AI/rocm-systems
2
0
Fork 0
Código Incidencias Pull Requests Acciones Paquetes Proyectos Lanzamientos Wiki Actividad

SWDEV-470698 - fix formatting, add format check workflow (#657)

Explorar el Código
Este commit está contenido en:
Danylo Lytovchenko
2025-08-20 16:28:06 +02:00
cometido por GitHub
padre 5840940caa
commit f7338717ae
Se han modificado 1574 ficheros con 162972 adiciones y 199346 borrados
Descargar archivo de parche Descargar archivo de diferencias Expandir todos los archivos Contraer todos los archivos
projects/hip/include/hip/channel_descriptor.h
+1 -1
Ver fichero
@@ -33,7 +33,7 @@ THE SOFTWARE.
#elif !defined(__HIP_PLATFORM_AMD__) && defined(__HIP_PLATFORM_NVIDIA__)
#include <hip/nvidia_detail/nvidia_channel_descriptor.h>
#else
#error("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#error ("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#endif
#endif
projects/hip/include/hip/device_functions.h
+1 -1
Ver fichero
@@ -32,7 +32,7 @@ THE SOFTWARE.
#elif !defined(__HIP_PLATFORM_AMD__) && defined(__HIP_PLATFORM_NVIDIA__)
#include <device_functions.h>
#else
#error("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#error ("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#endif
#endif
projects/hip/include/hip/driver_types.h
+390 -373
Ver fichero
@@ -28,7 +28,7 @@ THE SOFTWARE.
#if __cplusplus
#include <cstdlib>
#else
#include <stdlib.h> // size_t
#include <stdlib.h> // size_t
#endif
#endif
@@ -48,20 +48,20 @@ typedef void* hipDeviceptr_t;
* HIP channel format kinds
*/
typedef enum hipChannelFormatKind {
hipChannelFormatKindSigned = 0, ///< Signed channel format
hipChannelFormatKindUnsigned = 1, ///< Unsigned channel format
hipChannelFormatKindFloat = 2, ///< Float channel format
hipChannelFormatKindNone = 3 ///< No channel format
hipChannelFormatKindSigned = 0, ///< Signed channel format
hipChannelFormatKindUnsigned = 1, ///< Unsigned channel format
hipChannelFormatKindFloat = 2, ///< Float channel format
hipChannelFormatKindNone = 3 ///< No channel format
} hipChannelFormatKind;
/**
* HIP channel format descriptor
*/
typedef struct hipChannelFormatDesc {
int x;
int y;
int z;
int w;
enum hipChannelFormatKind f; ///< Channel format kind
int x;
int y;
int z;
int w;
enum hipChannelFormatKind f; ///< Channel format kind
} hipChannelFormatDesc;
/** @brief The hipTexRefSetArray function flags parameter override format value*/
#define HIP_TRSA_OVERRIDE_FORMAT 0x01
@@ -78,74 +78,74 @@ typedef const struct hipArray* hipArray_const_t;
* HIP array format
*/
typedef enum hipArray_Format {
HIP_AD_FORMAT_UNSIGNED_INT8 = 0x01, ///< Unsigned 8-bit array format
HIP_AD_FORMAT_UNSIGNED_INT16 = 0x02, ///< Unsigned 16-bit array format
HIP_AD_FORMAT_UNSIGNED_INT32 = 0x03, ///< Unsigned 32-bit array format
HIP_AD_FORMAT_SIGNED_INT8 = 0x08, ///< Signed 8-bit array format
HIP_AD_FORMAT_SIGNED_INT16 = 0x09, ///< Signed 16-bit array format
HIP_AD_FORMAT_SIGNED_INT32 = 0x0a, ///< Signed 32-bit array format
HIP_AD_FORMAT_HALF = 0x10, ///< Half array format
HIP_AD_FORMAT_FLOAT = 0x20 ///< Float array format
HIP_AD_FORMAT_UNSIGNED_INT8 = 0x01, ///< Unsigned 8-bit array format
HIP_AD_FORMAT_UNSIGNED_INT16 = 0x02, ///< Unsigned 16-bit array format
HIP_AD_FORMAT_UNSIGNED_INT32 = 0x03, ///< Unsigned 32-bit array format
HIP_AD_FORMAT_SIGNED_INT8 = 0x08, ///< Signed 8-bit array format
HIP_AD_FORMAT_SIGNED_INT16 = 0x09, ///< Signed 16-bit array format
HIP_AD_FORMAT_SIGNED_INT32 = 0x0a, ///< Signed 32-bit array format
HIP_AD_FORMAT_HALF = 0x10, ///< Half array format
HIP_AD_FORMAT_FLOAT = 0x20 ///< Float array format
} hipArray_Format;
/**
* HIP array descriptor
*/
typedef struct HIP_ARRAY_DESCRIPTOR {
size_t Width; ///< Width of the array
size_t Height; ///< Height of the array
enum hipArray_Format Format; ///< Format of the array
unsigned int NumChannels; ///< Number of channels of the array
size_t Width; ///< Width of the array
size_t Height; ///< Height of the array
enum hipArray_Format Format; ///< Format of the array
unsigned int NumChannels; ///< Number of channels of the array
} HIP_ARRAY_DESCRIPTOR;
/**
* HIP 3D array descriptor
*/
typedef struct HIP_ARRAY3D_DESCRIPTOR {
size_t Width; ///< Width of the array
size_t Height; ///< Height of the array
size_t Depth; ///< Depth of the array
enum hipArray_Format Format; ///< Format of the array
unsigned int NumChannels; ///< Number of channels of the array
unsigned int Flags; ///< Flags of the array
size_t Width; ///< Width of the array
size_t Height; ///< Height of the array
size_t Depth; ///< Depth of the array
enum hipArray_Format Format; ///< Format of the array
unsigned int NumChannels; ///< Number of channels of the array
unsigned int Flags; ///< Flags of the array
} HIP_ARRAY3D_DESCRIPTOR;
#if !defined(__HIPCC_RTC__)
/**
* HIP 2D memory copy parameters
*/
typedef struct hip_Memcpy2D {
size_t srcXInBytes; ///< Source width in bytes
size_t srcY; ///< Source height
hipMemoryType srcMemoryType; ///< Source memory type
const void* srcHost; ///< Source pointer
hipDeviceptr_t srcDevice; ///< Source device
hipArray_t srcArray; ///< Source array
size_t srcPitch; ///< Source pitch
size_t dstXInBytes; ///< Destination width in bytes
size_t dstY; ///< Destination height
hipMemoryType dstMemoryType; ///< Destination memory type
void* dstHost; ///< Destination pointer
hipDeviceptr_t dstDevice; ///< Destination device
hipArray_t dstArray; ///< Destination array
size_t dstPitch; ///< Destination pitch
size_t WidthInBytes; ///< Width in bytes of the 2D memory copy
size_t Height; ///< Height of the 2D memory copy
size_t srcXInBytes; ///< Source width in bytes
size_t srcY; ///< Source height
hipMemoryType srcMemoryType; ///< Source memory type
const void* srcHost; ///< Source pointer
hipDeviceptr_t srcDevice; ///< Source device
hipArray_t srcArray; ///< Source array
size_t srcPitch; ///< Source pitch
size_t dstXInBytes; ///< Destination width in bytes
size_t dstY; ///< Destination height
hipMemoryType dstMemoryType; ///< Destination memory type
void* dstHost; ///< Destination pointer
hipDeviceptr_t dstDevice; ///< Destination device
hipArray_t dstArray; ///< Destination array
size_t dstPitch; ///< Destination pitch
size_t WidthInBytes; ///< Width in bytes of the 2D memory copy
size_t Height; ///< Height of the 2D memory copy
} hip_Memcpy2D;
#endif // !defined(__HIPCC_RTC__)
#endif // !defined(__HIPCC_RTC__)
/**
* HIP mipmapped array
*/
typedef struct hipMipmappedArray {
void* data; ///< Data pointer of the mipmapped array
struct hipChannelFormatDesc desc; ///< Description of the mipmapped array
unsigned int type; ///< Type of the mipmapped array
unsigned int width; ///< Width of the mipmapped array
unsigned int height; ///< Height of the mipmapped array
unsigned int depth; ///< Depth of the mipmapped array
unsigned int min_mipmap_level; ///< Minimum level of the mipmapped array
unsigned int max_mipmap_level; ///< Maximum level of the mipmapped array
unsigned int flags; ///< Flags of the mipmapped array
enum hipArray_Format format; ///< Format of the mipmapped array
unsigned int num_channels; ///< Number of channels of the mipmapped array
void* data; ///< Data pointer of the mipmapped array
struct hipChannelFormatDesc desc; ///< Description of the mipmapped array
unsigned int type; ///< Type of the mipmapped array
unsigned int width; ///< Width of the mipmapped array
unsigned int height; ///< Height of the mipmapped array
unsigned int depth; ///< Depth of the mipmapped array
unsigned int min_mipmap_level; ///< Minimum level of the mipmapped array
unsigned int max_mipmap_level; ///< Maximum level of the mipmapped array
unsigned int flags; ///< Flags of the mipmapped array
enum hipArray_Format format; ///< Format of the mipmapped array
unsigned int num_channels; ///< Number of channels of the mipmapped array
} hipMipmappedArray;
/**
* HIP mipmapped array pointer
@@ -157,304 +157,306 @@ typedef const struct hipMipmappedArray* hipMipmappedArray_const_t;
* HIP resource types
*/
typedef enum hipResourceType {
hipResourceTypeArray = 0x00, ///< Array resource
hipResourceTypeMipmappedArray = 0x01, ///< Mipmapped array resource
hipResourceTypeLinear = 0x02, ///< Linear resource
hipResourceTypePitch2D = 0x03 ///< Pitch 2D resource
hipResourceTypeArray = 0x00, ///< Array resource
hipResourceTypeMipmappedArray = 0x01, ///< Mipmapped array resource
hipResourceTypeLinear = 0x02, ///< Linear resource
hipResourceTypePitch2D = 0x03 ///< Pitch 2D resource
} hipResourceType;
typedef enum HIPresourcetype_enum {
HIP_RESOURCE_TYPE_ARRAY = 0x00, ///< Array resource
HIP_RESOURCE_TYPE_MIPMAPPED_ARRAY = 0x01, ///< Mipmapped array resource
HIP_RESOURCE_TYPE_LINEAR = 0x02, ///< Linear resource
HIP_RESOURCE_TYPE_PITCH2D = 0x03 ///< Pitch 2D resource
} HIPresourcetype, hipResourcetype;
HIP_RESOURCE_TYPE_ARRAY = 0x00, ///< Array resource
HIP_RESOURCE_TYPE_MIPMAPPED_ARRAY = 0x01, ///< Mipmapped array resource
HIP_RESOURCE_TYPE_LINEAR = 0x02, ///< Linear resource
HIP_RESOURCE_TYPE_PITCH2D = 0x03 ///< Pitch 2D resource
} HIPresourcetype,
hipResourcetype;
/**
* HIP texture address modes
*/
typedef enum HIPaddress_mode_enum {
HIP_TR_ADDRESS_MODE_WRAP = 0, ///< Wrap address mode
HIP_TR_ADDRESS_MODE_CLAMP = 1, ///< Clamp address mode
HIP_TR_ADDRESS_MODE_MIRROR = 2, ///< Mirror address mode
HIP_TR_ADDRESS_MODE_BORDER = 3 ///< Border address mode
HIP_TR_ADDRESS_MODE_WRAP = 0, ///< Wrap address mode
HIP_TR_ADDRESS_MODE_CLAMP = 1, ///< Clamp address mode
HIP_TR_ADDRESS_MODE_MIRROR = 2, ///< Mirror address mode
HIP_TR_ADDRESS_MODE_BORDER = 3 ///< Border address mode
} HIPaddress_mode;
/**
* HIP filter modes
*/
typedef enum HIPfilter_mode_enum {
HIP_TR_FILTER_MODE_POINT = 0, ///< Filter mode point
HIP_TR_FILTER_MODE_LINEAR = 1 ///< Filter mode linear
HIP_TR_FILTER_MODE_POINT = 0, ///< Filter mode point
HIP_TR_FILTER_MODE_LINEAR = 1 ///< Filter mode linear
} HIPfilter_mode;
/**
* HIP texture descriptor
*/
typedef struct HIP_TEXTURE_DESC_st {
HIPaddress_mode addressMode[3]; ///< Address modes
HIPfilter_mode filterMode; ///< Filter mode
unsigned int flags; ///< Flags
unsigned int maxAnisotropy; ///< Maximum anisotropy ratio
HIPfilter_mode mipmapFilterMode; ///< Mipmap filter mode
float mipmapLevelBias; ///< Mipmap level bias
float minMipmapLevelClamp; ///< Mipmap minimum level clamp
float maxMipmapLevelClamp; ///< Mipmap maximum level clamp
float borderColor[4]; ///< Border Color
int reserved[12];
HIPaddress_mode addressMode[3]; ///< Address modes
HIPfilter_mode filterMode; ///< Filter mode
unsigned int flags; ///< Flags
unsigned int maxAnisotropy; ///< Maximum anisotropy ratio
HIPfilter_mode mipmapFilterMode; ///< Mipmap filter mode
float mipmapLevelBias; ///< Mipmap level bias
float minMipmapLevelClamp; ///< Mipmap minimum level clamp
float maxMipmapLevelClamp; ///< Mipmap maximum level clamp
float borderColor[4]; ///< Border Color
int reserved[12];
} HIP_TEXTURE_DESC;
/**
