From 2f782a796a2b0a4bb5ff772577f96eacdb9254c1 Mon Sep 17 00:00:00 2001
From: Nemanja Ivanovic <nemanja.i.ibm@gmail.com>
Date: Tue, 23 Mar 2021 05:51:19 -0500
Subject: [PATCH] [PowerPC] Add more missing overloads to altivec.h

Add overloads that perform subtraction on v1i128 that take and
produce vector unsigned char to avoid needing to use __int128.
The overloads are suffixed with _u128 and are needed for targets
where __int128 isn't supported (AIX).
---
 clang/lib/Headers/altivec.h                | 28 ++++++++++++++++++++++++++++
 clang/test/CodeGen/builtins-ppc-quadword.c | 18 ++++++++++++++++++
 2 files changed, 46 insertions(+)

diff --git a/clang/lib/Headers/altivec.h b/clang/lib/Headers/altivec.h
index f3340f2..c567441 100644
--- a/clang/lib/Headers/altivec.h
+++ b/clang/lib/Headers/altivec.h
@@ -11636,6 +11636,12 @@ static __inline__ vector signed __int128 __ATTRS_o_ai
 vec_subc(vector signed __int128 __a, vector signed __int128 __b) {
   return __builtin_altivec_vsubcuq(__a, __b);
 }
+
+static __inline__ vector unsigned char __attribute__((__always_inline__))
+vec_subc_u128(vector unsigned char __a, vector unsigned char __b) {
+  return (vector unsigned char)__builtin_altivec_vsubcuq(
+      (vector unsigned __int128)__a, (vector unsigned __int128)__b);
+}
 #endif // defined(__POWER8_VECTOR__) && defined(__powerpc64__)
 
 /* vec_vsubcuw */
@@ -11852,6 +11858,12 @@ vec_vsubuqm(vector unsigned __int128 __a, vector unsigned __int128 __b) {
   return __a - __b;
 }
 
+static __inline__ vector unsigned char __attribute__((__always_inline__))
+vec_sub_u128(vector unsigned char __a, vector unsigned char __b) {
+  return (vector unsigned char)((vector unsigned __int128)__a -
+                                (vector unsigned __int128)__b);
+}
+
 /* vec_vsubeuqm */
 
 
@@ -11879,6 +11891,14 @@ vec_sube(vector unsigned __int128 __a, vector unsigned __int128 __b,
   return __builtin_altivec_vsubeuqm(__a, __b, __c);
 }
 
+static __inline__ vector unsigned char __attribute__((__always_inline__))
+vec_sube_u128(vector unsigned char __a, vector unsigned char __b,
+              vector unsigned char __c) {
+  return (vector unsigned char)__builtin_altivec_vsubeuqm(
+      (vector unsigned __int128)__a, (vector unsigned __int128)__b,
+      (vector unsigned __int128)__c);
+}
+
 /* vec_vsubcuq */
 
 static __inline__ vector signed __int128 __ATTRS_o_ai
@@ -11928,6 +11948,14 @@ vec_subec(vector unsigned __int128 __a, vector unsigned __int128 __b,
              vector unsigned __int128 __c) {
   return __builtin_altivec_vsubecuq(__a, __b, __c);
 }
+
+static __inline__ vector unsigned char __attribute__((__always_inline__))
+vec_subec_u128(vector unsigned char __a, vector unsigned char __b,
+               vector unsigned char __c) {
+  return (vector unsigned char)__builtin_altivec_vsubecuq(
+      (vector unsigned __int128)__a, (vector unsigned __int128)__b,
+      (vector unsigned __int128)__c);
+}
 #endif // defined(__POWER8_VECTOR__) && defined(__powerpc64__)
 
 static __inline__ vector signed int __ATTRS_o_ai
diff --git a/clang/test/CodeGen/builtins-ppc-quadword.c b/clang/test/CodeGen/builtins-ppc-quadword.c
index 66cc9e8..561f0c2 100644
--- a/clang/test/CodeGen/builtins-ppc-quadword.c
+++ b/clang/test/CodeGen/builtins-ppc-quadword.c
@@ -14,6 +14,7 @@
 vector signed __int128 vlll = { -1 };
 // CHECK-PPC: error: __int128 is not supported on this target
 vector unsigned __int128 vulll = { 1 };
+vector unsigned char vuc;
 
 signed long long param_sll;
 // CHECK-PPC: error: __int128 is not supported on this target
@@ -25,6 +26,7 @@ unsigned __int128 param_ulll;
 vector signed __int128 res_vlll;
 // CHECK-PPC: error: __int128 is not supported on this target
 vector unsigned __int128 res_vulll;
+vector unsigned char res_vuc;
 
 
 // CHECK-LABEL: define{{.*}} void @test1
@@ -119,6 +121,10 @@ void test1() {
 // CHECK-LE: sub <1 x i128> 
 // CHECK-PPC: error: assigning to '__vector unsigned __int128' (vector of 1 'unsigned __int128' value) from incompatible type 'int'
   
+  res_vuc = vec_sub_u128(vuc, vuc);
+// CHECK: sub <1 x i128>
+// CHECK-LE: sub <1 x i128>
+
   /* vec_vsubeuqm */
   res_vlll = vec_vsubeuqm(vlll, vlll, vlll);
 // CHECK: @llvm.ppc.altivec.vsubeuqm
@@ -151,6 +157,10 @@ void test1() {
 // CHECK-LE: @llvm.ppc.altivec.vsubeuqm
 // CHECK-PPC: error: call to 'vec_sube' is ambiguous
 
+  res_vuc = vec_sube_u128(vuc, vuc, vuc);
+// CHECK: @llvm.ppc.altivec.vsubeuqm
+// CHECK-LE: @llvm.ppc.altivec.vsubeuqm
+
   /* vec_subc */
   res_vlll = vec_subc(vlll, vlll);
 // CHECK: @llvm.ppc.altivec.vsubcuq
@@ -162,6 +172,10 @@ void test1() {
 // CHECK-LE: @llvm.ppc.altivec.vsubcuq
 // KCHECK-PPC: error: call to 'vec_subc' is ambiguous
 
+  res_vuc = vec_subc_u128(vuc, vuc);
+// CHECK: @llvm.ppc.altivec.vsubcuq
+// CHECK-LE: @llvm.ppc.altivec.vsubcuq
+
   /* vec_vsubcuq */
   res_vlll = vec_vsubcuq(vlll, vlll);
 // CHECK: @llvm.ppc.altivec.vsubcuq
@@ -194,6 +208,10 @@ void test1() {
 // CHECK-LE: @llvm.ppc.altivec.vsubecuq
 // CHECK-PPC: error: assigning to '__vector unsigned __int128' (vector of 1 'unsigned __int128' value) from incompatible type 'int'  
 
+  res_vuc = vec_subec_u128(vuc, vuc, vuc);
+// CHECK: @llvm.ppc.altivec.vsubecuq
+// CHECK-LE: @llvm.ppc.altivec.vsubecuq
+
   res_vulll = vec_revb(vulll);
 // CHECK: store <16 x i8> <i8 15, i8 14, i8 13, i8 12, i8 11, i8 10, i8 9, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>, <16 x i8>* {{%.+}}, align 16
 // CHECK: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> {{%.+}}, <4 x i32> {{%.+}}, <16 x i8> {{%.+}})
-- 
2.7.4