[libclc] Add v3 variants of async_work_group_copy/async_work_group_strided_copy/prefetch (#137932)

3-component vector type is supported for them per OpenCL spec.
2025-04-30 12:19:08 +00:00
parent 101fd87f98
commit 5b6fc61091
7 changed files with 43 additions and 251 deletions
--- a/libclc/generic/include/clc/async/async_work_group_copy.h
+++ b/libclc/generic/include/clc/async/async_work_group_copy.h
@@ -9,15 +9,21 @@
 #define __CLC_DST_ADDR_SPACE local
 #define __CLC_SRC_ADDR_SPACE global
 #define __CLC_BODY <clc/async/async_work_group_copy.inc>
-#include <clc/async/gentype.inc>
+#include <clc/integer/gentype.inc>
+#undef __CLC_BODY
+#define __CLC_BODY <clc/async/async_work_group_copy.inc>
+#include <clc/math/gentype.inc>
+#undef __CLC_BODY
 #undef __CLC_DST_ADDR_SPACE
 #undef __CLC_SRC_ADDR_SPACE
-#undef __CLC_BODY

 #define __CLC_DST_ADDR_SPACE global
 #define __CLC_SRC_ADDR_SPACE local
 #define __CLC_BODY <clc/async/async_work_group_copy.inc>
-#include <clc/async/gentype.inc>
+#include <clc/integer/gentype.inc>
+#undef __CLC_BODY
+#define __CLC_BODY <clc/async/async_work_group_copy.inc>
+#include <clc/math/gentype.inc>
+#undef __CLC_BODY
 #undef __CLC_DST_ADDR_SPACE
 #undef __CLC_SRC_ADDR_SPACE
-#undef __CLC_BODY
--- a/libclc/generic/include/clc/async/async_work_group_strided_copy.h
+++ b/libclc/generic/include/clc/async/async_work_group_strided_copy.h
@@ -9,15 +9,21 @@
 #define __CLC_DST_ADDR_SPACE local
 #define __CLC_SRC_ADDR_SPACE global
 #define __CLC_BODY <clc/async/async_work_group_strided_copy.inc>
-#include <clc/async/gentype.inc>
+#include <clc/integer/gentype.inc>
+#undef __CLC_BODY
+#define __CLC_BODY <clc/async/async_work_group_strided_copy.inc>
+#include <clc/math/gentype.inc>
+#undef __CLC_BODY
 #undef __CLC_DST_ADDR_SPACE
 #undef __CLC_SRC_ADDR_SPACE
-#undef __CLC_BODY

 #define __CLC_DST_ADDR_SPACE global
 #define __CLC_SRC_ADDR_SPACE local
 #define __CLC_BODY <clc/async/async_work_group_strided_copy.inc>
-#include <clc/async/gentype.inc>
+#include <clc/integer/gentype.inc>
+#undef __CLC_BODY
+#define __CLC_BODY <clc/async/async_work_group_strided_copy.inc>
+#include <clc/math/gentype.inc>
+#undef __CLC_BODY
 #undef __CLC_DST_ADDR_SPACE
 #undef __CLC_SRC_ADDR_SPACE
-#undef __CLC_BODY
--- a/libclc/generic/include/clc/async/gentype.inc
+++ b/libclc/generic/include/clc/async/gentype.inc
@@ -1,239 +0,0 @@
-//===----------------------------------------------------------------------===//
-//
-// 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_GENTYPE char
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE char2
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE char4
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE char8
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE char16
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE uchar
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE uchar2
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE uchar4
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE uchar8
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE uchar16
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE short
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE short2
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE short4
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE short8
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE short16
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE ushort
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE ushort2
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE ushort4
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE ushort8
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE ushort16
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE int
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE int2
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE int4
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE int8
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE int16
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE uint
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE uint2
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE uint4
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE uint8
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE uint16
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE float
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE float2
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE float4
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE float8
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE float16
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE long
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE long2
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE long4
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE long8
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE long16
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE ulong
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE ulong2
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE ulong4
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE ulong8
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE ulong16
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#ifdef cl_khr_fp64
-#pragma OPENCL EXTENSION cl_khr_fp64 : enable
-
-#define __CLC_GENTYPE double
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE double2
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE double4
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE double8
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE double16
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#endif
-
-#ifdef cl_khr_fp16
-#pragma OPENCL EXTENSION cl_khr_fp16: enable
-
-#define __CLC_GENTYPE half
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE half2
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE half4
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE half8
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#define __CLC_GENTYPE half16
-#include __CLC_BODY
-#undef __CLC_GENTYPE
-
-#endif
-
-#undef __CLC_BODY
--- a/libclc/generic/include/clc/async/prefetch.h
+++ b/libclc/generic/include/clc/async/prefetch.h
@@ -7,5 +7,9 @@
 //===----------------------------------------------------------------------===//

 #define __CLC_BODY <clc/async/prefetch.inc>
-#include <clc/async/gentype.inc>
+#include <clc/integer/gentype.inc>
+#undef __CLC_BODY
+
+#define __CLC_BODY <clc/async/prefetch.inc>
+#include <clc/math/gentype.inc>
 #undef __CLC_BODY
--- a/libclc/generic/lib/async/async_work_group_copy.cl
+++ b/libclc/generic/lib/async/async_work_group_copy.cl
@@ -9,4 +9,9 @@
 #include <clc/clc.h>

 #define __CLC_BODY <async_work_group_copy.inc>
-#include <clc/async/gentype.inc>
+#include <clc/integer/gentype.inc>
+#undef __CLC_BODY
+
+#define __CLC_BODY <async_work_group_copy.inc>
+#include <clc/math/gentype.inc>
+#undef __CLC_BODY
--- a/libclc/generic/lib/async/async_work_group_strided_copy.cl
+++ b/libclc/generic/lib/async/async_work_group_strided_copy.cl
@@ -9,4 +9,9 @@
 #include <clc/clc.h>

 #define __CLC_BODY <async_work_group_strided_copy.inc>
-#include <clc/async/gentype.inc>
+#include <clc/integer/gentype.inc>
+#undef __CLC_BODY
+
+#define __CLC_BODY <async_work_group_strided_copy.inc>
+#include <clc/math/gentype.inc>
+#undef __CLC_BODY
--- a/libclc/generic/lib/async/prefetch.cl
+++ b/libclc/generic/lib/async/prefetch.cl
@@ -9,4 +9,9 @@
 #include <clc/clc.h>

 #define __CLC_BODY <prefetch.inc>
-#include <clc/async/gentype.inc>
+#include <clc/integer/gentype.inc>
+#undef __CLC_BODY
+
+#define __CLC_BODY <prefetch.inc>
+#include <clc/math/gentype.inc>
+#undef __CLC_BODY