[libclc] Move vload & vstore to CLC library

libclc/clc/include/clc/shared/clc_less_aligned_types.h

@@ -0,0 +1,23 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// Defines types to be used with (CLC) vstore and vload functions. These are
+// vector types whose alignment is that of their respective scalar types.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef __CLC_SHARED_CLC_LESS_ALIGNED_TYPES_H__
+#define __CLC_SHARED_CLC_LESS_ALIGNED_TYPES_H__
+
+#define __CLC_BODY <clc/shared/clc_less_aligned_types.inc>
+#include <clc/integer/gentype.inc>
+
+#define __CLC_BODY <clc/shared/clc_less_aligned_types.inc>
+#include <clc/math/gentype.inc>
+
+#endif // __CLC_SHARED_CLC_LESS_ALIGNED_TYPES_H__

libclc/clc/include/clc/shared/clc_less_aligned_types.inc

@@ -0,0 +1,23 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// Defines types to be used with (CLC) vstore and vload functions. These are
+// vector types whose alignment is that of their respective scalar types.
+//
+//===----------------------------------------------------------------------===//
+
+#ifdef __CLC_SCALAR
+
+typedef __CLC_GENTYPE __CLC_XCONCAT(less_aligned_, __CLC_GENTYPE);
+
+#else
+
+typedef __CLC_GENTYPE __CLC_XCONCAT(less_aligned_, __CLC_GENTYPE)
+    __attribute__((aligned(sizeof(__CLC_SCALAR_GENTYPE))));
+
+#endif

libclc/clc/include/clc/shared/clc_vload.h

@@ -0,0 +1,20 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef __CLC_SHARED_CLC_VLOAD_H__
+#define __CLC_SHARED_CLC_VLOAD_H__
+
+#include <clc/shared/clc_less_aligned_types.h>
+
+#define __CLC_BODY <clc/shared/clc_vload.inc>
+#include <clc/integer/gentype.inc>
+
+#define __CLC_BODY <clc/shared/clc_vload.inc>
+#include <clc/math/gentype.inc>
+
+#endif // __CLC_SHARED_CLC_VLOAD_H__

libclc/clc/include/clc/shared/clc_vload.inc

@@ -0,0 +1,64 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#define CLC_VLOAD_NAME __CLC_XCONCAT(__clc_vload, __CLC_VECSIZE)
+#define CLC_VLOAD_HALF_NAME __CLC_XCONCAT(__clc_vload_half, __CLC_VECSIZE)
+#define CLC_VLOADA_HALF_NAME __CLC_XCONCAT(__clc_vloada_half, __CLC_VECSIZE)
+
+#ifndef __CLC_SCALAR
+
+#define CLC_VLOAD_TY __CLC_XCONCAT(less_aligned_, __CLC_GENTYPE)
+
+#define CLC_VLOAD_DECL(ADDRSPACE)                                              \
+  _CLC_OVERLOAD _CLC_DECL CLC_VLOAD_TY CLC_VLOAD_NAME(                         \
+      size_t offset, const ADDRSPACE __CLC_SCALAR_GENTYPE *x);
+
+CLC_VLOAD_DECL(__private)
+CLC_VLOAD_DECL(__local)
+CLC_VLOAD_DECL(__constant)
+CLC_VLOAD_DECL(__global)
+
+#if _CLC_DISTINCT_GENERIC_AS_SUPPORTED
+CLC_VLOAD_DECL(__generic)
+#endif
+
+#undef CLC_VLOAD_DECL
+#undef CLC_VLOAD_TY
+
+#endif // __CLC_SCALAR
+
+// vload_half and vloada_half are available even if cl_khr_fp16 is unavailable.
+// Declare these functions when working on float types, which we know are
+// always available.
+#ifdef __CLC_FPSIZE
+#if __CLC_FPSIZE == 32
+
+#define CLC_VLOAD_HALF_DECL(ADDRSPACE)                                         \
+  _CLC_OVERLOAD _CLC_DECL __CLC_GENTYPE CLC_VLOAD_HALF_NAME(                   \
+      size_t offset, const ADDRSPACE half *mem);                               \
+                                                                               \
+  _CLC_OVERLOAD _CLC_DECL __CLC_GENTYPE CLC_VLOADA_HALF_NAME(                  \
+      size_t offset, const ADDRSPACE half *mem);
+
+CLC_VLOAD_HALF_DECL(__private)
+CLC_VLOAD_HALF_DECL(__local)
+CLC_VLOAD_HALF_DECL(__constant)
+CLC_VLOAD_HALF_DECL(__global)
+
+#if _CLC_DISTINCT_GENERIC_AS_SUPPORTED
+CLC_VLOAD_HALF_DECL(__generic)
+#endif
+
+#undef CLC_VLOAD_HALF_DECL
+
+#endif
+#endif
+
+#undef CLC_VLOAD_NAME
+#undef CLC_VLOAD_HALF_NAME
+#undef CLC_VLOADA_HALF_NAME