* HIP texture resource view formats
*/
typedef enum hipResourceViewFormat {
hipResViewFormatNone = 0x00, ///< No resource view format (use underlying resource format)
hipResViewFormatUnsignedChar1 = 0x01, ///< 1 channel, unsigned 8-bit integers
hipResViewFormatUnsignedChar2 = 0x02, ///< 2 channels, unsigned 8-bit integers
hipResViewFormatUnsignedChar4 = 0x03, ///< 4 channels, unsigned 8-bit integers
hipResViewFormatSignedChar1 = 0x04, ///< 1 channel, signed 8-bit integers
hipResViewFormatSignedChar2 = 0x05, ///< 2 channels, signed 8-bit integers
hipResViewFormatSignedChar4 = 0x06, ///< 4 channels, signed 8-bit integers
hipResViewFormatUnsignedShort1 = 0x07, ///< 1 channel, unsigned 16-bit integers
hipResViewFormatUnsignedShort2 = 0x08, ///< 2 channels, unsigned 16-bit integers
hipResViewFormatUnsignedShort4 = 0x09, ///< 4 channels, unsigned 16-bit integers
hipResViewFormatSignedShort1 = 0x0a, ///< 1 channel, signed 16-bit integers
hipResViewFormatSignedShort2 = 0x0b, ///< 2 channels, signed 16-bit integers
hipResViewFormatSignedShort4 = 0x0c, ///< 4 channels, signed 16-bit integers
hipResViewFormatUnsignedInt1 = 0x0d, ///< 1 channel, unsigned 32-bit integers
hipResViewFormatUnsignedInt2 = 0x0e, ///< 2 channels, unsigned 32-bit integers
hipResViewFormatUnsignedInt4 = 0x0f, ///< 4 channels, unsigned 32-bit integers
hipResViewFormatSignedInt1 = 0x10, ///< 1 channel, signed 32-bit integers
hipResViewFormatSignedInt2 = 0x11, ///< 2 channels, signed 32-bit integers
hipResViewFormatSignedInt4 = 0x12, ///< 4 channels, signed 32-bit integers
hipResViewFormatHalf1 = 0x13, ///< 1 channel, 16-bit floating point
hipResViewFormatHalf2 = 0x14, ///< 2 channels, 16-bit floating point
hipResViewFormatHalf4 = 0x15, ///< 4 channels, 16-bit floating point
hipResViewFormatFloat1 = 0x16, ///< 1 channel, 32-bit floating point
hipResViewFormatFloat2 = 0x17, ///< 2 channels, 32-bit floating point
hipResViewFormatFloat4 = 0x18, ///< 4 channels, 32-bit floating point
hipResViewFormatUnsignedBlockCompressed1 = 0x19, ///< Block-compressed 1
hipResViewFormatUnsignedBlockCompressed2 = 0x1a, ///< Block-compressed 2
hipResViewFormatUnsignedBlockCompressed3 = 0x1b, ///< Block-compressed 3
hipResViewFormatUnsignedBlockCompressed4 = 0x1c, ///< Block-compressed 4 unsigned
hipResViewFormatSignedBlockCompressed4 = 0x1d, ///< Block-compressed 4 signed
hipResViewFormatUnsignedBlockCompressed5 = 0x1e, ///< Block-compressed 5 unsigned
hipResViewFormatSignedBlockCompressed5 = 0x1f, ///< Block-compressed 5 signed
hipResViewFormatUnsignedBlockCompressed6H = 0x20, ///< Block-compressed 6 unsigned half-float
hipResViewFormatSignedBlockCompressed6H = 0x21, ///< Block-compressed 6 signed half-float
hipResViewFormatUnsignedBlockCompressed7 = 0x22 ///< Block-compressed 7
hipResViewFormatNone = 0x00, ///< No resource view format (use underlying resource format)
hipResViewFormatUnsignedChar1 = 0x01, ///< 1 channel, unsigned 8-bit integers
hipResViewFormatUnsignedChar2 = 0x02, ///< 2 channels, unsigned 8-bit integers
hipResViewFormatUnsignedChar4 = 0x03, ///< 4 channels, unsigned 8-bit integers
hipResViewFormatSignedChar1 = 0x04, ///< 1 channel, signed 8-bit integers
hipResViewFormatSignedChar2 = 0x05, ///< 2 channels, signed 8-bit integers
hipResViewFormatSignedChar4 = 0x06, ///< 4 channels, signed 8-bit integers
hipResViewFormatUnsignedShort1 = 0x07, ///< 1 channel, unsigned 16-bit integers
hipResViewFormatUnsignedShort2 = 0x08, ///< 2 channels, unsigned 16-bit integers
hipResViewFormatUnsignedShort4 = 0x09, ///< 4 channels, unsigned 16-bit integers
hipResViewFormatSignedShort1 = 0x0a, ///< 1 channel, signed 16-bit integers
hipResViewFormatSignedShort2 = 0x0b, ///< 2 channels, signed 16-bit integers
hipResViewFormatSignedShort4 = 0x0c, ///< 4 channels, signed 16-bit integers
hipResViewFormatUnsignedInt1 = 0x0d, ///< 1 channel, unsigned 32-bit integers
hipResViewFormatUnsignedInt2 = 0x0e, ///< 2 channels, unsigned 32-bit integers
hipResViewFormatUnsignedInt4 = 0x0f, ///< 4 channels, unsigned 32-bit integers
hipResViewFormatSignedInt1 = 0x10, ///< 1 channel, signed 32-bit integers
hipResViewFormatSignedInt2 = 0x11, ///< 2 channels, signed 32-bit integers
hipResViewFormatSignedInt4 = 0x12, ///< 4 channels, signed 32-bit integers
hipResViewFormatHalf1 = 0x13, ///< 1 channel, 16-bit floating point
hipResViewFormatHalf2 = 0x14, ///< 2 channels, 16-bit floating point
hipResViewFormatHalf4 = 0x15, ///< 4 channels, 16-bit floating point
hipResViewFormatFloat1 = 0x16, ///< 1 channel, 32-bit floating point
hipResViewFormatFloat2 = 0x17, ///< 2 channels, 32-bit floating point
hipResViewFormatFloat4 = 0x18, ///< 4 channels, 32-bit floating point
hipResViewFormatUnsignedBlockCompressed1 = 0x19, ///< Block-compressed 1
hipResViewFormatUnsignedBlockCompressed2 = 0x1a, ///< Block-compressed 2
hipResViewFormatUnsignedBlockCompressed3 = 0x1b, ///< Block-compressed 3
hipResViewFormatUnsignedBlockCompressed4 = 0x1c, ///< Block-compressed 4 unsigned
hipResViewFormatSignedBlockCompressed4 = 0x1d, ///< Block-compressed 4 signed
hipResViewFormatUnsignedBlockCompressed5 = 0x1e, ///< Block-compressed 5 unsigned
hipResViewFormatSignedBlockCompressed5 = 0x1f, ///< Block-compressed 5 signed
hipResViewFormatUnsignedBlockCompressed6H = 0x20, ///< Block-compressed 6 unsigned half-float
hipResViewFormatSignedBlockCompressed6H = 0x21, ///< Block-compressed 6 signed half-float
hipResViewFormatUnsignedBlockCompressed7 = 0x22 ///< Block-compressed 7
} hipResourceViewFormat;
/**
* HIP texture resource view formats
*/
typedef enum HIPresourceViewFormat_enum {
HIP_RES_VIEW_FORMAT_NONE = 0x00, ///< No resource view format (use underlying resource format)
HIP_RES_VIEW_FORMAT_UINT_1X8 = 0x01, ///< 1 channel, unsigned 8-bit integers
HIP_RES_VIEW_FORMAT_UINT_2X8 = 0x02, ///< 2 channels, unsigned 8-bit integers
HIP_RES_VIEW_FORMAT_UINT_4X8 = 0x03, ///< 4 channels, unsigned 8-bit integers
HIP_RES_VIEW_FORMAT_SINT_1X8 = 0x04, ///< 1 channel, signed 8-bit integers
HIP_RES_VIEW_FORMAT_SINT_2X8 = 0x05, ///< 2 channels, signed 8-bit integers
HIP_RES_VIEW_FORMAT_SINT_4X8 = 0x06, ///< 4 channels, signed 8-bit integers
HIP_RES_VIEW_FORMAT_UINT_1X16 = 0x07, ///< 1 channel, unsigned 16-bit integers
HIP_RES_VIEW_FORMAT_UINT_2X16 = 0x08, ///< 2 channels, unsigned 16-bit integers
HIP_RES_VIEW_FORMAT_UINT_4X16 = 0x09, ///< 4 channels, unsigned 16-bit integers
HIP_RES_VIEW_FORMAT_SINT_1X16 = 0x0a, ///< 1 channel, signed 16-bit integers
HIP_RES_VIEW_FORMAT_SINT_2X16 = 0x0b, ///< 2 channels, signed 16-bit integers
HIP_RES_VIEW_FORMAT_SINT_4X16 = 0x0c, ///< 4 channels, signed 16-bit integers
HIP_RES_VIEW_FORMAT_UINT_1X32 = 0x0d, ///< 1 channel, unsigned 32-bit integers
HIP_RES_VIEW_FORMAT_UINT_2X32 = 0x0e, ///< 2 channels, unsigned 32-bit integers
HIP_RES_VIEW_FORMAT_UINT_4X32 = 0x0f, ///< 4 channels, unsigned 32-bit integers
HIP_RES_VIEW_FORMAT_SINT_1X32 = 0x10, ///< 1 channel, signed 32-bit integers
HIP_RES_VIEW_FORMAT_SINT_2X32 = 0x11, ///< 2 channels, signed 32-bit integers
HIP_RES_VIEW_FORMAT_SINT_4X32 = 0x12, ///< 4 channels, signed 32-bit integers
HIP_RES_VIEW_FORMAT_FLOAT_1X16 = 0x13, ///< 1 channel, 16-bit floating point
HIP_RES_VIEW_FORMAT_FLOAT_2X16 = 0x14, ///< 2 channels, 16-bit floating point
HIP_RES_VIEW_FORMAT_FLOAT_4X16 = 0x15, ///< 4 channels, 16-bit floating point
HIP_RES_VIEW_FORMAT_FLOAT_1X32 = 0x16, ///< 1 channel, 32-bit floating point
HIP_RES_VIEW_FORMAT_FLOAT_2X32 = 0x17, ///< 2 channels, 32-bit floating point
HIP_RES_VIEW_FORMAT_FLOAT_4X32 = 0x18, ///< 4 channels, 32-bit floating point
HIP_RES_VIEW_FORMAT_UNSIGNED_BC1 = 0x19, ///< Block-compressed 1
HIP_RES_VIEW_FORMAT_UNSIGNED_BC2 = 0x1a, ///< Block-compressed 2
HIP_RES_VIEW_FORMAT_UNSIGNED_BC3 = 0x1b, ///< Block-compressed 3
HIP_RES_VIEW_FORMAT_UNSIGNED_BC4 = 0x1c, ///< Block-compressed 4 unsigned
HIP_RES_VIEW_FORMAT_SIGNED_BC4 = 0x1d, ///< Block-compressed 4 signed
HIP_RES_VIEW_FORMAT_UNSIGNED_BC5 = 0x1e, ///< Block-compressed 5 unsigned
HIP_RES_VIEW_FORMAT_SIGNED_BC5 = 0x1f, ///< Block-compressed 5 signed
HIP_RES_VIEW_FORMAT_UNSIGNED_BC6H = 0x20, ///< Block-compressed 6 unsigned half-float
HIP_RES_VIEW_FORMAT_SIGNED_BC6H = 0x21, ///< Block-compressed 6 signed half-float
HIP_RES_VIEW_FORMAT_UNSIGNED_BC7 = 0x22 ///< Block-compressed 7
HIP_RES_VIEW_FORMAT_NONE = 0x00, ///< No resource view format (use underlying resource format)
HIP_RES_VIEW_FORMAT_UINT_1X8 = 0x01, ///< 1 channel, unsigned 8-bit integers
HIP_RES_VIEW_FORMAT_UINT_2X8 = 0x02, ///< 2 channels, unsigned 8-bit integers
HIP_RES_VIEW_FORMAT_UINT_4X8 = 0x03, ///< 4 channels, unsigned 8-bit integers
HIP_RES_VIEW_FORMAT_SINT_1X8 = 0x04, ///< 1 channel, signed 8-bit integers
HIP_RES_VIEW_FORMAT_SINT_2X8 = 0x05, ///< 2 channels, signed 8-bit integers
HIP_RES_VIEW_FORMAT_SINT_4X8 = 0x06, ///< 4 channels, signed 8-bit integers
HIP_RES_VIEW_FORMAT_UINT_1X16 = 0x07, ///< 1 channel, unsigned 16-bit integers
HIP_RES_VIEW_FORMAT_UINT_2X16 = 0x08, ///< 2 channels, unsigned 16-bit integers
HIP_RES_VIEW_FORMAT_UINT_4X16 = 0x09, ///< 4 channels, unsigned 16-bit integers
HIP_RES_VIEW_FORMAT_SINT_1X16 = 0x0a, ///< 1 channel, signed 16-bit integers
HIP_RES_VIEW_FORMAT_SINT_2X16 = 0x0b, ///< 2 channels, signed 16-bit integers
HIP_RES_VIEW_FORMAT_SINT_4X16 = 0x0c, ///< 4 channels, signed 16-bit integers
HIP_RES_VIEW_FORMAT_UINT_1X32 = 0x0d, ///< 1 channel, unsigned 32-bit integers
HIP_RES_VIEW_FORMAT_UINT_2X32 = 0x0e, ///< 2 channels, unsigned 32-bit integers
HIP_RES_VIEW_FORMAT_UINT_4X32 = 0x0f, ///< 4 channels, unsigned 32-bit integers
HIP_RES_VIEW_FORMAT_SINT_1X32 = 0x10, ///< 1 channel, signed 32-bit integers
HIP_RES_VIEW_FORMAT_SINT_2X32 = 0x11, ///< 2 channels, signed 32-bit integers
HIP_RES_VIEW_FORMAT_SINT_4X32 = 0x12, ///< 4 channels, signed 32-bit integers
HIP_RES_VIEW_FORMAT_FLOAT_1X16 = 0x13, ///< 1 channel, 16-bit floating point
HIP_RES_VIEW_FORMAT_FLOAT_2X16 = 0x14, ///< 2 channels, 16-bit floating point
HIP_RES_VIEW_FORMAT_FLOAT_4X16 = 0x15, ///< 4 channels, 16-bit floating point
HIP_RES_VIEW_FORMAT_FLOAT_1X32 = 0x16, ///< 1 channel, 32-bit floating point
HIP_RES_VIEW_FORMAT_FLOAT_2X32 = 0x17, ///< 2 channels, 32-bit floating point
HIP_RES_VIEW_FORMAT_FLOAT_4X32 = 0x18, ///< 4 channels, 32-bit floating point
HIP_RES_VIEW_FORMAT_UNSIGNED_BC1 = 0x19, ///< Block-compressed 1
HIP_RES_VIEW_FORMAT_UNSIGNED_BC2 = 0x1a, ///< Block-compressed 2
HIP_RES_VIEW_FORMAT_UNSIGNED_BC3 = 0x1b, ///< Block-compressed 3
HIP_RES_VIEW_FORMAT_UNSIGNED_BC4 = 0x1c, ///< Block-compressed 4 unsigned
HIP_RES_VIEW_FORMAT_SIGNED_BC4 = 0x1d, ///< Block-compressed 4 signed
HIP_RES_VIEW_FORMAT_UNSIGNED_BC5 = 0x1e, ///< Block-compressed 5 unsigned
HIP_RES_VIEW_FORMAT_SIGNED_BC5 = 0x1f, ///< Block-compressed 5 signed
HIP_RES_VIEW_FORMAT_UNSIGNED_BC6H = 0x20, ///< Block-compressed 6 unsigned half-float
HIP_RES_VIEW_FORMAT_SIGNED_BC6H = 0x21, ///< Block-compressed 6 signed half-float
