Fix and re-enable R600 vload/vstore assembly

The assembly optimizations were making unsafe assumptions about which address
spaces had which identifiers.

Also, fix vload/vstore with 64-bit pointers. This was broken previously on
Radeon SI.

This version still only has assembly versions of int/uint 2/4/8/16 for global
loads and stores on R600, but it does it in a way that would be very easily
extended to private/local/constant and could also be handled easily on other
architectures.

v2: 1) Leave v[load|store]_impl.ll in generic/lib
    2) Remove vload_if.ll and vstore_if.ll interfaces
    3) Fix address+offset calculations
    3) Remove offset from assembly arg list

git-svn-id: https://llvm.org/svn/llvm-project/libclc/trunk@186416 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 198 insertions, 0 deletions

diff --git a/r600/lib/SOURCES b/r600/lib/SOURCES
index 16ef3ac..87df0b7 100644
--- a/r600/lib/SOURCES
+++ b/r600/lib/SOURCES
@@ -4,3 +4,5 @@ workitem/get_local_id.ll
 workitem/get_global_size.ll
 synchronization/barrier.cl
 synchronization/barrier_impl.ll
+shared/vload.cl
+shared/vstore.cl
\ No newline at end of file
diff --git a/r600/lib/shared/vload.cl b/r600/lib/shared/vload.cl
new file mode 100644
index 0000000..6144dde
--- /dev/null
+++ b/r600/lib/shared/vload.cl
@@ -0,0 +1,92 @@
+#include <clc/clc.h>
+
+#define VLOAD_VECTORIZE(PRIM_TYPE, ADDR_SPACE) \
+  _CLC_OVERLOAD _CLC_DEF PRIM_TYPE##2 vload2(size_t offset, const ADDR_SPACE PRIM_TYPE *x) { \
+    return (PRIM_TYPE##2)(x[2*offset] , x[2*offset+1]); \
+  } \
+\
+  _CLC_OVERLOAD _CLC_DEF PRIM_TYPE##3 vload3(size_t offset, const ADDR_SPACE PRIM_TYPE *x) { \
+    return (PRIM_TYPE##3)(x[3*offset] , x[3*offset+1], x[3*offset+2]); \
+  } \
+\
+  _CLC_OVERLOAD _CLC_DEF PRIM_TYPE##4 vload4(size_t offset, const ADDR_SPACE PRIM_TYPE *x) { \
+    return (PRIM_TYPE##4)(x[4*offset], x[4*offset+1], x[4*offset+2], x[4*offset+3]); \
+  } \
+\
+  _CLC_OVERLOAD _CLC_DEF PRIM_TYPE##8 vload8(size_t offset, const ADDR_SPACE PRIM_TYPE *x) { \
+    return (PRIM_TYPE##8)(vload4(0, &x[8*offset]), vload4(1, &x[8*offset])); \
+  } \
+\
+  _CLC_OVERLOAD _CLC_DEF PRIM_TYPE##16 vload16(size_t offset, const ADDR_SPACE PRIM_TYPE *x) { \
+    return (PRIM_TYPE##16)(vload8(0, &x[16*offset]), vload8(1, &x[16*offset])); \
+  } \
+
+#define VLOAD_ADDR_SPACES(SCALAR_GENTYPE) \
+    VLOAD_VECTORIZE(SCALAR_GENTYPE, __private) \
+    VLOAD_VECTORIZE(SCALAR_GENTYPE, __local) \
+    VLOAD_VECTORIZE(SCALAR_GENTYPE, __constant) \
+    VLOAD_VECTORIZE(SCALAR_GENTYPE, __global) \
+
