[X86] Allow _MM_FROUND_CUR_DIRECTION and _MM_FROUND_NO_EXC to be used together on instructions that only support SAE and not embedded rounding.

Current for SAE instructions we only allow _MM_FROUND_CUR_DIRECTION(bit 2) or _MM_FROUND_NO_EXC(bit 3) to be used as the immediate passed to the inrinsics. But these instructions don't perform rounding so _MM_FROUND_CUR_DIRECTION is just sort of a default placeholder when you don't want to suppress exceptions. Using _MM_FROUND_NO_EXC by itself is really bit equivalent to (_MM_FROUND_NO_EXC | _MM_FROUND_TO_NEAREST_INT) since _MM_FROUND_TO_NEAREST_INT is 0. Since we aren't rounding on these instructions we should also accept (_MM_FROUND_CUR_DIRECTION | _MM_FROUND_NO_EXC) as equivalent to (_MM_FROUND_NO_EXC). icc allows this, but gcc does not.

Differential Revision: https://reviews.llvm.org/D67289

llvm-svn: 371430

2 files changed, 16 insertions, 1 deletions

diff --git a/clang/lib/Sema/SemaChecking.cpp b/clang/lib/Sema/SemaChecking.cpp
index 7d17b0ecd49..c43f656c2f4 100644
--- a/clang/lib/Sema/SemaChecking.cpp
+++ b/clang/lib/Sema/SemaChecking.cpp
@@ -3546,9 +3546,11 @@ bool Sema::CheckX86BuiltinRoundingOrSAE(unsigned BuiltinID, CallExpr *TheCall) {
 
   // Make sure rounding mode is either ROUND_CUR_DIRECTION or ROUND_NO_EXC bit
   // is set. If the intrinsic has rounding control(bits 1:0), make sure its only
-  // combined with ROUND_NO_EXC.
+  // combined with ROUND_NO_EXC. If the intrinsic does not have rounding
+  // control, allow ROUND_NO_EXC and ROUND_CUR_DIRECTION together.
   if (Result == 4/*ROUND_CUR_DIRECTION*/ ||
       Result == 8/*ROUND_NO_EXC*/ ||
+      (!HasRC && Result == 12/*ROUND_CUR_DIRECTION|ROUND_NO_EXC*/) ||
       (HasRC && Result.getZExtValue() >= 8 && Result.getZExtValue() <= 11))
     return false;
 
diff --git a/clang/test/Sema/builtins-x86.c b/clang/test/Sema/builtins-x86.c
index 6a2a47d7792..dca0bdc720a 100644
--- a/clang/test/Sema/builtins-x86.c
+++ b/clang/test/Sema/builtins-x86.c
@@ -81,6 +81,19 @@ __mmask16 test__builtin_ia32_cmpps512_mask_rounding(__m512 __a, __m512 __b, __mm
   return __builtin_ia32_cmpps512_mask(__a, __b, 0, __u, 0); // expected-error {{invalid rounding argument}}
 }
 
+// Make sure we allow 4(CUR_DIRECTION), 8(NO_EXC), and 12(CUR_DIRECTION|NOEXC) for SAE arguments.
+__mmask16 test__builtin_ia32_cmpps512_mask_rounding_cur_dir(__m512 __a, __m512 __b, __mmask16 __u) {
+  return __builtin_ia32_cmpps512_mask(__a, __b, 0, __u, 4); // no-error
+}
+
+__mmask16 test__builtin_ia32_cmpps512_mask_rounding_sae1(__m512 __a, __m512 __b, __mmask16 __u) {
+  return __builtin_ia32_cmpps512_mask(__a, __b, 0, __u, 8); // no-error
+}
+
+__mmask16 test__builtin_ia32_cmpps512_mask_rounding_sae2(__m512 __a, __m512 __b, __mmask16 __u) {
+  return __builtin_ia32_cmpps512_mask(__a, __b, 0, __u, 12); // no-error
+}
+
 __m512 test__builtin_ia32_getmantps512_mask(__m512 a, __m512 b) {
   return __builtin_ia32_getmantps512_mask(a, 0, b, (__mmask16)-1, 10); // expected-error {{invalid rounding argument}}
 }