HIP_RES_VIEW_FORMAT_UNSIGNED_BC7 = 0x22 ///< Block-compressed 7
} HIPresourceViewFormat;
/**
* HIP resource descriptor
*/
typedef struct hipResourceDesc {
enum hipResourceType resType; ///< Resource type
union {
struct {
hipArray_t array; ///< HIP array
} array;
struct {
hipMipmappedArray_t mipmap; ///< HIP mipmapped array
} mipmap;
struct {
void* devPtr; ///< Device pointer
struct hipChannelFormatDesc desc; ///< Channel format description
size_t sizeInBytes; ///< Size in bytes
} linear;
struct {
void* devPtr; ///< Device pointer
struct hipChannelFormatDesc desc; ///< Channel format description
size_t width; ///< Width of the array in elements
size_t height; ///< Height of the array in elements
size_t pitchInBytes; ///< Pitch between two rows in bytes
} pitch2D;
} res;
enum hipResourceType resType; ///< Resource type
union {
struct {
hipArray_t array; ///< HIP array
} array;
struct {
hipMipmappedArray_t mipmap; ///< HIP mipmapped array
} mipmap;
struct {
void* devPtr; ///< Device pointer
struct hipChannelFormatDesc desc; ///< Channel format description
size_t sizeInBytes; ///< Size in bytes
} linear;
struct {
void* devPtr; ///< Device pointer
struct hipChannelFormatDesc desc; ///< Channel format description
size_t width; ///< Width of the array in elements
size_t height; ///< Height of the array in elements
size_t pitchInBytes; ///< Pitch between two rows in bytes
} pitch2D;
} res;
} hipResourceDesc;
/**
* HIP resource view descriptor struct
*/
typedef struct HIP_RESOURCE_DESC_st {
HIPresourcetype resType; ///< Resource type
union {
struct {
hipArray_t hArray; ///< HIP array
} array;
struct {
hipMipmappedArray_t hMipmappedArray; ///< HIP mipmapped array
} mipmap;
struct {
hipDeviceptr_t devPtr; ///< Device pointer
hipArray_Format format; ///< Array format
unsigned int numChannels; ///< Channels per array element
size_t sizeInBytes; ///< Size in bytes
} linear;
struct {
hipDeviceptr_t devPtr; ///< Device pointer
hipArray_Format format; ///< Array format
unsigned int numChannels; ///< Channels per array element
size_t width; ///< Width of the array in elements
size_t height; ///< Height of the array in elements
size_t pitchInBytes; ///< Pitch between two rows in bytes
} pitch2D;
struct {
int reserved[32];
} reserved;
} res;
unsigned int flags; ///< Flags (must be zero)
HIPresourcetype resType; ///< Resource type
union {
struct {
hipArray_t hArray; ///< HIP array
} array;
struct {
hipMipmappedArray_t hMipmappedArray; ///< HIP mipmapped array
} mipmap;
struct {
hipDeviceptr_t devPtr; ///< Device pointer
hipArray_Format format; ///< Array format
unsigned int numChannels; ///< Channels per array element
size_t sizeInBytes; ///< Size in bytes
} linear;
struct {
hipDeviceptr_t devPtr; ///< Device pointer
hipArray_Format format; ///< Array format
unsigned int numChannels; ///< Channels per array element
size_t width; ///< Width of the array in elements
size_t height; ///< Height of the array in elements
size_t pitchInBytes; ///< Pitch between two rows in bytes
} pitch2D;
struct {
int reserved[32];
} reserved;
} res;
unsigned int flags; ///< Flags (must be zero)
} HIP_RESOURCE_DESC;
/**
* HIP resource view descriptor
*/
struct hipResourceViewDesc {
enum hipResourceViewFormat format; ///< Resource view format
size_t width; ///< Width of the resource view
size_t height; ///< Height of the resource view
size_t depth; ///< Depth of the resource view
unsigned int firstMipmapLevel; ///< First defined mipmap level
unsigned int lastMipmapLevel; ///< Last defined mipmap level
unsigned int firstLayer; ///< First layer index
unsigned int lastLayer; ///< Last layer index
enum hipResourceViewFormat format; ///< Resource view format
size_t width; ///< Width of the resource view
size_t height; ///< Height of the resource view
size_t depth; ///< Depth of the resource view
unsigned int firstMipmapLevel; ///< First defined mipmap level
unsigned int lastMipmapLevel; ///< Last defined mipmap level
unsigned int firstLayer; ///< First layer index
unsigned int lastLayer; ///< Last layer index
};
/**
* Resource view descriptor
*/
typedef struct HIP_RESOURCE_VIEW_DESC_st {
HIPresourceViewFormat format; ///< Resource view format
size_t width; ///< Width of the resource view
size_t height; ///< Height of the resource view
size_t depth; ///< Depth of the resource view
unsigned int firstMipmapLevel; ///< First defined mipmap level
unsigned int lastMipmapLevel; ///< Last defined mipmap level
unsigned int firstLayer; ///< First layer index
unsigned int lastLayer; ///< Last layer index
unsigned int reserved[16];
HIPresourceViewFormat format; ///< Resource view format
size_t width; ///< Width of the resource view
size_t height; ///< Height of the resource view
size_t depth; ///< Depth of the resource view
unsigned int firstMipmapLevel; ///< First defined mipmap level
unsigned int lastMipmapLevel; ///< Last defined mipmap level
unsigned int firstLayer; ///< First layer index
unsigned int lastLayer; ///< Last layer index
unsigned int reserved[16];
} HIP_RESOURCE_VIEW_DESC;
/**
* Memory copy types
*/
#if !defined(__HIPCC_RTC__)
typedef enum hipMemcpyKind {
hipMemcpyHostToHost = 0, ///< Host-to-Host Copy
hipMemcpyHostToDevice = 1, ///< Host-to-Device Copy
hipMemcpyDeviceToHost = 2, ///< Device-to-Host Copy
hipMemcpyDeviceToDevice = 3, ///< Device-to-Device Copy
hipMemcpyDefault = 4, ///< Runtime will automatically determine
///<copy-kind based on virtual addresses.
hipMemcpyDeviceToDeviceNoCU = 1024 ///< Device-to-Device Copy without using compute units
hipMemcpyHostToHost = 0, ///< Host-to-Host Copy
hipMemcpyHostToDevice = 1, ///< Host-to-Device Copy
hipMemcpyDeviceToHost = 2, ///< Device-to-Host Copy
hipMemcpyDeviceToDevice = 3, ///< Device-to-Device Copy
hipMemcpyDefault = 4, ///< Runtime will automatically determine
///< copy-kind based on virtual addresses.
hipMemcpyDeviceToDeviceNoCU = 1024 ///< Device-to-Device Copy without using compute units
} hipMemcpyKind;
/**
* HIP pithed pointer
*/
typedef struct hipPitchedPtr {
void* ptr; ///< Pointer to the allocated memory
size_t pitch; ///< Pitch in bytes
size_t xsize; ///< Logical size of the first dimension of allocation in elements
size_t ysize; ///< Logical size of the second dimension of allocation in elements
void* ptr; ///< Pointer to the allocated memory
size_t pitch; ///< Pitch in bytes
size_t xsize; ///< Logical size of the first dimension of allocation in elements
size_t ysize; ///< Logical size of the second dimension of allocation in elements
} hipPitchedPtr;
/**
* HIP extent
*/
typedef struct hipExtent {
size_t width; // Width in elements when referring to array memory, in bytes when referring to
// linear memory
size_t height;
size_t depth;
size_t width; // Width in elements when referring to array memory, in bytes when referring to
// linear memory
size_t height;
size_t depth;
} hipExtent;
/**
* HIP position
*/
typedef struct hipPos {
size_t x; ///< X coordinate
size_t y; ///< Y coordinate
size_t z; ///< Z coordinate
size_t x; ///< X coordinate
size_t y; ///< Y coordinate
size_t z; ///< Z coordinate
} hipPos;
/**
* HIP 3D memory copy parameters
*/
typedef struct hipMemcpy3DParms {
hipArray_t srcArray; ///< Source array
struct hipPos srcPos; ///< Source position
struct hipPitchedPtr srcPtr; ///< Source pointer
hipArray_t dstArray; ///< Destination array
struct hipPos dstPos; ///< Destination position
struct hipPitchedPtr dstPtr; ///< Destination pointer
struct hipExtent extent; ///< Extent of 3D memory copy
enum hipMemcpyKind kind; ///< Kind of 3D memory copy
hipArray_t srcArray; ///< Source array
struct hipPos srcPos; ///< Source position
struct hipPitchedPtr srcPtr; ///< Source pointer
hipArray_t dstArray; ///< Destination array
struct hipPos dstPos; ///< Destination position
struct hipPitchedPtr dstPtr; ///< Destination pointer
struct hipExtent extent; ///< Extent of 3D memory copy
enum hipMemcpyKind kind; ///< Kind of 3D memory copy
} hipMemcpy3DParms;
/**
* HIP 3D memory copy
*/
typedef struct HIP_MEMCPY3D {
size_t srcXInBytes; ///< Source X in bytes
size_t srcY; ///< Source Y
size_t srcZ; ///< Source Z
size_t srcLOD; ///< Source LOD
hipMemoryType srcMemoryType; ///< Source memory type
const void* srcHost; ///< Source host pointer
hipDeviceptr_t srcDevice; ///< Source device
hipArray_t srcArray; ///< Source array
size_t srcPitch; ///< Source pitch
size_t srcHeight; ///< Source height
size_t dstXInBytes; ///< Destination X in bytes
size_t dstY; ///< Destination Y
size_t dstZ; ///< Destination Z
size_t dstLOD; ///< Destination LOD
hipMemoryType dstMemoryType; ///< Destination memory type
void* dstHost; ///< Destination host pointer
hipDeviceptr_t dstDevice; ///< Destination device
hipArray_t dstArray; ///< Destination array
size_t dstPitch; ///< Destination pitch
size_t dstHeight; ///< Destination height
size_t WidthInBytes; ///< Width in bytes of 3D memory copy
size_t Height; ///< Height in bytes of 3D memory copy
size_t Depth; ///< Depth in bytes of 3D memory copy
size_t srcXInBytes; ///< Source X in bytes
size_t srcY; ///< Source Y
size_t srcZ; ///< Source Z
size_t srcLOD; ///< Source LOD
hipMemoryType srcMemoryType; ///< Source memory type
const void* srcHost; ///< Source host pointer
hipDeviceptr_t srcDevice; ///< Source device
hipArray_t srcArray; ///< Source array
size_t srcPitch; ///< Source pitch
size_t srcHeight; ///< Source height
size_t dstXInBytes; ///< Destination X in bytes
size_t dstY; ///< Destination Y
size_t dstZ; ///< Destination Z
size_t dstLOD; ///< Destination LOD
hipMemoryType dstMemoryType; ///< Destination memory type
void* dstHost; ///< Destination host pointer
hipDeviceptr_t dstDevice; ///< Destination device
hipArray_t dstArray; ///< Destination array
size_t dstPitch; ///< Destination pitch
size_t dstHeight; ///< Destination height
size_t WidthInBytes; ///< Width in bytes of 3D memory copy
size_t Height; ///< Height in bytes of 3D memory copy
size_t Depth; ///< Depth in bytes of 3D memory copy
} HIP_MEMCPY3D;
/**
* Specifies the type of location
*/
typedef enum hipMemLocationType {
hipMemLocationTypeInvalid = 0,
hipMemLocationTypeNone = 0,
hipMemLocationTypeDevice = 1, ///< Device location, thus it's HIP device ID
hipMemLocationTypeHost = 2, ///< Host location, id is ignored
hipMemLocationTypeHostNuma = 3, ///< Host NUMA node location, id is host NUMA node id
hipMemLocationTypeHostNumaCurrent = 4 ///< Host NUMA node closest to current thread’s CPU, id is ignored
hipMemLocationTypeInvalid = 0,
hipMemLocationTypeNone = 0,
hipMemLocationTypeDevice = 1, ///< Device location, thus it's HIP device ID
hipMemLocationTypeHost = 2, ///< Host location, id is ignored
hipMemLocationTypeHostNuma = 3, ///< Host NUMA node location, id is host NUMA node id
hipMemLocationTypeHostNumaCurrent =
4 ///< Host NUMA node closest to current thread’s CPU, id is ignored
} hipMemLocationType;
/**
* Specifies a memory location.
