[PATCH] D28528: [PowerPC] [PowerPC] Fix the wrong implementation of builtin vec_rlnm.

[Tony Jiang via Phabricator via cfe-commits]

jtony created this revision.
jtony added reviewers: nemanjai, kbarton, sfertile, lei, syzaara.
jtony added subscribers: llvm-commits, cfe-commits, hfinkel, echristo.

vec_rlnm was implemented according to the old ABI, which was wrong. The ABI team have fixed the issue (although not published yet). We need to re-implement these builtins according to the new ABI.

From (old implementation):
__builtin_altivec_vrlwnm(__a, __b) & __c;

To (new implementation):
vector unsigned int OneByte = { 0x8, 0x8, 0x8, 0x8 };
__builtin_altivec_vrlwnm(a, ((c << OneByte) | b));


https://reviews.llvm.org/D28528

Files:
  lib/Headers/altivec.h
  test/CodeGen/builtins-ppc-p9vector.c


Index: test/CodeGen/builtins-ppc-p9vector.c
===================================================================
--- test/CodeGen/builtins-ppc-p9vector.c
+++ test/CodeGen/builtins-ppc-p9vector.c
@@ -868,20 +868,24 @@
   return vec_rlmi(vula, vula, vula);
 }
 vector unsigned int test77(void) {
+// CHECK-BE: %shl.i = shl <4 x i32
+// CHECK-BE: %or.i = or <4 x i32> %shl.i
 // CHECK-BE: @llvm.ppc.altivec.vrlwnm(<4 x i32
-// CHECK-BE: and <4 x i32
 // CHECK-BE: ret <4 x i32>
+// CHECK: %shl.i = shl <4 x i32
+// CHECK: %or.i = or <4 x i32> %shl.i
 // CHECK: @llvm.ppc.altivec.vrlwnm(<4 x i32
-// CHECK: and <4 x i32
 // CHECK: ret <4 x i32>
   return vec_rlnm(vuia, vuia, vuia);
 }
 vector unsigned long long test78(void) {
+// CHECK-BE: %shl.i = shl <2 x i64
+// CHECK-BE: %or.i = or <2 x i64> %shl.i
 // CHECK-BE: @llvm.ppc.altivec.vrldnm(<2 x i64
-// CHECK-BE: and <2 x i64
 // CHECK-BE-NEXT: ret <2 x i64>
+// CHECK: %shl.i = shl <2 x i64
+// CHECK: %or.i = or <2 x i64> %shl.i
 // CHECK: @llvm.ppc.altivec.vrldnm(<2 x i64
-// CHECK: and <2 x i64
 // CHECK-NEXT: ret <2 x i64>
   return vec_rlnm(vula, vula, vula);
 }
Index: lib/Headers/altivec.h
===================================================================
--- lib/Headers/altivec.h
+++ lib/Headers/altivec.h
@@ -7664,13 +7664,15 @@
 static __inline__ vector unsigned int __ATTRS_o_ai
 vec_rlnm(vector unsigned int __a, vector unsigned int __b,
          vector unsigned int __c) {
-  return __builtin_altivec_vrlwnm(__a, __b) & __c;
+  vector unsigned int OneByte = { 0x8, 0x8, 0x8, 0x8 };
+  return __builtin_altivec_vrlwnm(__a, ((__c << OneByte) | __b));
 }
 
 static __inline__ vector unsigned long long __ATTRS_o_ai
 vec_rlnm(vector unsigned long long __a, vector unsigned long long __b,
          vector unsigned long long __c) {
-  return __builtin_altivec_vrldnm(__a, __b) & __c;
+  vector unsigned long long OneByte = { 0x8, 0x8 };
+  return __builtin_altivec_vrldnm(__a, ((__c << OneByte) | __b));
 }
 #endif
 


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D28528.83846.patch
Type: text/x-patch
Size: 1982 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20170110/de7d89ea/attachment-0001.bin>