AI/rocm-systems
2
0
Derivar 0
Código Questões Pedidos de integração Operações Pacotes Planeamentos Lançamentos Wiki Trabalho

Re-sync with upstream.

Ver a proveniência 
[ROCm/hip commit: fdb3e3e4a6]
Este cometimento está contido em:
Alex Voicu
2018-06-29 00:16:34 +01:00
ascendente 158e2ed16d 54d7f22c92
cometimento 1320c586e0
7 ficheiros modificados com 82 adições e 48 eliminações
Descarregar ficheiro patch Descarregar ficheiro diff Expandir todos os ficheiros Colapsar todos os ficheiros
projects/hip/bin/hipcc
+6 -1
Ver ficheiro
@@ -115,8 +115,13 @@ if ($HIP_PLATFORM eq "clang") {
$ROCM_PATH=$ENV{'ROCM_PATH'} // "/opt/rocm";
$HIPCC="$HIP_CLANG_PATH/clang++";
$HIPCXXFLAGS .= "-std=c++11 -I$HIP_PATH/include";
$HIPLDFLAGS = "--hip-link --hip-device-lib-path=$DEVICE_LIB_PATH -L$HIP_PATH/lib -lhip_hcc";
# If $HIPCC clang++ is not compiled, use clang instead
if ( ! -e $HIPCC ) {
$HIPCC="$HIP_CLANG_PATH/clang";
$HIPLDFLAGS = "--driver-mode=g++ " . $HIPLDFLAGS;
}
} elsif ($HIP_PLATFORM eq "hcc") {
$HSA_PATH=$ENV{'HSA_PATH'} // "/opt/rocm/hsa";
projects/hip/cmake/FindHIP.cmake
+8 -1
Ver ficheiro
@@ -20,6 +20,7 @@ foreach(config ${_hip_configuration_types})
mark_as_advanced(HIP_HIPCC_FLAGS_${config_upper} HIP_HCC_FLAGS_${config_upper} HIP_NVCC_FLAGS_${config_upper})
endforeach()
option(HIP_HOST_COMPILATION_CPP "Host code compilation mode" ON)
option(HIP_VERBOSE_BUILD "Print out the commands run while compiling the HIP source file. With the Makefile generator this defaults to VERBOSE variable specified on the command line, but can be forced on with this option." OFF)
mark_as_advanced(HIP_HOST_COMPILATION_CPP)
###############################################################################
@@ -471,7 +472,13 @@ macro(HIP_PREPARE_TARGET_COMMANDS _target _format _generated_files _source_files
INPUT "${custom_target_script_pregen}"
)
set(main_dep DEPENDS ${source_file})
set(verbose_output "$(VERBOSE)")
if(CMAKE_GENERATOR MATCHES "Makefiles")
set(verbose_output "$(VERBOSE)")
elseif(HIP_VERBOSE_BUILD)
set(verbose_output ON)
else()
set(verbose_output OFF)
endif()
# Create up the comment string
file(RELATIVE_PATH generated_file_relative_path "${CMAKE_BINARY_DIR}" "${generated_file}")
projects/hip/include/hip/hcc_detail/hip_complex.h
+53 -5
Ver ficheiro
@@ -24,12 +24,34 @@ THE SOFTWARE.
#define HIP_INCLUDE_HIP_HCC_DETAIL_HIP_COMPLEX_H
#include "hip/hcc_detail/hip_vector_types.h"
#include <math.h>
// TODO: Clang has a bug which allows device functions to call std functions
// when std functions are introduced into default namespace by using statement.
// math.h may be included after this bug is fixed.
#if __cplusplus
#include <cmath>
#else
#include "math.h"
#endif
#if __cplusplus
#define MAKE_COMPONENT_CONSTRUCTOR_TWO_COMPONENT(type, type1) \
__device__ __host__ type(type1 val) : x(val), y(val) {} \
__device__ __host__ type(type1 val1, type1 val2) : x(val1), y(val2) {}
#define COMPLEX_NEG_OP_OVERLOAD(type) \
__device__ __host__ static inline type operator-(const type& op) { \
type ret; \
ret.x = -op.x; \
ret.y = -op.y; \
return ret; \
}
#define COMPLEX_EQ_OP_OVERLOAD(type) \
__device__ __host__ static inline bool operator==(const type& lhs, const type& rhs) { \
return lhs.x == rhs.x && lhs.y == rhs.y; \
}
#define COMPLEX_NE_OP_OVERLOAD(type) \
__device__ __host__ static inline bool operator!=(const type& lhs, const type& rhs) { \
return !(lhs == rhs); \
}
#define COMPLEX_ADD_OP_OVERLOAD(type) \
__device__ __host__ static inline type operator+(const type& lhs, const type& rhs) { \
@@ -98,12 +120,16 @@ THE SOFTWARE.
