbi = vec_cmpeq(vbi, vbi);
+// CHECK: @llvm.ppc.altivec.vcmpequw
+// CHECK-LE: @llvm.ppc.altivec.vcmpequw
+
+  res_vbi = vec_cmpeq(vbi, vbi);
+// CHECK: @llvm.ppc.altivec.vcmpequw
+// CHECK-LE: @llvm.ppc.altivec.vcmpequw
+
   res_vbi = vec_cmpeq(vf, vf);
 // CHECK: @llvm.ppc.altivec.vcmpeqfp
 // CHECK-LE: @llvm.ppc.altivec.vcmpeqfp

test/CodeGen/builtins-ppc-p8vector.c

 // CHECK-PPC: warning: implicit declaration of function 'vec_mergeo'
   
   /* vec_cmpeq */
+  res_vbll = vec_cmpeq(vbll, vbll);
+// CHECK: @llvm.ppc.altivec.vcmpequd
+// CHECK-LE: @llvm.ppc.altivec.vcmpequd
+// CHECK-PPC: error: call to 'vec_cmpeq' is ambiguous
+
   res_vbll = vec_cmpeq(vsll, vsll);
 // CHECK: @llvm.ppc.altivec.vcmpequd
 // CHECK-LE: @llvm.ppc.altivec.vcmpequd