@@ -462,37 +464,42 @@ typedef enum hipMemLocationType {
* To specify a gpu, set type = @p hipMemLocationTypeDevice and set id = the gpu's device ID
*/
typedef struct hipMemLocation {
hipMemLocationType type; ///< Specifies the location type, which describes the meaning of id
int id; ///< Identifier for the provided location type @p hipMemLocationType
hipMemLocationType type; ///< Specifies the location type, which describes the meaning of id
int id; ///< Identifier for the provided location type @p hipMemLocationType
} hipMemLocation;
/**
* Flags to specify for copies within a batch. Used with hipMemcpyBatchAsync
*/
typedef enum hipMemcpyFlags {
hipMemcpyFlagDefault = 0x0, ///< Default flag
hipMemcpyFlagPreferOverlapWithCompute = 0x1 ///< Tries to overlap copy with compute work.
hipMemcpyFlagDefault = 0x0, ///< Default flag
hipMemcpyFlagPreferOverlapWithCompute = 0x1 ///< Tries to overlap copy with compute work.
} hipMemcpyFlags;
/**
* Flags to specify order in which source pointer is accessed by Batch memcpy
*/
typedef enum hipMemcpySrcAccessOrder {
hipMemcpySrcAccessOrderInvalid = 0x0, ///< Default Invalid.
hipMemcpySrcAccessOrderStream = 0x1, ///< Access to source pointer must be in stream order.
hipMemcpySrcAccessOrderDuringApiCall = 0x2, ///< Access to source pointer can be out of stream order and all accesses must be complete before API call returns.
hipMemcpySrcAccessOrderAny = 0x3, ///< Access to the source pointer can be out of stream order and the accesses can happen even after the API call return.
hipMemcpySrcAccessOrderMax = 0x7FFFFFFF
hipMemcpySrcAccessOrderInvalid = 0x0, ///< Default Invalid.
hipMemcpySrcAccessOrderStream = 0x1, ///< Access to source pointer must be in stream order.
hipMemcpySrcAccessOrderDuringApiCall =
0x2, ///< Access to source pointer can be out of stream order and all accesses must be
///< complete before API call returns.
hipMemcpySrcAccessOrderAny =
0x3, ///< Access to the source pointer can be out of stream order and the accesses can happen
///< even after the API call return.
hipMemcpySrcAccessOrderMax = 0x7FFFFFFF
} hipMemcpySrcAccessOrder;
/**
* Attributes for copies within a batch.
*/
typedef struct hipMemcpyAttributes {
hipMemcpySrcAccessOrder srcAccessOrder; ///< Source access ordering to be observed for copies with this attribute.
hipMemLocation srcLocHint; ///< Location hint for src operand.
hipMemLocation dstLocHint; ///< Location hint for destination operand.
unsigned int flags; ///< Additional Flags for copies. See hipMemcpyFlags.
hipMemcpySrcAccessOrder
srcAccessOrder; ///< Source access ordering to be observed for copies with this attribute.
hipMemLocation srcLocHint; ///< Location hint for src operand.
hipMemLocation dstLocHint; ///< Location hint for destination operand.
unsigned int flags; ///< Additional Flags for copies. See hipMemcpyFlags.
} hipMemcpyAttributes;
/**
* Operand types for individual copies within a batch
@@ -507,51 +514,50 @@ typedef enum hipMemcpy3DOperandType {
* Struct representing offset into a hipArray_t in elements.
*/
typedef struct hipOffset3D {
size_t x;
size_t y;
size_t z;
size_t x;
size_t y;
size_t z;
} hipOffset3D;
/**
* Struct representing an operand for copy with hipMemcpy3DBatchAsync.
*/
typedef struct hipMemcpy3DOperand {
hipMemcpy3DOperandType type;
union {
struct {
void *ptr;
size_t rowLength; ///< Length of each row in elements.
size_t layerHeight; ///< Height of each layer in elements.
hipMemLocation locHint; ///< Location Hint for the operand.
} ptr;
struct {
hipArray_t array; ///< Array struct for hipMemcpyOperandTypeArray.
hipOffset3D offset; ///< Offset into array in elements.
} array;
} op;
hipMemcpy3DOperandType type;
union {
struct {
void* ptr;
size_t rowLength; ///< Length of each row in elements.
size_t layerHeight; ///< Height of each layer in elements.
hipMemLocation locHint; ///< Location Hint for the operand.
} ptr;
struct {
hipArray_t array; ///< Array struct for hipMemcpyOperandTypeArray.
hipOffset3D offset; ///< Offset into array in elements.
} array;
} op;
} hipMemcpy3DOperand;
/**
* HIP 3D Batch Op
*/
typedef struct hipMemcpy3DBatchOp {
hipMemcpy3DOperand src;
hipMemcpy3DOperand dst;
hipExtent extent;
hipMemcpySrcAccessOrder srcAccessOrder;
unsigned int flags;
hipMemcpy3DOperand src;
hipMemcpy3DOperand dst;
hipExtent extent;
hipMemcpySrcAccessOrder srcAccessOrder;
unsigned int flags;
} hipMemcpy3DBatchOp;
typedef struct hipMemcpy3DPeerParms
{
hipArray_t srcArray; ///< Source memory address
hipPos srcPos; ///< Source position offset
hipPitchedPtr srcPtr; ///< Pitched source memory address
int srcDevice; ///< Source device
hipArray_t dstArray; ///< Destination memory address
hipPos dstPos; ///< Destination position offset
hipPitchedPtr dstPtr; ///< Pitched destination memory address
int dstDevice; ///< Destination device
hipExtent extent; ///< Requested memory copy size
typedef struct hipMemcpy3DPeerParms {
hipArray_t srcArray; ///< Source memory address
hipPos srcPos; ///< Source position offset
hipPitchedPtr srcPtr; ///< Pitched source memory address
int srcDevice; ///< Source device
hipArray_t dstArray; ///< Destination memory address
hipPos dstPos; ///< Destination position offset
hipPitchedPtr dstPtr; ///< Pitched destination memory address
int dstDevice; ///< Destination device
hipExtent extent; ///< Requested memory copy size
} hipMemcpy3DPeerParms;
/**
@@ -564,14 +570,13 @@ typedef struct hipMemcpy3DPeerParms
*
* @returns The created hipPitchedPtr
*/
static inline struct hipPitchedPtr make_hipPitchedPtr(void* d, size_t p, size_t xsz,
size_t ysz) {
struct hipPitchedPtr s;
s.ptr = d;
s.pitch = p;
s.xsize = xsz;
s.ysize = ysz;
return s;
static inline struct hipPitchedPtr make_hipPitchedPtr(void* d, size_t p, size_t xsz, size_t ysz) {
struct hipPitchedPtr s;
s.ptr = d;
s.pitch = p;
s.xsize = xsz;
s.ysize = ysz;
return s;
}
/**
* @brief Make hipPos struct
@@ -583,11 +588,11 @@ static inline struct hipPitchedPtr make_hipPitchedPtr(void* d, size_t p, size_t
* @returns The created hipPos struct
*/
static inline struct hipPos make_hipPos(size_t x, size_t y, size_t z) {
struct hipPos p;
p.x = x;
p.y = y;
p.z = z;
return p;
struct hipPos p;
p.x = x;
p.y = y;
p.z = z;
return p;
}
/**
* @brief Make hipExtent struct
@@ -599,57 +604,69 @@ static inline struct hipPos make_hipPos(size_t x, size_t y, size_t z) {
* @returns The created hipExtent struct
*/
static inline struct hipExtent make_hipExtent(size_t w, size_t h, size_t d) {
struct hipExtent e;
e.width = w;
e.height = h;
e.depth = d;
return e;
struct hipExtent e;
e.width = w;
e.height = h;
e.depth = d;
return e;
}
typedef enum hipFunction_attribute {
HIP_FUNC_ATTRIBUTE_MAX_THREADS_PER_BLOCK, ///< The maximum number of threads per block. Depends on function and device.
HIP_FUNC_ATTRIBUTE_SHARED_SIZE_BYTES, ///< The statically allocated shared memory size in bytes per block required by the function.
HIP_FUNC_ATTRIBUTE_CONST_SIZE_BYTES, ///< The user-allocated constant memory by the function in bytes.
HIP_FUNC_ATTRIBUTE_LOCAL_SIZE_BYTES, ///< The local memory usage of each thread by this function in bytes.
HIP_FUNC_ATTRIBUTE_NUM_REGS, ///< The number of registers used by each thread of this function.
HIP_FUNC_ATTRIBUTE_PTX_VERSION, ///< PTX version
HIP_FUNC_ATTRIBUTE_BINARY_VERSION, ///< Binary version
HIP_FUNC_ATTRIBUTE_CACHE_MODE_CA, ///< Cache mode
HIP_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES, ///< The maximum dynamic shared memory per block for this function in bytes.
HIP_FUNC_ATTRIBUTE_PREFERRED_SHARED_MEMORY_CARVEOUT, ///< The shared memory carveout preference in percent of the maximum shared memory.
HIP_FUNC_ATTRIBUTE_MAX
HIP_FUNC_ATTRIBUTE_MAX_THREADS_PER_BLOCK, ///< The maximum number of threads per block. Depends
///< on function and device.
HIP_FUNC_ATTRIBUTE_SHARED_SIZE_BYTES, ///< The statically allocated shared memory size in bytes
///< per block required by the function.
HIP_FUNC_ATTRIBUTE_CONST_SIZE_BYTES, ///< The user-allocated constant memory by the function in
///< bytes.
HIP_FUNC_ATTRIBUTE_LOCAL_SIZE_BYTES, ///< The local memory usage of each thread by this function
///< in bytes.
HIP_FUNC_ATTRIBUTE_NUM_REGS, ///< The number of registers used by each thread of this function.
HIP_FUNC_ATTRIBUTE_PTX_VERSION, ///< PTX version
HIP_FUNC_ATTRIBUTE_BINARY_VERSION, ///< Binary version
HIP_FUNC_ATTRIBUTE_CACHE_MODE_CA, ///< Cache mode
HIP_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES, ///< The maximum dynamic shared memory per
///< block for this function in bytes.
HIP_FUNC_ATTRIBUTE_PREFERRED_SHARED_MEMORY_CARVEOUT, ///< The shared memory carveout preference
///< in percent of the maximum shared
///< memory.