libclc/clc/include/clc/shared/clc_vstore.h

@@ -0,0 +1,20 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef __CLC_SHARED_CLC_VSTORE_H__
+#define __CLC_SHARED_CLC_VSTORE_H__
+
+#include <clc/shared/clc_less_aligned_types.h>
+
+#define __CLC_BODY <clc/shared/clc_vstore.inc>
+#include <clc/integer/gentype.inc>
+
+#define __CLC_BODY <clc/shared/clc_vstore.inc>
+#include <clc/math/gentype.inc>
+
+#endif // __CLC_SHARED_CLC_VSTORE_H__

libclc/clc/include/clc/shared/clc_vstore.inc

@@ -0,0 +1,70 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#define CLC_VSTORE_TY __CLC_XCONCAT(less_aligned_, __CLC_GENTYPE)
+#define CLC_VSTORE_NAME __CLC_XCONCAT(__clc_vstore, __CLC_VECSIZE)
+#define CLC_VSTORE_HALF_NAME(x)                                                \
+  __CLC_XCONCAT(__CLC_XCONCAT(__clc_vstore_half, __CLC_VECSIZE), x)
+#define CLC_VSTOREA_HALF_NAME(x)                                               \
+  __CLC_XCONCAT(__CLC_XCONCAT(__clc_vstorea_half, __CLC_VECSIZE), x)
+
+#ifndef __CLC_SCALAR
+
+#define CLC_VSTORE_DECL(ADDRSPACE)                                             \
+  _CLC_OVERLOAD _CLC_DECL void CLC_VSTORE_NAME(                                \
+      CLC_VSTORE_TY data, size_t offset, ADDRSPACE __CLC_SCALAR_GENTYPE *p);
+
+CLC_VSTORE_DECL(__private)
+CLC_VSTORE_DECL(__local)
+CLC_VSTORE_DECL(__global)
+
+#if _CLC_DISTINCT_GENERIC_AS_SUPPORTED
+CLC_VSTORE_DECL(__generic)
+#endif
+
+#undef CLC_VSTORE_DECL
+
+#endif // __CLC_SCALAR
+
+// vstore_half and vstorea_half are available even if cl_khr_fp16 is
+// unavailable.
+#ifdef __CLC_FPSIZE
+#if __CLC_FPSIZE == 32 || __CLC_FPSIZE == 64
+
+#define CLC_VSTORE_HALF_DECL(ADDRSPACE, SUFFIX)                                \
+  _CLC_OVERLOAD _CLC_DECL void CLC_VSTORE_HALF_NAME(SUFFIX)(                   \
+      CLC_VSTORE_TY data, size_t offset, ADDRSPACE half *p);                   \
+                                                                               \
+  _CLC_OVERLOAD _CLC_DECL void CLC_VSTOREA_HALF_NAME(SUFFIX)(                  \
+      CLC_VSTORE_TY data, size_t offset, ADDRSPACE half *p);
+
+#define CLC_VSTORE_HALF_DECL_ALL_MODES(ADDRSPACE)                              \
+  CLC_VSTORE_HALF_DECL(ADDRSPACE, )                                            \
+  CLC_VSTORE_HALF_DECL(ADDRSPACE, _rtz)                                        \
+  CLC_VSTORE_HALF_DECL(ADDRSPACE, _rtn)                                        \
+  CLC_VSTORE_HALF_DECL(ADDRSPACE, _rtp)                                        \
+  CLC_VSTORE_HALF_DECL(ADDRSPACE, _rte)
+
+CLC_VSTORE_HALF_DECL_ALL_MODES(__private)
+CLC_VSTORE_HALF_DECL_ALL_MODES(__local)
+CLC_VSTORE_HALF_DECL_ALL_MODES(__global)
+
+#if _CLC_DISTINCT_GENERIC_AS_SUPPORTED
+CLC_VSTORE_HALF_DECL_ALL_MODES(__generic)
+#endif
+
+#undef CLC_VSTORE_HALF_DECL
+#undef CLC_VSTORE_HALF_DECL_ALL_MODES
+
+#endif
+#endif
+
+#undef CLC_VSTORE_TY
+#undef CLC_VSTORE_NAME
+#undef CLC_VSTORE_HALF_NAME
+#undef CLC_VSTOREA_HALF_NAME