+//int/uint are special... see below
+#define VLOAD_TYPES() \
+    VLOAD_ADDR_SPACES(char) \
+    VLOAD_ADDR_SPACES(uchar) \
+    VLOAD_ADDR_SPACES(short) \
+    VLOAD_ADDR_SPACES(ushort) \
+    VLOAD_ADDR_SPACES(long) \
+    VLOAD_ADDR_SPACES(ulong) \
+    VLOAD_ADDR_SPACES(float) \
+
+VLOAD_TYPES()
+
+#ifdef cl_khr_fp64
+#pragma OPENCL EXTENSION cl_khr_fp64 : enable
+    VLOAD_ADDR_SPACES(double)
+#endif
+
+//Assembly overrides start here
+
+VLOAD_VECTORIZE(int, __private)
+VLOAD_VECTORIZE(int, __local)
+VLOAD_VECTORIZE(int, __constant)
+VLOAD_VECTORIZE(uint, __private)
+VLOAD_VECTORIZE(uint, __local)
+VLOAD_VECTORIZE(uint, __constant)
+
+_CLC_OVERLOAD _CLC_DEF int3 vload3(size_t offset, const global int *x) {
+  return (int3)(vload2(0, &x[3*offset]), x[3*offset+2]);
+}
+_CLC_OVERLOAD _CLC_DEF uint3 vload3(size_t offset, const global uint *x) {
+  return (uint3)(vload2(0, &x[3*offset]), x[3*offset+2]);
+}
+
+//We only define functions for typeN vloadN(), and then just bitcast the result for unsigned types
+#define _CLC_VLOAD_ASM_DECL(PRIM_TYPE,LLVM_SCALAR_TYPE,ADDR_SPACE,ADDR_SPACE_ID) \
+_CLC_DECL PRIM_TYPE##2 __clc_vload2_##LLVM_SCALAR_TYPE##__addr##ADDR_SPACE_ID (const ADDR_SPACE PRIM_TYPE *); \
+_CLC_DECL PRIM_TYPE##4 __clc_vload4_##LLVM_SCALAR_TYPE##__addr##ADDR_SPACE_ID (const ADDR_SPACE PRIM_TYPE *); \
+_CLC_DECL PRIM_TYPE##8 __clc_vload8_##LLVM_SCALAR_TYPE##__addr##ADDR_SPACE_ID (const ADDR_SPACE PRIM_TYPE *); \
+_CLC_DECL PRIM_TYPE##16 __clc_vload16_##LLVM_SCALAR_TYPE##__addr##ADDR_SPACE_ID (const ADDR_SPACE PRIM_TYPE *); \
+
+#define _CLC_VLOAD_ASM_DEFINE(PRIM_TYPE,S_PRIM_TYPE, LLVM_SCALAR_TYPE,VEC_WIDTH,ADDR_SPACE,ADDR_SPACE_ID) \
+  _CLC_OVERLOAD _CLC_DEF PRIM_TYPE##VEC_WIDTH vload##VEC_WIDTH (size_t offset, const ADDR_SPACE PRIM_TYPE *x) { \
+    return __builtin_astype(__clc_vload##VEC_WIDTH##_##LLVM_SCALAR_TYPE##__addr##ADDR_SPACE_ID ((const ADDR_SPACE S_PRIM_TYPE *)&x[VEC_WIDTH * offset]), PRIM_TYPE##VEC_WIDTH); \
+  } \
+
+/*Note: R600 back-end doesn't support load <3 x ?>... so
+ * those functions aren't actually overridden here
+ */
+#define _CLC_VLOAD_ASM_OVERLOAD_SIZES(PRIM_TYPE,S_PRIM_TYPE,LLVM_TYPE,ADDR_SPACE,ADDR_SPACE_ID) \
+  _CLC_VLOAD_ASM_DEFINE(PRIM_TYPE, S_PRIM_TYPE, LLVM_TYPE, 2, ADDR_SPACE, ADDR_SPACE_ID) \
+  _CLC_VLOAD_ASM_DEFINE(PRIM_TYPE, S_PRIM_TYPE, LLVM_TYPE, 4, ADDR_SPACE, ADDR_SPACE_ID) \
+  _CLC_VLOAD_ASM_DEFINE(PRIM_TYPE, S_PRIM_TYPE, LLVM_TYPE, 8, ADDR_SPACE, ADDR_SPACE_ID) \
+  _CLC_VLOAD_ASM_DEFINE(PRIM_TYPE, S_PRIM_TYPE, LLVM_TYPE, 16, ADDR_SPACE, ADDR_SPACE_ID) \
+
