================
@@ -0,0 +1,207 @@
+#ifndef __SPIRVINTRIN_H
+#define __SPIRVINTRIN_H
+
+#ifndef __SPIRV__
+#error "This file is intended for SPIRV targets or offloading to SPIRV"
+#endif
+
+#ifndef __GPUINTRIN_H
+#error "Never use <spirvintrin.h> directly; include <gpuintrin.h> instead"
+#endif
+
+#include <stdint.h>
+#if !defined(__cplusplus)
+_Pragma("push_macro(\"bool\")");
+#define bool _Bool
+#define true 1
+#define false 0
+#endif
+
+_Pragma("omp begin declare target device_type(nohost)");
+_Pragma("omp begin declare variant match(device = {arch(spirv64)})");
+
+// Type aliases to the address spaces used by the SPIR-V backend.
+//
+// TODO: FIX
+#define __gpu_private
+#define __gpu_constant
+#define __gpu_local
+#define __gpu_global __attribute__((address_space(1)))
+#define __gpu_generic __attribute__((address_space(4)))
+// Attribute to declare a function as a kernel.
+#define __gpu_kernel __attribute__((spirv_kernel, visibility("protected")))
+#define __SPIRV_VAR_QUALIFIERS extern const
+// Workgroup and invocation ID functions
+uint64_t __spirv_BuiltInNumWorkgroups(int i);
+uint64_t __spirv_BuiltInWorkgroupId(int i);
+uint64_t __spirv_BuiltInWorkgroupSize(int i);
+uint64_t __spirv_BuiltInLocalInvocationId(int i);
+
+#ifdef __cplusplus
+template <typename... Args>
+int __spirv_ocl_printf(Args...);
+#endif
+
+// Subgroup functions
+__SPIRV_VAR_QUALIFIERS uint32_t __spirv_BuiltInSubgroupLocalInvocationId;
+__SPIRV_VAR_QUALIFIERS uint32_t __spirv_BuiltInSubgroupSize;
+
+// Group non-uniform operations
+uint64_t __spirv_GroupNonUniformBallot(uint32_t execution_scope, bool
predicate);
+uint32_t __spirv_GroupNonUniformBroadcastFirst(uint32_t execution_scope,
uint32_t value);
+uint32_t __spirv_GroupNonUniformShuffle(uint32_t execution_scope, uint32_t
value, uint32_t id);
+
+// Synchronization
+void __spirv_ControlBarrier(uint32_t execution_scope, uint32_t memory_scope,
uint32_t semantics);
+void __spirv_MemoryBarrier(uint32_t memory_scope, uint32_t semantics);
+
+
+// Returns the number of blocks in the 'x' dimension.
+_DEFAULT_FN_ATTRS static __inline__ uint32_t __gpu_num_blocks_x(void) {
+ return __spirv_BuiltInNumWorkgroups(0);
+}
+
+// Returns the number of blocks in the 'y' dimension.
+_DEFAULT_FN_ATTRS static __inline__ uint32_t __gpu_num_blocks_y(void) {
+ return __spirv_BuiltInNumWorkgroups(1);
+}
+
+// Returns the number of blocks in the 'z' dimension.
+_DEFAULT_FN_ATTRS static __inline__ uint32_t __gpu_num_blocks_z(void) {
+ return __spirv_BuiltInNumWorkgroups(2);
+}
+
+// Returns the 'x' dimension of the current block's id.
+_DEFAULT_FN_ATTRS static __inline__ uint32_t __gpu_block_id_x(void) {
+ return __spirv_BuiltInWorkgroupId(0);
+}
+
+// Returns the 'y' dimension of the current block's id.
+_DEFAULT_FN_ATTRS static __inline__ uint32_t __gpu_block_id_y(void) {
+ return __spirv_BuiltInWorkgroupId(1);
+}
+
+// Returns the 'z' dimension of the current block's id.
+_DEFAULT_FN_ATTRS static __inline__ uint32_t __gpu_block_id_z(void) {
+ return __spirv_BuiltInWorkgroupId(2);
+}
+
+// Returns the number of threads in the 'x' dimension.
+_DEFAULT_FN_ATTRS static __inline__ uint32_t __gpu_num_threads_x(void) {
+ return __spirv_BuiltInWorkgroupSize(0);
+}
+
+// Returns the number of threads in the 'y' dimension.
+_DEFAULT_FN_ATTRS static __inline__ uint32_t __gpu_num_threads_y(void) {
+ return __spirv_BuiltInWorkgroupSize(1);
+}
+
+// Returns the number of threads in the 'z' dimension.
+_DEFAULT_FN_ATTRS static __inline__ uint32_t __gpu_num_threads_z(void) {
+ return __spirv_BuiltInWorkgroupSize(2);
+}
+
+// Returns the 'x' dimension id of the thread in the current block.
+_DEFAULT_FN_ATTRS static __inline__ uint32_t __gpu_thread_id_x(void) {
+ return __spirv_BuiltInLocalInvocationId(0);
+}
+
+// Returns the 'y' dimension id of the thread in the current block.
+_DEFAULT_FN_ATTRS static __inline__ uint32_t __gpu_thread_id_y(void) {
+ return __spirv_BuiltInLocalInvocationId(1);
+}
+
+// Returns the 'z' dimension id of the thread in the current block.
+_DEFAULT_FN_ATTRS static __inline__ uint32_t __gpu_thread_id_z(void) {
+ return __spirv_BuiltInLocalInvocationId(2);
+}
+
+// Returns the size of a warp, always 32 on NVIDIA hardware.
+_DEFAULT_FN_ATTRS static __inline__ uint32_t __gpu_num_lanes(void) { return
__spirv_BuiltInSubgroupSize; }
----------------
adurang wrote:
I don't think this line is following the correct format
https://github.com/llvm/llvm-project/pull/171011
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