ret.y = lhs.y * rhs; \
return ret; \
}
#define MAKE_COMPONENT_CONSTRUCTOR_TWO_COMPONENT(ComplexT, T) \
__device__ __host__ ComplexT(T val) : x(val), y(val) {} \
__device__ __host__ ComplexT(T val1, T val2) : x(val1), y(val2) {}
#endif
struct hipFloatComplex {
#ifdef __cplusplus
public:
typedef float value_type;
__device__ __host__ hipFloatComplex() : x(0.0f), y(0.0f) {}
__device__ __host__ hipFloatComplex(float x) : x(x), y(0.0f) {}
__device__ __host__ hipFloatComplex(float x, float y) : x(x), y(y) {}
@@ -123,6 +149,7 @@ struct hipFloatComplex {
struct hipDoubleComplex {
#ifdef __cplusplus
public:
typedef double value_type;
__device__ __host__ hipDoubleComplex() : x(0.0f), y(0.0f) {}
__device__ __host__ hipDoubleComplex(double x) : x(x), y(0.0f) {}
__device__ __host__ hipDoubleComplex(double x, double y) : x(x), y(y) {}
@@ -141,6 +168,9 @@ struct hipDoubleComplex {
#if __cplusplus
COMPLEX_NEG_OP_OVERLOAD(hipFloatComplex)
COMPLEX_EQ_OP_OVERLOAD(hipFloatComplex)
COMPLEX_NE_OP_OVERLOAD(hipFloatComplex)
COMPLEX_ADD_OP_OVERLOAD(hipFloatComplex)
COMPLEX_SUB_OP_OVERLOAD(hipFloatComplex)
COMPLEX_MUL_OP_OVERLOAD(hipFloatComplex)
@@ -160,6 +190,9 @@ COMPLEX_SCALAR_PRODUCT(hipFloatComplex, double)
COMPLEX_SCALAR_PRODUCT(hipFloatComplex, signed long long)
COMPLEX_SCALAR_PRODUCT(hipFloatComplex, unsigned long long)
COMPLEX_NEG_OP_OVERLOAD(hipDoubleComplex)
COMPLEX_EQ_OP_OVERLOAD(hipDoubleComplex)
COMPLEX_NE_OP_OVERLOAD(hipDoubleComplex)
COMPLEX_ADD_OP_OVERLOAD(hipDoubleComplex)
COMPLEX_SUB_OP_OVERLOAD(hipDoubleComplex)
COMPLEX_MUL_OP_OVERLOAD(hipDoubleComplex)
@@ -225,7 +258,6 @@ __device__ __host__ static inline hipFloatComplex hipCdivf(hipFloatComplex p, hi
__device__ __host__ static inline float hipCabsf(hipFloatComplex z) { return sqrtf(hipCsqabsf(z)); }
__device__ __host__ static inline double hipCreal(hipDoubleComplex z) { return z.x; }
__device__ __host__ static inline double hipCimag(hipDoubleComplex z) { return z.y; }
@@ -305,4 +337,20 @@ __device__ __host__ static inline hipDoubleComplex hipCfma(hipDoubleComplex p, h
return make_hipDoubleComplex(real, imag);
}
// Complex functions returning real numbers.
#define __DEFINE_HIP_COMPLEX_REAL_FUN(func, hipFun) \
__device__ __host__ inline float func(const hipFloatComplex& z) { return hipFun##f(z); } \
__device__ __host__ inline double func(const hipDoubleComplex& z) { return hipFun(z); }
__DEFINE_HIP_COMPLEX_REAL_FUN(abs, hipCabs)
__DEFINE_HIP_COMPLEX_REAL_FUN(real, hipCreal)
__DEFINE_HIP_COMPLEX_REAL_FUN(imag, hipCimag)
// Complex functions returning complex numbers.
#define __DEFINE_HIP_COMPLEX_FUN(func, hipFun) \
__device__ __host__ inline hipFloatComplex func(const hipFloatComplex& z) { return hipFun##f(z); } \
__device__ __host__ inline hipDoubleComplex func(const hipDoubleComplex& z) { return hipFun(z); }
__DEFINE_HIP_COMPLEX_FUN(conj, hipConj)
#endif
projects/hip/include/hip/hcc_detail/hip_fp16.h
+1 -1
Ver ficheiro
@@ -21,7 +21,7 @@ THE SOFTWARE.