+#define _CLC_VLOAD_ASM_OVERLOAD_ADDR_SPACES(PRIM_TYPE,S_PRIM_TYPE,LLVM_TYPE) \
+  _CLC_VLOAD_ASM_OVERLOAD_SIZES(PRIM_TYPE, S_PRIM_TYPE, LLVM_TYPE, global, 1) \
+
+#define _CLC_VLOAD_ASM_OVERLOADS() \
+  _CLC_VLOAD_ASM_DECL(int,i32,__global,1) \
+  _CLC_VLOAD_ASM_OVERLOAD_ADDR_SPACES(int,int,i32) \
+  _CLC_VLOAD_ASM_OVERLOAD_ADDR_SPACES(uint,int,i32) \
+
+_CLC_VLOAD_ASM_OVERLOADS()
\ No newline at end of file
diff --git a/r600/lib/shared/vstore.cl b/r600/lib/shared/vstore.cl
new file mode 100644
index 0000000..a150849
--- /dev/null
+++ b/r600/lib/shared/vstore.cl
@@ -0,0 +1,104 @@
+#include <clc/clc.h>
+
+#pragma OPENCL EXTENSION cl_khr_byte_addressable_store : enable
+
+#define VSTORE_VECTORIZE(PRIM_TYPE, ADDR_SPACE) \
+  _CLC_OVERLOAD _CLC_DEF void vstore2(PRIM_TYPE##2 vec, size_t offset, ADDR_SPACE PRIM_TYPE *mem) { \
+    mem[2*offset] = vec.s0; \
+    mem[2*offset+1] = vec.s1; \
+  } \
+\
+  _CLC_OVERLOAD _CLC_DEF void vstore3(PRIM_TYPE##3 vec, size_t offset, ADDR_SPACE PRIM_TYPE *mem) { \
+    mem[3*offset] = vec.s0; \
+    mem[3*offset+1] = vec.s1; \
+    mem[3*offset+2] = vec.s2; \
+  } \
+\
+  _CLC_OVERLOAD _CLC_DEF void vstore4(PRIM_TYPE##4 vec, size_t offset, ADDR_SPACE PRIM_TYPE *mem) { \
+    vstore2(vec.lo, 0, &mem[offset*4]); \
+    vstore2(vec.hi, 1, &mem[offset*4]); \
+  } \
+\
+  _CLC_OVERLOAD _CLC_DEF void vstore8(PRIM_TYPE##8 vec, size_t offset, ADDR_SPACE PRIM_TYPE *mem) { \
+    vstore4(vec.lo, 0, &mem[offset*8]); \
+    vstore4(vec.hi, 1, &mem[offset*8]); \
+  } \
+\
+  _CLC_OVERLOAD _CLC_DEF void vstore16(PRIM_TYPE##16 vec, size_t offset, ADDR_SPACE PRIM_TYPE *mem) { \
+    vstore8(vec.lo, 0, &mem[offset*16]); \
+    vstore8(vec.hi, 1, &mem[offset*16]); \
+  } \
+
+#define VSTORE_ADDR_SPACES(SCALAR_GENTYPE) \
+    VSTORE_VECTORIZE(SCALAR_GENTYPE, __private) \
+    VSTORE_VECTORIZE(SCALAR_GENTYPE, __local) \
+    VSTORE_VECTORIZE(SCALAR_GENTYPE, __global) \
+
+//int/uint are special... see below
+#define VSTORE_TYPES() \
+    VSTORE_ADDR_SPACES(char) \
+    VSTORE_ADDR_SPACES(uchar) \
+    VSTORE_ADDR_SPACES(short) \
+    VSTORE_ADDR_SPACES(ushort) \
+    VSTORE_ADDR_SPACES(long) \
+    VSTORE_ADDR_SPACES(ulong) \
+    VSTORE_ADDR_SPACES(float) \
+
+VSTORE_TYPES()
+
+#ifdef cl_khr_fp64
+#pragma OPENCL EXTENSION cl_khr_fp64 : enable
+    VSTORE_ADDR_SPACES(double)
+#endif
+
+VSTORE_VECTORIZE(int, __private)
+VSTORE_VECTORIZE(int, __local)
+VSTORE_VECTORIZE(uint, __private)
+VSTORE_VECTORIZE(uint, __local)
+
+_CLC_OVERLOAD _CLC_DEF void vstore3(int3 vec, size_t offset, global int *mem) {
+    mem[3*offset] = vec.s0;
