[Libclc-dev] [PATCH 11/15] R600: Map addr spaces and use atomic_max

[Aaron Watry]

Signed-off-by: Aaron Watry <awatry at gmail.com>
---
 r600/lib/atomic/atomic.cl | 21 ++++++++++++++++-----
 1 file changed, 16 insertions(+), 5 deletions(-)

diff --git a/r600/lib/atomic/atomic.cl b/r600/lib/atomic/atomic.cl
index 0eba779..d38bf81 100644
--- a/r600/lib/atomic/atomic.cl
+++ b/r600/lib/atomic/atomic.cl
@@ -1,14 +1,15 @@
 #include <clc/clc.h>
 
-#define ATOMIC_FUNC_TYPE(SIGN, TYPE, FUNCTION, CL_ADDRSPACE, LLVM_ADDRSPACE) \
-_CLC_OVERLOAD _CLC_DEF SIGN TYPE FUNCTION (volatile CL_ADDRSPACE SIGN TYPE *p, SIGN TYPE val) { \
-	return (SIGN TYPE)__clc_##FUNCTION##_addr##LLVM_ADDRSPACE((volatile CL_ADDRSPACE signed TYPE*)p, (signed TYPE)val); \
+#define ATOMIC_FUNC_DEFINE(RET_SIGN, ARG_SIGN, TYPE, CL_FUNCTION, CLC_FUNCTION, CL_ADDRSPACE, LLVM_ADDRSPACE) \
+_CLC_OVERLOAD _CLC_DEF RET_SIGN TYPE CL_FUNCTION (volatile CL_ADDRSPACE RET_SIGN TYPE *p, RET_SIGN TYPE val) { \
+	return (RET_SIGN TYPE)__clc_##CLC_FUNCTION##_addr##LLVM_ADDRSPACE((volatile CL_ADDRSPACE ARG_SIGN TYPE*)p, (ARG_SIGN TYPE)val); \
 }
 
+/* For atomic functions that don't need different bitcode dependending on argument signedness */
 #define ATOMIC_FUNC_SIGN(TYPE, FUNCTION, CL_ADDRSPACE, LLVM_ADDRSPACE) \
 	_CLC_DECL signed TYPE __clc_##FUNCTION##_addr##LLVM_ADDRSPACE(volatile CL_ADDRSPACE signed TYPE*, signed TYPE); \
-	ATOMIC_FUNC_TYPE(signed, TYPE, FUNCTION, CL_ADDRSPACE, LLVM_ADDRSPACE) \
-	ATOMIC_FUNC_TYPE(unsigned, TYPE, FUNCTION, CL_ADDRSPACE, LLVM_ADDRSPACE)
+	ATOMIC_FUNC_DEFINE(signed, signed, TYPE, FUNCTION, FUNCTION, CL_ADDRSPACE, LLVM_ADDRSPACE) \
+	ATOMIC_FUNC_DEFINE(unsigned, signed, TYPE, FUNCTION, FUNCTION, CL_ADDRSPACE, LLVM_ADDRSPACE)
 
 #define ATOMIC_FUNC_ADDRSPACE(TYPE, FUNCTION) \
 	ATOMIC_FUNC_SIGN(TYPE, FUNCTION, global, 1) \
@@ -21,3 +22,13 @@ ATOMIC_FUNC(atomic_add)
 ATOMIC_FUNC(atomic_and)
 ATOMIC_FUNC(atomic_or)
 ATOMIC_FUNC(atomic_sub)
+
+_CLC_DECL signed int __clc_atomic_max_addr1(volatile global signed int*, signed int);
+_CLC_DECL signed int __clc_atomic_max_addr3(volatile local signed int*, signed int);
+_CLC_DECL uint __clc_atomic_umax_addr1(volatile global uint*, uint);
+_CLC_DECL uint __clc_atomic_umax_addr3(volatile local uint*, uint);
+
+ATOMIC_FUNC_DEFINE(signed, signed, int, atomic_max, atomic_max, global, 1)
+ATOMIC_FUNC_DEFINE(signed, signed, int, atomic_max, atomic_max, local, 3)
+ATOMIC_FUNC_DEFINE(unsigned, unsigned, int, atomic_max, atomic_umax, global, 1)
+ATOMIC_FUNC_DEFINE(unsigned, unsigned, int, atomic_max, atomic_umax, local, 3)
-- 
1.9.1