libclc/clc/lib/generic/SOURCES

@@ -147,3 +147,5 @@ relational/clc_signbit.cl
 shared/clc_clamp.cl
 shared/clc_max.cl
 shared/clc_min.cl
+shared/clc_vload.cl
+shared/clc_vstore.cl

libclc/clc/lib/generic/shared/clc_vload.cl

@@ -0,0 +1,130 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include <clc/internal/clc.h>
+#include <clc/shared/clc_vload.h>
+
+#define VLOAD_VECTORIZE(PRIM_TYPE, ADDR_SPACE)                                 \
+  _CLC_OVERLOAD _CLC_DEF PRIM_TYPE##2 __clc_vload2(                            \
+      size_t offset, const ADDR_SPACE PRIM_TYPE *x) {                          \
+    return *(                                                                  \
+        (const ADDR_SPACE less_aligned_##PRIM_TYPE##2 *)(&x[2 * offset]));     \
+  }                                                                            \
+                                                                               \
+  _CLC_OVERLOAD _CLC_DEF PRIM_TYPE##3 __clc_vload3(                            \
+      size_t offset, const ADDR_SPACE PRIM_TYPE *x) {                          \
+    PRIM_TYPE##2 vec =                                                         \
+        *((const ADDR_SPACE less_aligned_##PRIM_TYPE##2 *)(&x[3 * offset]));   \
+    return (PRIM_TYPE##3)(vec.s0, vec.s1, x[offset * 3 + 2]);                  \
+  }                                                                            \
+                                                                               \
+  _CLC_OVERLOAD _CLC_DEF PRIM_TYPE##4 __clc_vload4(                            \
+      size_t offset, const ADDR_SPACE PRIM_TYPE *x) {                          \
+    return *(                                                                  \
+        (const ADDR_SPACE less_aligned_##PRIM_TYPE##4 *)(&x[4 * offset]));     \
+  }                                                                            \
+                                                                               \
+  _CLC_OVERLOAD _CLC_DEF PRIM_TYPE##8 __clc_vload8(                            \
+      size_t offset, const ADDR_SPACE PRIM_TYPE *x) {                          \
+    return *(                                                                  \
+        (const ADDR_SPACE less_aligned_##PRIM_TYPE##8 *)(&x[8 * offset]));     \
+  }                                                                            \
+                                                                               \
+  _CLC_OVERLOAD _CLC_DEF PRIM_TYPE##16 __clc_vload16(                          \
+      size_t offset, const ADDR_SPACE PRIM_TYPE *x) {                          \
+    return *(                                                                  \
+        (const ADDR_SPACE less_aligned_##PRIM_TYPE##16 *)(&x[16 * offset]));   \
+  }
+
+#if _CLC_DISTINCT_GENERIC_AS_SUPPORTED
+#define VLOAD_VECTORIZE_GENERIC VLOAD_VECTORIZE
+#else
+// The generic address space isn't available, so make the macro do nothing
+#define VLOAD_VECTORIZE_GENERIC(X, Y)
+#endif
+
+#define VLOAD_ADDR_SPACES(__CLC_SCALAR_GENTYPE)                                \
+  VLOAD_VECTORIZE(__CLC_SCALAR_GENTYPE, __private)                             \
+  VLOAD_VECTORIZE(__CLC_SCALAR_GENTYPE, __local)                               \
+  VLOAD_VECTORIZE(__CLC_SCALAR_GENTYPE, __constant)                            \
+  VLOAD_VECTORIZE(__CLC_SCALAR_GENTYPE, __global)                              \