+    mem[3*offset+1] = vec.s1;
+    mem[3*offset+2] = vec.s2;
+}
+_CLC_OVERLOAD _CLC_DEF void vstore3(uint3 vec, size_t offset, global uint *mem) {
+    mem[3*offset] = vec.s0;
+    mem[3*offset+1] = vec.s1;
+    mem[3*offset+2] = vec.s2;
+}
+
+/*Note: R600 doesn't support store <3 x ?>... so
+ * those functions aren't actually overridden here... lowest-common-denominator
+ */
+
+//We only define functions for signed_type vstoreN(), and then just cast the pointers/vectors for unsigned types
+#define _CLC_VSTORE_ASM_DECL(PRIM_TYPE,LLVM_SCALAR_TYPE,ADDR_SPACE,ADDR_SPACE_ID) \
+_CLC_DECL void __clc_vstore2_##LLVM_SCALAR_TYPE##__addr##ADDR_SPACE_ID (PRIM_TYPE##2, ADDR_SPACE PRIM_TYPE *); \
+_CLC_DECL void __clc_vstore4_##LLVM_SCALAR_TYPE##__addr##ADDR_SPACE_ID (PRIM_TYPE##4, ADDR_SPACE PRIM_TYPE *); \
+_CLC_DECL void __clc_vstore8_##LLVM_SCALAR_TYPE##__addr##ADDR_SPACE_ID (PRIM_TYPE##8, ADDR_SPACE PRIM_TYPE *); \
+_CLC_DECL void __clc_vstore16_##LLVM_SCALAR_TYPE##__addr##ADDR_SPACE_ID (PRIM_TYPE##16, ADDR_SPACE PRIM_TYPE *); \
+
+#define _CLC_VSTORE_ASM_DEFINE(PRIM_TYPE, S_PRIM_TYPE, LLVM_SCALAR_TYPE, VEC_WIDTH, ADDR_SPACE, ADDR_SPACE_ID) \
+  _CLC_OVERLOAD _CLC_DEF void vstore##VEC_WIDTH(PRIM_TYPE##VEC_WIDTH vec, size_t offset, ADDR_SPACE PRIM_TYPE *x) { \
+    __clc_vstore##VEC_WIDTH##_##LLVM_SCALAR_TYPE##__addr##ADDR_SPACE_ID (__builtin_astype(vec, S_PRIM_TYPE##VEC_WIDTH), (ADDR_SPACE S_PRIM_TYPE *)&x[ VEC_WIDTH * offset]); \
+  } \
+
+/*Note: R600 back-end doesn't support load <3 x ?>... so
+ * those functions aren't actually overridden here... When the back-end supports
+ * that, then clean add here, and remove the vstore3 definitions from above.
+ */
+#define _CLC_VSTORE_ASM_OVERLOAD_SIZES(PRIM_TYPE,S_PRIM_TYPE,LLVM_TYPE,ADDR_SPACE,ADDR_SPACE_ID) \
+  _CLC_VSTORE_ASM_DEFINE(PRIM_TYPE, S_PRIM_TYPE, LLVM_TYPE, 2, ADDR_SPACE, ADDR_SPACE_ID) \
+  _CLC_VSTORE_ASM_DEFINE(PRIM_TYPE, S_PRIM_TYPE, LLVM_TYPE, 4, ADDR_SPACE, ADDR_SPACE_ID) \
+  _CLC_VSTORE_ASM_DEFINE(PRIM_TYPE, S_PRIM_TYPE, LLVM_TYPE, 8, ADDR_SPACE, ADDR_SPACE_ID) \
+  _CLC_VSTORE_ASM_DEFINE(PRIM_TYPE, S_PRIM_TYPE, LLVM_TYPE, 16, ADDR_SPACE, ADDR_SPACE_ID) \
+
+#define _CLC_VSTORE_ASM_OVERLOAD_ADDR_SPACES(PRIM_TYPE,S_PRIM_TYPE,LLVM_TYPE) \
+  _CLC_VSTORE_ASM_OVERLOAD_SIZES(PRIM_TYPE, S_PRIM_TYPE, LLVM_TYPE, global, 1) \
+
+#define _CLC_VSTORE_ASM_OVERLOADS() \
+  _CLC_VSTORE_ASM_DECL(int,i32,__global,1) \
+  _CLC_VSTORE_ASM_OVERLOAD_ADDR_SPACES(int,int,i32) \
+  _CLC_VSTORE_ASM_OVERLOAD_ADDR_SPACES(uint,int,i32) \
+
+_CLC_VSTORE_ASM_OVERLOADS()
\ No newline at end of file