HIP_FUNC_ATTRIBUTE_MAX
} hipFunction_attribute;
typedef enum hipPointer_attribute {
HIP_POINTER_ATTRIBUTE_CONTEXT = 1, ///< The context on which a pointer was allocated
HIP_POINTER_ATTRIBUTE_CONTEXT = 1, ///< The context on which a pointer was allocated
///< @warning This attribute is not supported in HIP
HIP_POINTER_ATTRIBUTE_MEMORY_TYPE, ///< memory type describing the location of a pointer
HIP_POINTER_ATTRIBUTE_DEVICE_POINTER,///< address at which the pointer is allocated on the device
HIP_POINTER_ATTRIBUTE_HOST_POINTER, ///< address at which the pointer is allocated on the host
HIP_POINTER_ATTRIBUTE_P2P_TOKENS, ///< A pair of tokens for use with Linux kernel interface
HIP_POINTER_ATTRIBUTE_MEMORY_TYPE, ///< memory type describing the location of a pointer
HIP_POINTER_ATTRIBUTE_DEVICE_POINTER, ///< address at which the pointer is allocated on the
///< device
HIP_POINTER_ATTRIBUTE_HOST_POINTER, ///< address at which the pointer is allocated on the host
HIP_POINTER_ATTRIBUTE_P2P_TOKENS, ///< A pair of tokens for use with Linux kernel interface
///< @warning This attribute is not supported in HIP
HIP_POINTER_ATTRIBUTE_SYNC_MEMOPS, ///< Synchronize every synchronous memory operation
HIP_POINTER_ATTRIBUTE_SYNC_MEMOPS, ///< Synchronize every synchronous memory operation
///< initiated on this region
HIP_POINTER_ATTRIBUTE_BUFFER_ID, ///< Unique ID for an allocated memory region
HIP_POINTER_ATTRIBUTE_IS_MANAGED, ///< Indicates if the pointer points to managed memory
HIP_POINTER_ATTRIBUTE_DEVICE_ORDINAL,///< device ordinal of a device on which a pointer
HIP_POINTER_ATTRIBUTE_BUFFER_ID, ///< Unique ID for an allocated memory region
HIP_POINTER_ATTRIBUTE_IS_MANAGED, ///< Indicates if the pointer points to managed memory
HIP_POINTER_ATTRIBUTE_DEVICE_ORDINAL, ///< device ordinal of a device on which a pointer
///< was allocated or registered
HIP_POINTER_ATTRIBUTE_IS_LEGACY_HIP_IPC_CAPABLE, ///< if this pointer maps to an allocation
///< that is suitable for hipIpcGetMemHandle
///< @warning This attribute is not supported in HIP
HIP_POINTER_ATTRIBUTE_RANGE_START_ADDR,///< Starting address for this requested pointer
HIP_POINTER_ATTRIBUTE_RANGE_SIZE, ///< Size of the address range for this requested pointer
HIP_POINTER_ATTRIBUTE_MAPPED, ///< tells if this pointer is in a valid address range
///< that is mapped to a backing allocation
HIP_POINTER_ATTRIBUTE_ALLOWED_HANDLE_TYPES,///< Bitmask of allowed hipmemAllocationHandleType
///< for this allocation @warning This attribute is not supported in HIP
HIP_POINTER_ATTRIBUTE_IS_GPU_DIRECT_RDMA_CAPABLE, ///< returns if the memory referenced by
///< this pointer can be used with the GPUDirect RDMA API
///< @warning This attribute is not supported in HIP
HIP_POINTER_ATTRIBUTE_ACCESS_FLAGS, ///< Returns the access flags the device associated with
///< for the corresponding memory referenced by the ptr
HIP_POINTER_ATTRIBUTE_MEMPOOL_HANDLE ///< Returns the mempool handle for the allocation if
///< it was allocated from a mempool
///< @warning This attribute is not supported in HIP
HIP_POINTER_ATTRIBUTE_IS_LEGACY_HIP_IPC_CAPABLE, ///< if this pointer maps to an allocation
///< that is suitable for hipIpcGetMemHandle
///< @warning This attribute is not supported in
///< HIP
HIP_POINTER_ATTRIBUTE_RANGE_START_ADDR, ///< Starting address for this requested pointer
HIP_POINTER_ATTRIBUTE_RANGE_SIZE, ///< Size of the address range for this requested pointer
HIP_POINTER_ATTRIBUTE_MAPPED, ///< tells if this pointer is in a valid address range
///< that is mapped to a backing allocation
HIP_POINTER_ATTRIBUTE_ALLOWED_HANDLE_TYPES, ///< Bitmask of allowed hipmemAllocationHandleType
///< for this allocation @warning This attribute is
///< not supported in HIP
HIP_POINTER_ATTRIBUTE_IS_GPU_DIRECT_RDMA_CAPABLE, ///< returns if the memory referenced by
///< this pointer can be used with the
///< GPUDirect RDMA API
///< @warning This attribute is not supported
///< in HIP
HIP_POINTER_ATTRIBUTE_ACCESS_FLAGS, ///< Returns the access flags the device associated with
///< for the corresponding memory referenced by the ptr
HIP_POINTER_ATTRIBUTE_MEMPOOL_HANDLE ///< Returns the mempool handle for the allocation if
///< it was allocated from a mempool
///< @warning This attribute is not supported in HIP
} hipPointer_attribute;
// doxygen end DriverTypes
@@ -657,8 +674,8 @@ typedef enum hipPointer_attribute {
* @}
*/
#endif // !defined(__HIPCC_RTC__)
#endif // !defined(__HIPCC_RTC__)
#else
#error("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#error ("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#endif
#endif
projects/hip/include/hip/hip_bf16.h
+1 -1
Ver fichero
@@ -30,7 +30,7 @@ THE SOFTWARE.
#elif !defined(__HIP_PLATFORM_AMD__) && defined(__HIP_PLATFORM_NVIDIA__)
#include <hip/nvidia_detail/nvidia_hip_bf16.h>
#else
#error("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#error ("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#endif
#endif // HIP_INCLUDE_HIP_HIP_BF16_H
projects/hip/include/hip/hip_bfloat16.h
+1 -1
Ver fichero
@@ -38,7 +38,7 @@
#elif !defined(__HIP_PLATFORM_AMD__) && defined(__HIP_PLATFORM_NVIDIA__)
#warning "hip_bfloat16.h is not supported on nvidia platform"
#else
#error("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#error ("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#endif
#endif // _HIP_BFLOAT16_H_
projects/hip/include/hip/hip_common.h
+2 -2
Ver fichero
@@ -55,8 +55,8 @@ THE SOFTWARE.
#endif
#ifdef __GNUC__
#define HIP_PUBLIC_API __attribute__ ((visibility ("default")))
#define HIP_INTERNAL_EXPORTED_API __attribute__ ((visibility ("default")))
#define HIP_PUBLIC_API __attribute__((visibility("default")))
#define HIP_INTERNAL_EXPORTED_API __attribute__((visibility("default")))
#else
#define HIP_PUBLIC_API
#define HIP_INTERNAL_EXPORTED_API
projects/hip/include/hip/hip_complex.h
+1 -1
Ver fichero
@@ -32,7 +32,7 @@ THE SOFTWARE.
#elif !defined(__HIP_PLATFORM_AMD__) && defined(__HIP_PLATFORM_NVIDIA__)
#include <hip/nvidia_detail/nvidia_hip_complex.h>
#else
#error("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#error ("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#endif
#endif
projects/hip/include/hip/hip_cooperative_groups.h
+4 -4
Ver fichero
@@ -27,8 +27,8 @@ THE SOFTWARE.
* feature.
*/
#ifndef HIP_INCLUDE_HIP_HIP_COOPERATIVE_GROUP_H
#define HIP_INCLUDE_HIP_HIP_COOPERATIVE_GROUP_H
#ifndef HIP_INCLUDE_HIP_HIP_COOPERATIVE_GROUP_H
#define HIP_INCLUDE_HIP_HIP_COOPERATIVE_GROUP_H
#include <hip/hip_version.h>
#include <hip/hip_common.h>
@@ -40,7 +40,7 @@ THE SOFTWARE.
#elif !defined(__HIP_PLATFORM_AMD__) && defined(__HIP_PLATFORM_NVIDIA__)
#include <hip/nvidia_detail/nvidia_hip_cooperative_groups.h>
#else
#error("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#error ("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#endif
#endif // HIP_INCLUDE_HIP_HIP_COOPERATIVE_GROUP_H
#endif // HIP_INCLUDE_HIP_HIP_COOPERATIVE_GROUP_H
projects/hip/include/hip/hip_ext.h
+28 -27
Ver fichero
@@ -49,9 +49,9 @@ THE SOFTWARE.
* @param [in] kernelParams pointer to kernel parameters.
* @param [in] extra Pointer to kernel arguments. These are passed directly to the kernel and
* must be in the memory layout and alignment expected by the kernel.
* All passed arguments must be naturally aligned according to their type. The memory address of each
* argument should be a multiple of its size in bytes. Please refer to hip_porting_driver_api.md
* for sample usage.
* All passed arguments must be naturally aligned according to their type. The memory address of
* each argument should be a multiple of its size in bytes. Please refer to
* hip_porting_driver_api.md for sample usage.
* @param [in] startEvent If non-null, specified event will be updated to track the start time of
* the kernel launch. The event must be created before calling this API.
* @param [in] stopEvent If non-null, specified event will be updated to track the stop time of
@@ -68,14 +68,14 @@ THE SOFTWARE.
*
*/
HIP_PUBLIC_API
extern "C" hipError_t hipExtModuleLaunchKernel(hipFunction_t f, uint32_t globalWorkSizeX,
uint32_t globalWorkSizeY, uint32_t globalWorkSizeZ,
uint32_t localWorkSizeX, uint32_t localWorkSizeY,
uint32_t localWorkSizeZ, size_t sharedMemBytes,
hipStream_t hStream, void** kernelParams, void** extra,
hipEvent_t startEvent __dparm(NULL),
hipEvent_t stopEvent __dparm(NULL),
uint32_t flags __dparm(0));
extern "C" hipError_t hipExtModuleLaunchKernel(hipFunction_t f, uint32_t globalWorkSizeX,
uint32_t globalWorkSizeY, uint32_t globalWorkSizeZ,
uint32_t localWorkSizeX, uint32_t localWorkSizeY,
uint32_t localWorkSizeZ, size_t sharedMemBytes,
hipStream_t hStream, void** kernelParams,
void** extra, hipEvent_t startEvent __dparm(NULL),
hipEvent_t stopEvent __dparm(NULL),
uint32_t flags __dparm(0));
/**
* @brief This HIP API is deprecated, please use hipExtModuleLaunchKernel() instead.
*
@@ -83,12 +83,12 @@ HIP_PUBLIC_API
HIP_DEPRECATED("use hipExtModuleLaunchKernel instead")
HIP_PUBLIC_API
extern "C" hipError_t hipHccModuleLaunchKernel(hipFunction_t f, uint32_t globalWorkSizeX,
uint32_t globalWorkSizeY, uint32_t globalWorkSizeZ,
uint32_t localWorkSizeX, uint32_t localWorkSizeY,
uint32_t localWorkSizeZ, size_t sharedMemBytes,
hipStream_t hStream, void** kernelParams, void** extra,
hipEvent_t startEvent __dparm(NULL),
hipEvent_t stopEvent __dparm(NULL));
uint32_t globalWorkSizeY, uint32_t globalWorkSizeZ,
uint32_t localWorkSizeX, uint32_t localWorkSizeY,
uint32_t localWorkSizeZ, size_t sharedMemBytes,
hipStream_t hStream, void** kernelParams,
void** extra, hipEvent_t startEvent __dparm(NULL),
hipEvent_t stopEvent __dparm(NULL));
#if defined(__cplusplus)
@@ -118,7 +118,8 @@ extern "C" hipError_t hipExtLaunchKernel(const void* function_address, dim3 numB
hipEvent_t stopEvent, int flags);
/**
* @brief Launches kernel with dimention parameters and shared memory on stream with templated kernel and arguments.
* @brief Launches kernel with dimention parameters and shared memory on stream with templated
* kernel and arguments.
*
* @param [in] kernel Kernel to launch.
* @param [in] numBlocks const number of blocks.
@@ -141,18 +142,18 @@ inline void hipExtLaunchKernelGGL(F kernel, const dim3& numBlocks, const dim3& d
std::uint32_t sharedMemBytes, hipStream_t stream,
hipEvent_t startEvent, hipEvent_t stopEvent, std::uint32_t flags,
Args... args) {
constexpr size_t count = sizeof...(Args);
auto tup_ = std::tuple<Args...>{args...};
auto tup = validateArgsCountType(kernel, tup_);
void* _Args[count];
pArgs<0>(tup, _Args);
constexpr size_t count = sizeof...(Args);
auto tup_ = std::tuple<Args...>{args...};
auto tup = validateArgsCountType(kernel, tup_);
void* _Args[count];
pArgs<0>(tup, _Args);
auto k = reinterpret_cast<void*>(kernel);
hipExtLaunchKernel(k, numBlocks, dimBlocks, _Args, sharedMemBytes, stream, startEvent,
stopEvent, (int)flags);
auto k = reinterpret_cast<void*>(kernel);
hipExtLaunchKernel(k, numBlocks, dimBlocks, _Args, sharedMemBytes, stream, startEvent, stopEvent,
(int)flags);
}
#endif // defined(__cplusplus)
#endif // defined(__cplusplus)
// doxygen end AMD-specific features
/**
projects/hip/include/hip/hip_fp16.h
+1 -1
Ver fichero
@@ -38,7 +38,7 @@ THE SOFTWARE.
#include "cuda_fp16.h"
#else
#error("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#error ("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#endif
#endif
projects/hip/include/hip/hip_math_constants.h
+1 -1
Ver fichero
@@ -31,6 +31,6 @@ THE SOFTWARE.
#elif !defined(__HIP_PLATFORM_AMD__) && defined(__HIP_PLATFORM_NVIDIA__)
#include "hip/nvidia_detail/nvidia_hip_math_constants.h"
#else
#error("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#error ("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#endif
#endif
projects/hip/include/hip/hip_runtime.h
+4 -4
Ver fichero
@@ -47,8 +47,8 @@ THE SOFTWARE.
#else
#include <stdint.h>
#include <stdlib.h>
#endif // __cplusplus
#endif // !defined(__HIPCC_RTC__)
#endif // __cplusplus
#endif // !defined(__HIPCC_RTC__)
#include <hip/hip_version.h>
#include <hip/hip_common.h>
@@ -58,13 +58,13 @@ THE SOFTWARE.