+  VLOAD_VECTORIZE_GENERIC(__CLC_SCALAR_GENTYPE, __generic)
+
+#define VLOAD_TYPES()                                                          \
+  VLOAD_ADDR_SPACES(char)                                                      \
+  VLOAD_ADDR_SPACES(uchar)                                                     \
+  VLOAD_ADDR_SPACES(short)                                                     \
+  VLOAD_ADDR_SPACES(ushort)                                                    \
+  VLOAD_ADDR_SPACES(int)                                                       \
+  VLOAD_ADDR_SPACES(uint)                                                      \
+  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
+#ifdef cl_khr_fp16
+#pragma OPENCL EXTENSION cl_khr_fp16 : enable
+VLOAD_ADDR_SPACES(half)
+#endif
+
+/* vload_half are legal even without cl_khr_fp16 */
+/* no vload_half for double */
+#define VEC_LOAD1(val, AS) val = __builtin_load_halff(&mem[offset++]);
+#define VEC_LOAD2(val, AS)                                                     \
+  VEC_LOAD1(val.lo, AS)                                                        \
+  VEC_LOAD1(val.hi, AS)
+#define VEC_LOAD3(val, AS)                                                     \
+  VEC_LOAD1(val.s0, AS)                                                        \
+  VEC_LOAD1(val.s1, AS)                                                        \
+  VEC_LOAD1(val.s2, AS)
+#define VEC_LOAD4(val, AS)                                                     \
+  VEC_LOAD2(val.lo, AS)                                                        \
+  VEC_LOAD2(val.hi, AS)
+#define VEC_LOAD8(val, AS)                                                     \
+  VEC_LOAD4(val.lo, AS)                                                        \
+  VEC_LOAD4(val.hi, AS)
+#define VEC_LOAD16(val, AS)                                                    \
+  VEC_LOAD8(val.lo, AS)                                                        \
+  VEC_LOAD8(val.hi, AS)
+
+#define __FUNC(SUFFIX, VEC_SIZE, OFFSET_SIZE, TYPE, AS)                        \
+  _CLC_OVERLOAD _CLC_DEF TYPE __clc_vload_half##SUFFIX(size_t offset,          \
+                                                       const AS half *mem) {   \
+    offset *= VEC_SIZE;                                                        \
+    TYPE __tmp;                                                                \
+    VEC_LOAD##VEC_SIZE(__tmp, AS) return __tmp;                                \
+  }                                                                            \
+  _CLC_OVERLOAD _CLC_DEF TYPE __clc_vloada_half##SUFFIX(size_t offset,         \
+                                                        const AS half *mem) {  \
+    offset *= OFFSET_SIZE;                                                     \
+    TYPE __tmp;                                                                \
+    VEC_LOAD##VEC_SIZE(__tmp, AS) return __tmp;                                \
+  }
+
+#define FUNC(SUFFIX, VEC_SIZE, OFFSET_SIZE, TYPE, AS)                          \
+  __FUNC(SUFFIX, VEC_SIZE, OFFSET_SIZE, TYPE, AS)
+
+#define __CLC_BODY "clc_vload_half.inc"
+#include <clc/math/gentype.inc>
+#undef FUNC
+#undef __FUNC
+#undef VEC_LOAD16
+#undef VEC_LOAD8
+#undef VEC_LOAD4
+#undef VEC_LOAD3
+#undef VEC_LOAD2
+#undef VEC_LOAD1
+#undef VLOAD_TYPES
+#undef VLOAD_ADDR_SPACES
+#undef VLOAD_VECTORIZE
+#undef VLOAD_VECTORIZE_GENERIC