*/
#pragma once
#include "hip/hcc_detail/host_defines.h"
#include <assert.h>
#if defined(__cplusplus)
#include <algorithm>
projects/hip/include/hip/hcc_detail/hip_runtime.h
+5 -5
Ver ficheiro
@@ -98,6 +98,9 @@ struct Empty_launch_parm {};
#include "grid_launch_GGL.hpp"
#endif // GENERIC_GRID_LAUNCH
#endif // HCC
#if __HCC_OR_HIP_CLANG__
extern int HIP_TRACE_API;
#ifdef __cplusplus
@@ -107,15 +110,16 @@ extern int HIP_TRACE_API;
#include <hip/hcc_detail/host_defines.h>
#include <hip/hcc_detail/math_functions.h>
#include <hip/hcc_detail/device_functions.h>
#if __HCC__
#include <hip/hcc_detail/texture_functions.h>
#include <hip/hcc_detail/surface_functions.h>
#endif // __HCC__
// TODO-HCC remove old definitions ; ~1602 hcc supports __HCC_ACCELERATOR__ define.
#if defined(__KALMAR_ACCELERATOR__) && !defined(__HCC_ACCELERATOR__)
#define __HCC_ACCELERATOR__ __KALMAR_ACCELERATOR__
#endif
// TODO-HCC add a dummy implementation of assert, need to replace with a proper kernel exit call.
#if __HIP_DEVICE_COMPILE__ == 1
#undef assert
@@ -180,10 +184,6 @@ extern int HIP_TRACE_API;
#define __HCC_C__
#endif
#endif // defined __HCC__
#if __HCC_OR_HIP_CLANG__
// TODO - hipify-clang - change to use the function call.
//#define warpSize hc::__wavesize()
static constexpr int warpSize = 64;
projects/hip/include/hip/hcc_detail/math_functions.h
+9 -9
Ver ficheiro
@@ -508,19 +508,19 @@ inline
float __fadd_rz(float x, float y) { return __ocml_add_rtz_f32(x, y); }
__device__
inline
float __fdiv_rd(float x, float y) { return __ocml_div_rtp_f32(x, y); }
float __fdiv_rd(float x, float y) { return x / y; }
__device__
inline
float __fdiv_rn(float x, float y) { return __ocml_div_rte_f32(x, y); }
float __fdiv_rn(float x, float y) { return x / y; }
__device__
inline
float __fdiv_ru(float x, float y) { return __ocml_div_rtn_f32(x, y); }
float __fdiv_ru(float x, float y) { return x / y; }
__device__
inline
float __fdiv_rz(float x, float y) { return __ocml_div_rtz_f32(x, y); }
float __fdiv_rz(float x, float y) { return x / y; }
__device__
inline
float __fdividef(float x, float y) { return __ocml_div_rte_f32(x, y); }
float __fdividef(float x, float y) { return x / y; }
__device__
inline
float __fmaf_rd(float x, float y, float z)
@@ -1028,16 +1028,16 @@ inline
double __dadd_rz(double x, double y) { return __ocml_add_rtz_f64(x, y); }
__device__
inline
double __ddiv_rd(double x, double y) { return __ocml_div_rtp_f64(x, y); }
double __ddiv_rd(double x, double y) { return x / y; }
__device__
inline
double __ddiv_rn(double x, double y) { return __ocml_div_rte_f64(x, y); }
double __ddiv_rn(double x, double y) { return x / y; }
__device__
inline
double __ddiv_ru(double x, double y) { return __ocml_div_rtn_f64(x, y); }
double __ddiv_ru(double x, double y) { return x / y; }
__device__
inline
double __ddiv_rz(double x, double y) { return __ocml_div_rtz_f64(x, y); }
double __ddiv_rz(double x, double y) { return x / y; }
__device__
inline
double __dmul_rd(double x, double y) { return __ocml_mul_rtp_f64(x, y); }
projects/hip/include/hip/hcc_detail/math_fwd.h
-26
Ver ficheiro
@@ -69,8 +69,6 @@ float __ocml_cosh_f32(float);
__device__
float __ocml_cospi_f32(float);
__device__
float __ocml_div_rtz_f32(float, float);
__device__
float __ocml_i0_f32(float);
__device__
float __ocml_i1_f32(float);
@@ -290,18 +288,6 @@ __attribute__((const))
float __ocml_mul_rtz_f32(float, float);
__device__
__attribute__((const))
float __ocml_div_rte_f32(float, float);
__device__
__attribute__((const))
float __ocml_div_rtn_f32(float, float);
__device__
__attribute__((const))
float __ocml_div_rtp_f32(float, float);
__device__
__attribute__((const))
float __ocml_div_rtz_f32(float, float);
__device__
__attribute__((const))
float __ocml_sqrt_rte_f32(float);
__device__
__attribute__((const))
@@ -598,18 +584,6 @@ __attribute__((const))
double __ocml_mul_rtz_f64(double, double);
__device__
__attribute__((const))
double __ocml_div_rte_f64(double, double);
__device__
__attribute__((const))
double __ocml_div_rtn_f64(double, double);
__device__
__attribute__((const))
double __ocml_div_rtp_f64(double, double);
__device__
__attribute__((const))
double __ocml_div_rtz_f64(double, double);
__device__
__attribute__((const))
double __ocml_sqrt_rte_f64(double);
__device__
__attribute__((const))