#elif !defined(__HIP_PLATFORM_AMD__) && defined(__HIP_PLATFORM_NVIDIA__)
#include <hip/nvidia_detail/nvidia_hip_runtime.h>
#else
#error("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#error ("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#endif
#if !defined(__HIPCC_RTC__)
#include <hip/hip_runtime_api.h>
#include <hip/library_types.h>
#endif // !defined(__HIPCC_RTC__)
#endif // !defined(__HIPCC_RTC__)
#include <hip/hip_vector_types.h>
#endif
projects/hip/include/hip/hip_runtime_api.h
+1285 -1360
Ver fichero
La diferencia del archivo ha sido suprimido porque es demasiado grande Cargar Diff
projects/hip/include/hip/hip_vector_types.h
+1 -1
Ver fichero
@@ -35,7 +35,7 @@ THE SOFTWARE.
#elif !defined(__HIP_PLATFORM_AMD__) && defined(__HIP_PLATFORM_NVIDIA__)
#include <vector_types.h>
#else
#error("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#error ("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#endif
#endif
projects/hip/include/hip/hiprtc.h
+125 -102
Ver fichero
@@ -48,106 +48,137 @@ extern "C" {
* @{
*
*/
/**
/**
* hiprtc error code
*/
typedef enum hiprtcResult {
HIPRTC_SUCCESS = 0, ///< Success
HIPRTC_ERROR_OUT_OF_MEMORY = 1, ///< Out of memory
HIPRTC_ERROR_PROGRAM_CREATION_FAILURE = 2, ///< Failed to create program
HIPRTC_ERROR_INVALID_INPUT = 3, ///< Invalid input
HIPRTC_ERROR_INVALID_PROGRAM = 4, ///< Invalid program
HIPRTC_ERROR_INVALID_OPTION = 5, ///< Invalid option
HIPRTC_ERROR_COMPILATION = 6, ///< Compilation error
HIPRTC_ERROR_BUILTIN_OPERATION_FAILURE = 7, ///< Failed in builtin operation
HIPRTC_ERROR_NO_NAME_EXPRESSIONS_AFTER_COMPILATION = 8, ///< No name expression after compilation
HIPRTC_ERROR_NO_LOWERED_NAMES_BEFORE_COMPILATION = 9, ///< No lowered names before compilation
HIPRTC_ERROR_NAME_EXPRESSION_NOT_VALID = 10, ///< Invalid name expression
HIPRTC_ERROR_INTERNAL_ERROR = 11, ///< Internal error
HIPRTC_ERROR_LINKING = 100 ///< Error in linking
HIPRTC_SUCCESS = 0, ///< Success
HIPRTC_ERROR_OUT_OF_MEMORY = 1, ///< Out of memory
HIPRTC_ERROR_PROGRAM_CREATION_FAILURE = 2, ///< Failed to create program
HIPRTC_ERROR_INVALID_INPUT = 3, ///< Invalid input
HIPRTC_ERROR_INVALID_PROGRAM = 4, ///< Invalid program
HIPRTC_ERROR_INVALID_OPTION = 5, ///< Invalid option
HIPRTC_ERROR_COMPILATION = 6, ///< Compilation error
HIPRTC_ERROR_BUILTIN_OPERATION_FAILURE = 7, ///< Failed in builtin operation
HIPRTC_ERROR_NO_NAME_EXPRESSIONS_AFTER_COMPILATION = 8, ///< No name expression after compilation
HIPRTC_ERROR_NO_LOWERED_NAMES_BEFORE_COMPILATION = 9, ///< No lowered names before compilation
HIPRTC_ERROR_NAME_EXPRESSION_NOT_VALID = 10, ///< Invalid name expression
HIPRTC_ERROR_INTERNAL_ERROR = 11, ///< Internal error
HIPRTC_ERROR_LINKING = 100 ///< Error in linking
} hiprtcResult;
/**
* hiprtc JIT option
*/
#define hiprtcJIT_option hipJitOption
#define HIPRTC_JIT_MAX_REGISTERS hipJitOptionMaxRegisters ///< CUDA Only Maximum registers may be used in a
///< thread, passed to compiler
#define HIPRTC_JIT_THREADS_PER_BLOCK hipJitOptionThreadsPerBlock ///< CUDA Only Number of thread per block
#define HIPRTC_JIT_WALL_TIME hipJitOptionWallTime ///< CUDA Only Value for total wall clock time
#define HIPRTC_JIT_INFO_LOG_BUFFER hipJitOptionInfoLogBuffer ///< CUDA Only Pointer to the buffer with
///< logged information
#define HIPRTC_JIT_INFO_LOG_BUFFER_SIZE_BYTES hipJitOptionInfoLogBufferSizeBytes ///< CUDA Only Size of the buffer
///< in bytes for logged info
#define HIPRTC_JIT_ERROR_LOG_BUFFER hipJitOptionErrorLogBuffer ///< CUDA Only Pointer to the buffer
///< with logged error(s)
#define HIPRTC_JIT_ERROR_LOG_BUFFER_SIZE_BYTES hipJitOptionErrorLogBufferSizeBytes ///< CUDA Only Size of the buffer in
///< bytes for logged error(s)
#define HIPRTC_JIT_OPTIMIZATION_LEVEL hipJitOptionOptimizationLevel ///< Value of optimization level for
///< generated codes, acceptable
///< options -O0, -O1, -O2, -O3
#define HIPRTC_JIT_TARGET_FROM_HIPCONTEXT hipJitOptionTargetFromContext ///< CUDA Only The target context,
///< which is the default
#define HIPRTC_JIT_TARGET hipJitOptionTarget ///< CUDA Only JIT target
#define HIPRTC_JIT_FALLBACK_STRATEGY hipJitOptionFallbackStrategy ///< CUDA Only Fallback strategy
#define HIPRTC_JIT_GENERATE_DEBUG_INFO hipJitOptionGenerateDebugInfo ///< CUDA Only Generate debug information
#define HIPRTC_JIT_LOG_VERBOSE hipJitOptionLogVerbose ///< CUDA Only Generate log verbose
#define HIPRTC_JIT_GENERATE_LINE_INFO hipJitOptionGenerateLineInfo ///< CUDA Only Generate line number information
#define HIPRTC_JIT_CACHE_MODE hipJitOptionCacheMode ///< CUDA Only Set cache mode
#define HIPRTC_JIT_NEW_SM3X_OPT hipJitOptionSm3xOpt ///< @deprecated CUDA Only New SM3X option.
#define HIPRTC_JIT_FAST_COMPILE hipJitOptionFastCompile ///< CUDA Only Set fast compile
#define HIPRTC_JIT_GLOBAL_SYMBOL_NAMES hipJitOptionGlobalSymbolNames ///< CUDA Only Array of device symbol names to be
///< relocated to the host
#define HIPRTC_JIT_GLOBAL_SYMBOL_ADDRESS hipJitOptionGlobalSymbolAddresses ///< CUDA Only Array of host addresses to be
///< relocated to the device
#define HIPRTC_JIT_GLOBAL_SYMBOL_COUNT hipJitOptionGlobalSymbolCount ///< CUDA Only Number of symbol count.
#define HIPRTC_JIT_LTO hipJitOptionLto ///< @deprecated CUDA Only Enable link-time
///< optimization for device code
#define HIPRTC_JIT_FTZ hipJitOptionFtz ///< @deprecated CUDA Only Set
///< single-precision denormals.
#define HIPRTC_JIT_PREC_DIV hipJitOptionPrecDiv ///< @deprecated CUDA Only Set
///< single-precision floating-point division
///< and reciprocals
#define HIPRTC_JIT_PREC_SQRT hipJitOptionPrecSqrt ///< @deprecated CUDA Only Set
///< single-precision floating-point
///< square root
#define HIPRTC_JIT_FMA hipJitOptionFma ///< @deprecated CUDA Only Enable
///< floating-point multiplies and
///< adds/subtracts operations
#define HIPRTC_JIT_POSITION_INDEPENDENT_CODE hipJitOptionPositionIndependentCode ///< CUDA Only Generates
///< Position Independent code
#define HIPRTC_JIT_MIN_CTA_PER_SM hipJitOptionMinCTAPerSM ///< CUDA Only Hints to JIT compiler
///< the minimum number of CTAs frin
///< kernel's grid to be mapped to SM
#define HIPRTC_JIT_MAX_THREADS_PER_BLOCK hipJitOptionMaxThreadsPerBlock ///< CUDA only Maximum number of
///< threads in a thread block
#define HIPRTC_JIT_OVERRIDE_DIRECT_VALUES hipJitOptionOverrideDirectiveValues ///< CUDA only Override Directive
///< Values
#define HIPRTC_JIT_NUM_OPTIONS hipJitOptionNumOptions ///< Number of options
#define HIPRTC_JIT_IR_TO_ISA_OPT_EXT hipJitOptionIRtoISAOptExt ///< HIP Only Linker options to be
///< passed on to compiler
#define HIPRTC_JIT_IR_TO_ISA_OPT_COUNT_EXT hipJitOptionIRtoISAOptCountExt ///< HIP Only Count of linker options
///< to be passed on to
#define HIPRTC_JIT_MAX_REGISTERS \
hipJitOptionMaxRegisters ///< CUDA Only Maximum registers may be used in a
///< thread, passed to compiler
#define HIPRTC_JIT_THREADS_PER_BLOCK \
hipJitOptionThreadsPerBlock ///< CUDA Only Number of thread per block
#define HIPRTC_JIT_WALL_TIME hipJitOptionWallTime ///< CUDA Only Value for total wall clock time
#define HIPRTC_JIT_INFO_LOG_BUFFER \
hipJitOptionInfoLogBuffer ///< CUDA Only Pointer to the buffer with
///< logged information
#define HIPRTC_JIT_INFO_LOG_BUFFER_SIZE_BYTES \
hipJitOptionInfoLogBufferSizeBytes ///< CUDA Only Size of the buffer
///< in bytes for logged info
#define HIPRTC_JIT_ERROR_LOG_BUFFER \
hipJitOptionErrorLogBuffer ///< CUDA Only Pointer to the buffer
///< with logged error(s)
#define HIPRTC_JIT_ERROR_LOG_BUFFER_SIZE_BYTES \
hipJitOptionErrorLogBufferSizeBytes ///< CUDA Only Size of the buffer in
///< bytes for logged error(s)
#define HIPRTC_JIT_OPTIMIZATION_LEVEL \
hipJitOptionOptimizationLevel ///< Value of optimization level for
///< generated codes, acceptable
///< options -O0, -O1, -O2, -O3
#define HIPRTC_JIT_TARGET_FROM_HIPCONTEXT \
hipJitOptionTargetFromContext ///< CUDA Only The target context,
///< which is the default
#define HIPRTC_JIT_TARGET hipJitOptionTarget ///< CUDA Only JIT target
#define HIPRTC_JIT_FALLBACK_STRATEGY hipJitOptionFallbackStrategy ///< CUDA Only Fallback strategy
#define HIPRTC_JIT_GENERATE_DEBUG_INFO \
hipJitOptionGenerateDebugInfo ///< CUDA Only Generate debug information
#define HIPRTC_JIT_LOG_VERBOSE hipJitOptionLogVerbose ///< CUDA Only Generate log verbose
#define HIPRTC_JIT_GENERATE_LINE_INFO \
hipJitOptionGenerateLineInfo ///< CUDA Only Generate line number information
#define HIPRTC_JIT_CACHE_MODE hipJitOptionCacheMode ///< CUDA Only Set cache mode
#define HIPRTC_JIT_NEW_SM3X_OPT hipJitOptionSm3xOpt ///< @deprecated CUDA Only New SM3X option.
#define HIPRTC_JIT_FAST_COMPILE hipJitOptionFastCompile ///< CUDA Only Set fast compile
#define HIPRTC_JIT_GLOBAL_SYMBOL_NAMES \
hipJitOptionGlobalSymbolNames ///< CUDA Only Array of device symbol names to be
///< relocated to the host
#define HIPRTC_JIT_GLOBAL_SYMBOL_ADDRESS \
hipJitOptionGlobalSymbolAddresses ///< CUDA Only Array of host addresses to be
///< relocated to the device
#define HIPRTC_JIT_GLOBAL_SYMBOL_COUNT \
hipJitOptionGlobalSymbolCount ///< CUDA Only Number of symbol count.
#define HIPRTC_JIT_LTO \
hipJitOptionLto ///< @deprecated CUDA Only Enable link-time
///< optimization for device code
#define HIPRTC_JIT_FTZ \
hipJitOptionFtz ///< @deprecated CUDA Only Set
///< single-precision denormals.
