[Builtins] Implement half-precision conversions.

Mostly uninteresting, except:
- in __extendXfYf2, when checking if the number is normal, the old
  code relied on the unsignedness of src_rep_t, which is a problem
  when sizeof(src_rep_t) < sizeof(int): the result gets promoted to
  int, the signedness of which breaks the comparison.
  I added an explicit cast; it shouldn't affect other types.
- we can't pass __fp16, so src_t and src_rep_t are the same.
- the gnu_*_ieee symbols are simply duplicated definitions, as aliases
  are problematic on mach-o (where only weak aliases are supported;
  that's not what we want).

Differential Revision: http://reviews.llvm.org/D9693


git-svn-id: https://llvm.org/svn/llvm-project/compiler-rt/trunk@237161 91177308-0d34-0410-b5e6-96231b3b80d8

7 files changed, 92 insertions, 5 deletions

diff --git a/lib/builtins/CMakeLists.txt b/lib/builtins/CMakeLists.txt
index 4d102c615..d79feae13 100644
--- a/lib/builtins/CMakeLists.txt
+++ b/lib/builtins/CMakeLists.txt
@@ -44,6 +44,7 @@ set(GENERIC_SOURCES
   enable_execute_stack.c
   eprintf.c
   extendsfdf2.c
+  extendhfsf2.c
   ffsdi2.c
   ffsti2.c
   fixdfdi.c
@@ -123,7 +124,9 @@ set(GENERIC_SOURCES
   subvti3.c
   subtf3.c
   trampoline_setup.c
+  truncdfhf2.c
   truncdfsf2.c
+  truncsfhf2.c
   ucmpdi2.c
   ucmpti2.c
   udivdi3.c
diff --git a/lib/builtins/extendhfsf2.c b/lib/builtins/extendhfsf2.c
new file mode 100644
index 000000000..f0b3c37c3
--- /dev/null
+++ b/lib/builtins/extendhfsf2.c
@@ -0,0 +1,21 @@
+//===-- lib/extendhfsf2.c - half -> single conversion -------------*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+
+#define SRC_HALF
+#define DST_SINGLE
+#include "fp_extend_impl.inc"
+
+COMPILER_RT_ABI float __extendhfsf2(uint16_t a) {
+    return __extendXfYf2__(a);
+}
+
+COMPILER_RT_ABI float __gnu_h2f_ieee(uint16_t a) {
+    return __extendXfYf2__(a);
+}
diff --git a/lib/builtins/fp_extend.h b/lib/builtins/fp_extend.h
index fff676eec..5c2b92310 100644
--- a/lib/builtins/fp_extend.h
+++ b/lib/builtins/fp_extend.h
@@ -39,11 +39,24 @@ static inline int src_rep_t_clz(src_rep_t a) {
 #endif
 }
 
+#elif defined SRC_HALF
+typedef uint16_t src_t;
+typedef uint16_t src_rep_t;
+#define SRC_REP_C UINT16_C
+static const int srcSigBits = 10;
+#define src_rep_t_clz __builtin_clz
+
 #else
-#error Source should be single precision or double precision!
+#error Source should be half, single, or double precision!
 #endif //end source precision
 
-#if defined DST_DOUBLE
+#if defined DST_SINGLE
+typedef float dst_t;
+typedef uint32_t dst_rep_t;
+#define DST_REP_C UINT32_C
+static const int dstSigBits = 23;
+
+#elif defined DST_DOUBLE
 typedef double dst_t;
 typedef uint64_t dst_rep_t;
 #define DST_REP_C UINT64_C
@@ -56,7 +69,7 @@ typedef __uint128_t dst_rep_t;
 static const int dstSigBits = 112;
 
 #else
-#error Destination should be double precision or quad precision!
+#error Destination should be single, double, or quad precision!
 #endif //end destination precision
 
 // End of specialization parameters.  Two helper routines for conversion to and
diff --git a/lib/builtins/fp_extend_impl.inc b/lib/builtins/fp_extend_impl.inc
index f6953ff4b..edcfa8d23 100644
--- a/lib/builtins/fp_extend_impl.inc
+++ b/lib/builtins/fp_extend_impl.inc
@@ -66,7 +66,9 @@ static inline dst_t __extendXfYf2__(src_t a) {
     const src_rep_t sign = aRep & srcSignMask;
     dst_rep_t absResult;
 
-    if (aAbs - srcMinNormal < srcInfinity - srcMinNormal) {
+    // If sizeof(src_rep_t) < sizeof(int), the subtraction result is promoted
+    // to (signed) int.  To avoid that, explicitly cast to src_rep_t.
+    if ((src_rep_t)(aAbs - srcMinNormal) < srcInfinity - srcMinNormal) {
         // a is a normal number.
         // Extend to the destination type by shifting the significand and
         // exponent into the proper position and rebiasing the exponent.
diff --git a/lib/builtins/fp_trunc.h b/lib/builtins/fp_trunc.h
index 49a9aebbc..373ba1b04 100644
--- a/lib/builtins/fp_trunc.h
+++ b/lib/builtins/fp_trunc.h
@@ -16,7 +16,13 @@
 
 #include "int_lib.h"
 
-#if defined SRC_DOUBLE
+#if defined SRC_SINGLE
+typedef float src_t;
+typedef uint32_t src_rep_t;
+#define SRC_REP_C UINT32_C
+static const int srcSigBits = 23;
+
+#elif defined SRC_DOUBLE
 typedef double src_t;
 typedef uint64_t src_rep_t;
 #define SRC_REP_C UINT64_C
@@ -44,6 +50,12 @@ typedef uint32_t dst_rep_t;
 #define DST_REP_C UINT32_C
 static const int dstSigBits = 23;
 
+#elif defined DST_HALF
+typedef uint16_t dst_t;
+typedef uint16_t dst_rep_t;
+#define DST_REP_C UINT16_C
+static const int dstSigBits = 10;
+
 #else
 #error Destination should be single precision or double precision!
 #endif //end destination precision
diff --git a/lib/builtins/truncdfhf2.c b/lib/builtins/truncdfhf2.c
new file mode 100644
index 000000000..0852df369
--- /dev/null
+++ b/lib/builtins/truncdfhf2.c
@@ -0,0 +1,16 @@
+//===-- lib/truncdfhf2.c - double -> half conversion --------------*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#define SRC_DOUBLE
+#define DST_HALF
+#include "fp_trunc_impl.inc"
+
+COMPILER_RT_ABI uint16_t __truncdfhf2(double a) {
+    return __truncXfYf2__(a);
+}
diff --git a/lib/builtins/truncsfhf2.c b/lib/builtins/truncsfhf2.c
new file mode 100644
index 000000000..a25dd83d8
--- /dev/null
+++ b/lib/builtins/truncsfhf2.c
@@ -0,0 +1,20 @@
+//===-- lib/truncsfhf2.c - single -> half conversion --------------*- C -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#define SRC_SINGLE
+#define DST_HALF
+#include "fp_trunc_impl.inc"
+
+COMPILER_RT_ABI uint16_t __truncsfhf2(float a) {
+    return __truncXfYf2__(a);
+}
+
+COMPILER_RT_ABI uint16_t __gnu_f2h_ieee(float a) {
+    return __truncXfYf2__(a);
+}