#define HIPRTC_JIT_PREC_DIV \
hipJitOptionPrecDiv ///< @deprecated CUDA Only Set
///< single-precision floating-point division
///< and reciprocals
#define HIPRTC_JIT_PREC_SQRT \
hipJitOptionPrecSqrt ///< @deprecated CUDA Only Set
///< single-precision floating-point
///< square root
#define HIPRTC_JIT_FMA \
hipJitOptionFma ///< @deprecated CUDA Only Enable
///< floating-point multiplies and
///< adds/subtracts operations
#define HIPRTC_JIT_POSITION_INDEPENDENT_CODE \
hipJitOptionPositionIndependentCode ///< CUDA Only Generates
///< Position Independent code
#define HIPRTC_JIT_MIN_CTA_PER_SM \
hipJitOptionMinCTAPerSM ///< CUDA Only Hints to JIT compiler
///< the minimum number of CTAs frin
///< kernel's grid to be mapped to SM
#define HIPRTC_JIT_MAX_THREADS_PER_BLOCK \
hipJitOptionMaxThreadsPerBlock ///< CUDA only Maximum number of
///< threads in a thread block
#define HIPRTC_JIT_OVERRIDE_DIRECT_VALUES \
hipJitOptionOverrideDirectiveValues ///< CUDA only Override Directive
///< Values
#define HIPRTC_JIT_NUM_OPTIONS hipJitOptionNumOptions ///< Number of options
#define HIPRTC_JIT_IR_TO_ISA_OPT_EXT \
hipJitOptionIRtoISAOptExt ///< HIP Only Linker options to be
///< passed on to compiler
#define HIPRTC_JIT_IR_TO_ISA_OPT_COUNT_EXT \
hipJitOptionIRtoISAOptCountExt ///< HIP Only Count of linker options
///< to be passed on to
/**
* hiprtc JIT input type
*/
#define hiprtcJITInputType hipJitInputType
#define HIPRTC_JIT_INPUT_CUBIN hipJitInputCubin ///< Cuda only Input Cubin
#define HIPRTC_JIT_INPUT_PTX hipJitInputPtx ///< Cuda only Input PTX
#define HIPRTC_JIT_INPUT_FATBINARY hipJitInputFatBinary ///< Cuda Only Input FAT Binary
#define HIPRTC_JIT_INPUT_OBJECT hipJitInputObject ///< Cuda Only Host Object with embedded device code
#define HIPRTC_JIT_INPUT_LIBRARY hipJitInputLibrary ///< Cuda Only Archive of Host Objects with embedded device code
#define HIPRTC_JIT_INPUT_NVVM hipJitInputNvvm ///< @deprecated CUDA only High Level intermediate code for LTO
#define HIPRTC_JIT_NUM_LEGACY_INPUT_TYPES hipJitNumLegacyInputTypes ///< Count of Legacy Input Types
#define HIPRTC_JIT_INPUT_LLVM_BITCODE hipJitInputLLVMBitcode ///< HIP Only LLVM Bitcode or IR assembly
#define HIPRTC_JIT_INPUT_LLVM_BUNDLED_BITCODE hipJitInputLLVMBundledBitcode ///< HIP Only LLVM Clang Bundled Code
#define HIPRTC_JIT_INPUT_LLVM_ARCHIVES_OF_BUNDLED_BITCODE hipJitInputLLVMArchivesOfBundledBitcode ///< HIP Only LLVM
///< Archives of
///< Bundled Bitcode
#define HIPRTC_JIT_INPUT_CUBIN hipJitInputCubin ///< Cuda only Input Cubin
#define HIPRTC_JIT_INPUT_PTX hipJitInputPtx ///< Cuda only Input PTX
#define HIPRTC_JIT_INPUT_FATBINARY hipJitInputFatBinary ///< Cuda Only Input FAT Binary
#define HIPRTC_JIT_INPUT_OBJECT \
hipJitInputObject ///< Cuda Only Host Object with embedded device code
#define HIPRTC_JIT_INPUT_LIBRARY \
hipJitInputLibrary ///< Cuda Only Archive of Host Objects with embedded device code
#define HIPRTC_JIT_INPUT_NVVM \
hipJitInputNvvm ///< @deprecated CUDA only High Level intermediate code for LTO
#define HIPRTC_JIT_NUM_LEGACY_INPUT_TYPES \
hipJitNumLegacyInputTypes ///< Count of Legacy Input Types
#define HIPRTC_JIT_INPUT_LLVM_BITCODE \
hipJitInputLLVMBitcode ///< HIP Only LLVM Bitcode or IR assembly
#define HIPRTC_JIT_INPUT_LLVM_BUNDLED_BITCODE \
hipJitInputLLVMBundledBitcode ///< HIP Only LLVM Clang Bundled Code
#define HIPRTC_JIT_INPUT_LLVM_ARCHIVES_OF_BUNDLED_BITCODE \
hipJitInputLLVMArchivesOfBundledBitcode ///< HIP Only LLVM
///< Archives of
///< Bundled Bitcode
#define HIPRTC_JIT_INPUT_SPIRV hipJitInputSpirv ///< HIP Only SPIRV Code Object
#define HIPRTC_JIT_NUM_INPUT_TYPES hipJitNumInputTypes ///< Count of Input Types
/**
* @}
*/
* @}
*/
/**
* hiprtc link state
@@ -199,8 +230,7 @@ typedef struct _hiprtcProgram* hiprtcProgram;
*
* @see hiprtcResult
*/
hiprtcResult hiprtcAddNameExpression(hiprtcProgram prog,
const char* name_expression);
hiprtcResult hiprtcAddNameExpression(hiprtcProgram prog, const char* name_expression);
/**
* @ingroup Runtime
@@ -216,9 +246,7 @@ hiprtcResult hiprtcAddNameExpression(hiprtcProgram prog,
*
* @see hiprtcResult
*/
hiprtcResult hiprtcCompileProgram(hiprtcProgram prog,
int numOptions,
const char* const* options);
hiprtcResult hiprtcCompileProgram(hiprtcProgram prog, int numOptions, const char* const* options);
/**
* @ingroup Runtime
@@ -240,11 +268,8 @@ hiprtcResult hiprtcCompileProgram(hiprtcProgram prog,
*
* @see hiprtcResult
*/
hiprtcResult hiprtcCreateProgram(hiprtcProgram* prog,
const char* src,
const char* name,
int numHeaders,
const char* const* headers,
hiprtcResult hiprtcCreateProgram(hiprtcProgram* prog, const char* src, const char* name,
int numHeaders, const char* const* headers,
const char* const* includeNames);
/**
@@ -260,8 +285,8 @@ hiprtcResult hiprtcCreateProgram(hiprtcProgram* prog,
hiprtcResult hiprtcDestroyProgram(hiprtcProgram* prog);
/**
* @brief Gets the lowered (mangled) name from an instance of hiprtcProgram with the given input parameters,
* and sets the output lowered_name with it.
* @brief Gets the lowered (mangled) name from an instance of hiprtcProgram with the given input
* parameters, and sets the output lowered_name with it.
* @ingroup Runtime
* @param [in] prog runtime compilation program instance.
* @param [in] name_expression const char pointer to the name expression.
@@ -276,8 +301,7 @@ hiprtcResult hiprtcDestroyProgram(hiprtcProgram* prog);
*
* @see hiprtcResult
*/
hiprtcResult hiprtcGetLoweredName(hiprtcProgram prog,
const char* name_expression,
hiprtcResult hiprtcGetLoweredName(hiprtcProgram prog, const char* name_expression,
const char** lowered_name);
/**
@@ -300,8 +324,7 @@ hiprtcResult hiprtcGetProgramLog(hiprtcProgram prog, char* log);
*
* @see hiprtcResult
*/
hiprtcResult hiprtcGetProgramLogSize(hiprtcProgram prog,
size_t* logSizeRet);
hiprtcResult hiprtcGetProgramLogSize(hiprtcProgram prog, size_t* logSizeRet);
/**
* @brief Gets the pointer of compilation binary by the runtime compilation program instance.
@@ -446,5 +469,5 @@ hiprtcResult hiprtcLinkDestroy(hiprtcLinkState hip_link_state);
#endif /* __cplusplus */
#else
#error("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#error ("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#endif
projects/hip/include/hip/library_types.h
+32 -32
Ver fichero
@@ -30,40 +30,40 @@ THE SOFTWARE.
#if defined(__HIP_PLATFORM_AMD__) && !defined(__HIP_PLATFORM_NVIDIA__)
typedef enum hipDataType {
HIP_R_32F = 0,
HIP_R_64F = 1,
HIP_R_16F = 2,
HIP_R_8I = 3,
HIP_C_32F = 4,
HIP_C_64F = 5,
HIP_C_16F = 6,
HIP_C_8I = 7,
HIP_R_8U = 8,
HIP_C_8U = 9,
HIP_R_32I = 10,
HIP_C_32I = 11,
HIP_R_32U = 12,
HIP_C_32U = 13,
HIP_R_16BF = 14,
HIP_C_16BF = 15,
HIP_R_4I = 16,
HIP_C_4I = 17,
HIP_R_4U = 18,
HIP_C_4U = 19,
HIP_R_16I = 20,
HIP_C_16I = 21,
HIP_R_16U = 22,
HIP_C_16U = 23,
HIP_R_64I = 24,
HIP_C_64I = 25,
HIP_R_64U = 26,
HIP_C_64U = 27,
HIP_R_32F = 0,
HIP_R_64F = 1,
HIP_R_16F = 2,
HIP_R_8I = 3,
HIP_C_32F = 4,
HIP_C_64F = 5,
HIP_C_16F = 6,
HIP_C_8I = 7,
HIP_R_8U = 8,
HIP_C_8U = 9,
HIP_R_32I = 10,
HIP_C_32I = 11,
HIP_R_32U = 12,
HIP_C_32U = 13,
HIP_R_16BF = 14,
HIP_C_16BF = 15,
HIP_R_4I = 16,
HIP_C_4I = 17,
HIP_R_4U = 18,
HIP_C_4U = 19,
HIP_R_16I = 20,
HIP_C_16I = 21,
HIP_R_16U = 22,
HIP_C_16U = 23,
HIP_R_64I = 24,
HIP_C_64I = 25,
HIP_R_64U = 26,
HIP_C_64U = 27,
HIP_R_8F_E4M3 = 28,
HIP_R_8F_E5M2 = 29,
HIP_R_8F_UE8M0 = 30,
HIP_R_6F_E2M3 = 31,
HIP_R_6F_E3M2 = 32,
HIP_R_4F_E2M1 = 33,
HIP_R_6F_E2M3 = 31,
HIP_R_6F_E3M2 = 32,
HIP_R_4F_E2M1 = 33,
// HIP specific Data Types
HIP_R_8F_E4M3_FNUZ = 1000,
HIP_R_8F_E5M2_FNUZ = 1001,
@@ -78,7 +78,7 @@ typedef enum hipLibraryPropertyType {
#elif !defined(__HIP_PLATFORM_AMD__) && defined(__HIP_PLATFORM_NVIDIA__)
#include "library_types.h"
#else
#error("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#error ("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#endif
#endif
projects/hip/include/hip/linker_types.h
+61 -58
Ver fichero
@@ -45,77 +45,80 @@ THE SOFTWARE.
* hipJitOption
*/
typedef enum hipJitOption {
hipJitOptionMaxRegisters = 0, ///< CUDA Only Maximum registers may be used in a thread,
///< passed to compiler
hipJitOptionThreadsPerBlock, ///< CUDA Only Number of thread per block
hipJitOptionWallTime, ///< CUDA Only Value for total wall clock time
hipJitOptionInfoLogBuffer, ///< CUDA Only Pointer to the buffer with logged information
hipJitOptionInfoLogBufferSizeBytes, ///< CUDA Only Size of the buffer in bytes for logged info
hipJitOptionErrorLogBuffer, ///< CUDA Only Pointer to the buffer with logged error(s)
hipJitOptionErrorLogBufferSizeBytes, ///< CUDA Only Size of the buffer in bytes for logged error(s)
hipJitOptionOptimizationLevel, ///< Value of optimization level for generated codes, acceptable options
///< -O0, -O1, -O2, -O3
hipJitOptionTargetFromContext, ///< CUDA Only The target context, which is the default
hipJitOptionTarget, ///< CUDA Only JIT target
hipJitOptionFallbackStrategy, ///< CUDA Only Fallback strategy
hipJitOptionGenerateDebugInfo, ///< CUDA Only Generate debug information
hipJitOptionLogVerbose, ///< CUDA Only Generate log verbose
hipJitOptionGenerateLineInfo, ///< CUDA Only Generate line number information
hipJitOptionCacheMode, ///< CUDA Only Set cache mode
hipJitOptionSm3xOpt, ///< @deprecated CUDA Only New SM3X option.
hipJitOptionFastCompile, ///< CUDA Only Set fast compile
hipJitOptionGlobalSymbolNames, ///< CUDA Only Array of device symbol names to be relocated to the host
hipJitOptionGlobalSymbolAddresses, ///< CUDA Only Array of host addresses to be relocated to the device
hipJitOptionGlobalSymbolCount, ///< CUDA Only Number of symbol count.
hipJitOptionLto, ///< @deprecated CUDA Only Enable link-time optimization for device code
hipJitOptionFtz, ///< @deprecated CUDA Only Set single-precision denormals.
hipJitOptionPrecDiv, ///< @deprecated CUDA Only Set single-precision floating-point division
///< and reciprocals
hipJitOptionPrecSqrt, ///< @deprecated CUDA Only Set single-precision floating-point square root
hipJitOptionFma, ///< @deprecated CUDA Only Enable floating-point multiplies and
///< adds/subtracts operations
hipJitOptionPositionIndependentCode, ///< CUDA Only Generates Position Independent code
hipJitOptionMinCTAPerSM, ///< CUDA Only Hints to JIT compiler the minimum number of CTAs frin kernel's
///< grid to be mapped to SM
hipJitOptionMaxThreadsPerBlock, ///< CUDA only Maximum number of threads in a thread block
hipJitOptionOverrideDirectiveValues, ///< Cuda only Override Directive values
hipJitOptionNumOptions, ///< Number of options
hipJitOptionIRtoISAOptExt = 10000, ///< Hip Only Linker options to be passed on to compiler
hipJitOptionIRtoISAOptCountExt, ///< Hip Only Count of linker options to be passed on to compiler
hipJitOptionMaxRegisters = 0, ///< CUDA Only Maximum registers may be used in a thread,
///< passed to compiler
hipJitOptionThreadsPerBlock, ///< CUDA Only Number of thread per block
hipJitOptionWallTime, ///< CUDA Only Value for total wall clock time
hipJitOptionInfoLogBuffer, ///< CUDA Only Pointer to the buffer with logged information
hipJitOptionInfoLogBufferSizeBytes, ///< CUDA Only Size of the buffer in bytes for logged info
hipJitOptionErrorLogBuffer, ///< CUDA Only Pointer to the buffer with logged error(s)
hipJitOptionErrorLogBufferSizeBytes, ///< CUDA Only Size of the buffer in bytes for logged
///< error(s)
hipJitOptionOptimizationLevel, ///< Value of optimization level for generated codes, acceptable
///< options -O0, -O1, -O2, -O3
hipJitOptionTargetFromContext, ///< CUDA Only The target context, which is the default
hipJitOptionTarget, ///< CUDA Only JIT target
hipJitOptionFallbackStrategy, ///< CUDA Only Fallback strategy
hipJitOptionGenerateDebugInfo, ///< CUDA Only Generate debug information
hipJitOptionLogVerbose, ///< CUDA Only Generate log verbose
hipJitOptionGenerateLineInfo, ///< CUDA Only Generate line number information
hipJitOptionCacheMode, ///< CUDA Only Set cache mode
hipJitOptionSm3xOpt, ///< @deprecated CUDA Only New SM3X option.
hipJitOptionFastCompile, ///< CUDA Only Set fast compile
hipJitOptionGlobalSymbolNames, ///< CUDA Only Array of device symbol names to be relocated to the
///< host
hipJitOptionGlobalSymbolAddresses, ///< CUDA Only Array of host addresses to be relocated to the
///< device
hipJitOptionGlobalSymbolCount, ///< CUDA Only Number of symbol count.
hipJitOptionLto, ///< @deprecated CUDA Only Enable link-time optimization for device code
hipJitOptionFtz, ///< @deprecated CUDA Only Set single-precision denormals.
hipJitOptionPrecDiv, ///< @deprecated CUDA Only Set single-precision floating-point division
///< and reciprocals
hipJitOptionPrecSqrt, ///< @deprecated CUDA Only Set single-precision floating-point square root
hipJitOptionFma, ///< @deprecated CUDA Only Enable floating-point multiplies and
///< adds/subtracts operations
hipJitOptionPositionIndependentCode, ///< CUDA Only Generates Position Independent code
hipJitOptionMinCTAPerSM, ///< CUDA Only Hints to JIT compiler the minimum number of CTAs frin
///< kernel's grid to be mapped to SM
hipJitOptionMaxThreadsPerBlock, ///< CUDA only Maximum number of threads in a thread block
hipJitOptionOverrideDirectiveValues, ///< Cuda only Override Directive values
hipJitOptionNumOptions, ///< Number of options
hipJitOptionIRtoISAOptExt = 10000, ///< Hip Only Linker options to be passed on to compiler
hipJitOptionIRtoISAOptCountExt, ///< Hip Only Count of linker options to be passed on to compiler
} hipJitOption;
/**
* hipJitInputType
*/
typedef enum hipJitInputType {
hipJitInputCubin = 0, ///< Cuda only Input cubin
hipJitInputPtx, ///< Cuda only Input PTX
hipJitInputFatBinary, ///< Cuda Only Input FAT Binary
hipJitInputObject, ///< Cuda Only Host Object with embedded device code
hipJitInputLibrary, ///< Cuda Only Archive of Host Objects with embedded
///< device code
hipJitInputNvvm, ///< @deprecated Cuda only High Level intermediate
///< code for LTO
hipJitNumLegacyInputTypes, ///< Count of Legacy Input Types
hipJitInputLLVMBitcode = 100, ///< HIP Only LLVM Bitcode or IR assembly
hipJitInputLLVMBundledBitcode = 101, ///< HIP Only LLVM Clang Bundled Code
hipJitInputLLVMArchivesOfBundledBitcode = 102, ///< HIP Only LLVM Archive of Bundled Bitcode
hipJitInputSpirv = 103, ///< HIP Only SPIRV Code Object
hipJitNumInputTypes = 10 ///< Count of Input Types
hipJitInputCubin = 0, ///< Cuda only Input cubin
hipJitInputPtx, ///< Cuda only Input PTX
hipJitInputFatBinary, ///< Cuda Only Input FAT Binary
hipJitInputObject, ///< Cuda Only Host Object with embedded device code
hipJitInputLibrary, ///< Cuda Only Archive of Host Objects with embedded
///< device code
hipJitInputNvvm, ///< @deprecated Cuda only High Level intermediate
///< code for LTO
hipJitNumLegacyInputTypes, ///< Count of Legacy Input Types
hipJitInputLLVMBitcode = 100, ///< HIP Only LLVM Bitcode or IR assembly
hipJitInputLLVMBundledBitcode = 101, ///< HIP Only LLVM Clang Bundled Code
hipJitInputLLVMArchivesOfBundledBitcode = 102, ///< HIP Only LLVM Archive of Bundled Bitcode
hipJitInputSpirv = 103, ///< HIP Only SPIRV Code Object
hipJitNumInputTypes = 10 ///< Count of Input Types
} hipJitInputType;
/**
* hipJitCacheMode
*/
typedef enum hipJitCacheMode {
hipJitCacheOptionNone = 0,
hipJitCacheOptionCG,
hipJitCacheOptionCA
hipJitCacheOptionNone = 0,
hipJitCacheOptionCG,
hipJitCacheOptionCA
} hipJitCacheMode;
/**
* hipJitFallback
*/
typedef enum hipJitFallback {
hipJitPreferPTX = 0,
hipJitPreferBinary,
hipJitPreferPTX = 0,
hipJitPreferBinary,
} hipJitFallback;
// doxygen end LinkerTypes
@@ -124,7 +127,7 @@ typedef enum hipJitFallback {
*/
#else
#error("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#error ("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#endif
#endif // HIP_INCLUDE_HIP_LINKER_TYPES_H
#endif // HIP_INCLUDE_HIP_LINKER_TYPES_H
projects/hip/include/hip/math_functions.h
+2 -2
Ver fichero
@@ -34,9 +34,9 @@ THE SOFTWARE.
#if defined(__HIP_PLATFORM_AMD__) && !defined(__HIP_PLATFORM_NVIDIA__)
#include <hip/amd_detail/amd_math_functions.h>
#elif !defined(__HIP_PLATFORM_AMD__) && defined(__HIP_PLATFORM_NVIDIA__)
//#include <hip/nvidia_detail/math_functions.h>
// #include <hip/nvidia_detail/math_functions.h>
#else
#error("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#error ("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#endif
#endif
projects/hip/include/hip/surface_types.h
+4 -4
Ver fichero
@@ -44,16 +44,16 @@ typedef struct __hip_surface* hipSurfaceObject_t;
* hip surface reference
*/
struct surfaceReference {
hipSurfaceObject_t surfaceObject;
hipSurfaceObject_t surfaceObject;
};
/**
* hip surface boundary modes
*/
enum hipSurfaceBoundaryMode {
hipBoundaryModeZero = 0,
hipBoundaryModeTrap = 1,
hipBoundaryModeClamp = 2
hipBoundaryModeZero = 0,
hipBoundaryModeTrap = 1,
hipBoundaryModeClamp = 2
};
#if defined(__clang__)
projects/hip/include/hip/texture_types.h
+65 -65
Ver fichero
@@ -45,7 +45,7 @@ THE SOFTWARE.
#if !defined(__HIPCC_RTC__)
#include <hip/channel_descriptor.h>
#include <hip/driver_types.h>
#endif // !defined(__HIPCC_RTC__)
#endif // !defined(__HIPCC_RTC__)
#define hipTextureType1D 0x01
#define hipTextureType2D 0x02
@@ -73,10 +73,10 @@ typedef struct __hip_texture* hipTextureObject_t;
* hip texture address modes
*/
enum hipTextureAddressMode {
hipAddressModeWrap = 0,
hipAddressModeClamp = 1,
hipAddressModeMirror = 2,
hipAddressModeBorder = 3
hipAddressModeWrap = 0,
hipAddressModeClamp = 1,
hipAddressModeMirror = 2,
hipAddressModeBorder = 3
};
/**
@@ -93,39 +93,39 @@ enum hipTextureReadMode { hipReadModeElementType = 0, hipReadModeNormalizedFloat
* hip texture reference
*/
typedef struct textureReference {
int normalized;
enum hipTextureReadMode readMode;// used only for driver API's
enum hipTextureFilterMode filterMode;
enum hipTextureAddressMode addressMode[3]; // Texture address mode for up to 3 dimensions
struct hipChannelFormatDesc channelDesc;
int sRGB; // Perform sRGB->linear conversion during texture read
unsigned int maxAnisotropy; // Limit to the anisotropy ratio
enum hipTextureFilterMode mipmapFilterMode;
float mipmapLevelBias;
float minMipmapLevelClamp;
float maxMipmapLevelClamp;
int normalized;
enum hipTextureReadMode readMode; // used only for driver API's
enum hipTextureFilterMode filterMode;
enum hipTextureAddressMode addressMode[3]; // Texture address mode for up to 3 dimensions
struct hipChannelFormatDesc channelDesc;
int sRGB; // Perform sRGB->linear conversion during texture read
unsigned int maxAnisotropy; // Limit to the anisotropy ratio
enum hipTextureFilterMode mipmapFilterMode;
float mipmapLevelBias;
float minMipmapLevelClamp;
float maxMipmapLevelClamp;
hipTextureObject_t textureObject;
int numChannels;
enum hipArray_Format format;
}textureReference;
hipTextureObject_t textureObject;
int numChannels;
enum hipArray_Format format;
} textureReference;
/**
* hip texture descriptor
*/
typedef struct hipTextureDesc {
enum hipTextureAddressMode addressMode[3]; // Texture address mode for up to 3 dimensions
enum hipTextureFilterMode filterMode;
enum hipTextureReadMode readMode;
int sRGB; // Perform sRGB->linear conversion during texture read
float borderColor[4];
int normalizedCoords;
unsigned int maxAnisotropy;
enum hipTextureFilterMode mipmapFilterMode;
float mipmapLevelBias;
float minMipmapLevelClamp;
float maxMipmapLevelClamp;
}hipTextureDesc;
enum hipTextureAddressMode addressMode[3]; // Texture address mode for up to 3 dimensions
enum hipTextureFilterMode filterMode;
enum hipTextureReadMode readMode;
int sRGB; // Perform sRGB->linear conversion during texture read
float borderColor[4];
int normalizedCoords;
unsigned int maxAnisotropy;
enum hipTextureFilterMode mipmapFilterMode;
float mipmapLevelBias;
float minMipmapLevelClamp;
float maxMipmapLevelClamp;
} hipTextureDesc;
#if __cplusplus
@@ -145,45 +145,45 @@ typedef textureReference* hipTexRef;
template <class T, int texType = hipTextureType1D,
enum hipTextureReadMode mode = hipReadModeElementType>
struct __HIP_TEXTURE_ATTRIB texture : public textureReference {
texture(int norm = 0, enum hipTextureFilterMode fMode = hipFilterModePoint,
enum hipTextureAddressMode aMode = hipAddressModeClamp) {
normalized = norm;
readMode = mode;
filterMode = fMode;
addressMode[0] = aMode;
addressMode[1] = aMode;
addressMode[2] = aMode;
channelDesc = hipCreateChannelDesc<T>();
sRGB = 0;
textureObject = nullptr;
maxAnisotropy = 0;
mipmapLevelBias = 0;
minMipmapLevelClamp = 0;
maxMipmapLevelClamp = 0;
}
texture(int norm = 0, enum hipTextureFilterMode fMode = hipFilterModePoint,
enum hipTextureAddressMode aMode = hipAddressModeClamp) {
normalized = norm;
readMode = mode;
filterMode = fMode;
addressMode[0] = aMode;
addressMode[1] = aMode;
addressMode[2] = aMode;
channelDesc = hipCreateChannelDesc<T>();
sRGB = 0;
textureObject = nullptr;
maxAnisotropy = 0;
mipmapLevelBias = 0;
minMipmapLevelClamp = 0;
maxMipmapLevelClamp = 0;
}
texture(int norm, enum hipTextureFilterMode fMode, enum hipTextureAddressMode aMode,
struct hipChannelFormatDesc desc) {
normalized = norm;
readMode = mode;
filterMode = fMode;
addressMode[0] = aMode;
addressMode[1] = aMode;
addressMode[2] = aMode;
channelDesc = desc;
sRGB = 0;
textureObject = nullptr;
maxAnisotropy = 0;
mipmapLevelBias = 0;
minMipmapLevelClamp = 0;
maxMipmapLevelClamp = 0;
}
texture(int norm, enum hipTextureFilterMode fMode, enum hipTextureAddressMode aMode,
struct hipChannelFormatDesc desc) {
normalized = norm;
readMode = mode;
filterMode = fMode;
addressMode[0] = aMode;
addressMode[1] = aMode;
addressMode[2] = aMode;
channelDesc = desc;
sRGB = 0;
textureObject = nullptr;
maxAnisotropy = 0;
mipmapLevelBias = 0;
minMipmapLevelClamp = 0;
maxMipmapLevelClamp = 0;
}
};
#endif /* __cplusplus */
#else
#error("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#error ("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#endif
#if defined(__clang__)