AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

SWDEV-283266 - __HIPCC_RTC__ macro added to HIP headers

Browse Source 
For hipRTC on Windows, add macro __HIPCC_RTC__ to allow
online compilation of with device functions excluding standard
C/C++ headers, system headers, and host HIP APIs.

Change-Id: I1d91f042baf1359856ec83ab7030dc58785e0334
This commit is contained in:
Aaron En Ye Shi
2021-05-05 21:26:21 +00:00
parent 73964037c1
commit 811bc363c8
12 changed files with 368 additions and 310 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/hip/amd_detail/hip_complex.h
+42 -37
View File
@@ -25,6 +25,10 @@ THE SOFTWARE.
#include "hip/amd_detail/hip_vector_types.h"
#if defined(__HIPCC_RTC__)
#define __HOST_DEVICE__ __device__
#else
#define __HOST_DEVICE__ __host__ __device__
// 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.
@@ -33,10 +37,11 @@ THE SOFTWARE.
#else
#include "math.h"
#endif
#endif // !defined(__HIPCC_RTC__)
#if __cplusplus
#define COMPLEX_NEG_OP_OVERLOAD(type) \
__device__ __host__ static inline type operator-(const type& op) { \
__HOST_DEVICE__ static inline type operator-(const type& op) { \
type ret; \
ret.x = -op.x; \
ret.y = -op.y; \
@@ -44,17 +49,17 @@ THE SOFTWARE.
}
#define COMPLEX_EQ_OP_OVERLOAD(type) \
__device__ __host__ static inline bool operator==(const type& lhs, const type& rhs) { \
__HOST_DEVICE__ 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) { \
__HOST_DEVICE__ 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) { \
__HOST_DEVICE__ static inline type operator+(const type& lhs, const type& rhs) { \
type ret; \
ret.x = lhs.x + rhs.x; \
ret.y = lhs.y + rhs.y; \
@@ -62,7 +67,7 @@ THE SOFTWARE.
}
#define COMPLEX_SUB_OP_OVERLOAD(type) \
__device__ __host__ static inline type operator-(const type& lhs, const type& rhs) { \
__HOST_DEVICE__ static inline type operator-(const type& lhs, const type& rhs) { \
type ret; \
ret.x = lhs.x - rhs.x; \
ret.y = lhs.y - rhs.y; \
@@ -70,7 +75,7 @@ THE SOFTWARE.
}
#define COMPLEX_MUL_OP_OVERLOAD(type) \
__device__ __host__ static inline type operator*(const type& lhs, const type& rhs) { \
__HOST_DEVICE__ static inline type operator*(const type& lhs, const type& rhs) { \
type ret; \
ret.x = lhs.x * rhs.x - lhs.y * rhs.y; \
ret.y = lhs.x * rhs.y + lhs.y * rhs.x; \
@@ -78,7 +83,7 @@ THE SOFTWARE.
}
#define COMPLEX_DIV_OP_OVERLOAD(type) \
__device__ __host__ static inline type operator/(const type& lhs, const type& rhs) { \
__HOST_DEVICE__ static inline type operator/(const type& lhs, const type& rhs) { \
type ret; \
ret.x = (lhs.x * rhs.x + lhs.y * rhs.y); \
ret.y = (rhs.x * lhs.y - lhs.x * rhs.y); \
@@ -88,33 +93,33 @@ THE SOFTWARE.
}
#define COMPLEX_ADD_PREOP_OVERLOAD(type) \
__device__ __host__ static inline type& operator+=(type& lhs, const type& rhs) { \
__HOST_DEVICE__ static inline type& operator+=(type& lhs, const type& rhs) { \
lhs.x += rhs.x; \
lhs.y += rhs.y; \
return lhs; \
}
#define COMPLEX_SUB_PREOP_OVERLOAD(type) \
__device__ __host__ static inline type& operator-=(type& lhs, const type& rhs) { \
__HOST_DEVICE__ static inline type& operator-=(type& lhs, const type& rhs) { \
lhs.x -= rhs.x; \
lhs.y -= rhs.y; \
return lhs; \
}
#define COMPLEX_MUL_PREOP_OVERLOAD(type) \
__device__ __host__ static inline type& operator*=(type& lhs, const type& rhs) { \
__HOST_DEVICE__ static inline type& operator*=(type& lhs, const type& rhs) { \
lhs = lhs * rhs; \
return lhs; \
}
#define COMPLEX_DIV_PREOP_OVERLOAD(type) \
__device__ __host__ static inline type& operator/=(type& lhs, const type& rhs) { \
__HOST_DEVICE__ static inline type& operator/=(type& lhs, const type& rhs) { \
lhs = lhs / rhs; \
return lhs; \
}
#define COMPLEX_SCALAR_PRODUCT(type, type1) \
__device__ __host__ static inline type operator*(const type& lhs, type1 rhs) { \
__HOST_DEVICE__ static inline type operator*(const type& lhs, type1 rhs) { \
type ret; \
ret.x = lhs.x * rhs; \
ret.y = lhs.y * rhs; \
@@ -125,41 +130,41 @@ THE SOFTWARE.
typedef float2 hipFloatComplex;
__device__ __host__ static inline float hipCrealf(hipFloatComplex z) { return z.x; }
__HOST_DEVICE__ static inline float hipCrealf(hipFloatComplex z) { return z.x; }
__device__ __host__ static inline float hipCimagf(hipFloatComplex z) { return z.y; }
__HOST_DEVICE__ static inline float hipCimagf(hipFloatComplex z) { return z.y; }
__device__ __host__ static inline hipFloatComplex make_hipFloatComplex(float a, float b) {
__HOST_DEVICE__ static inline hipFloatComplex make_hipFloatComplex(float a, float b) {
hipFloatComplex z;
z.x = a;
z.y = b;
return z;
}
__device__ __host__ static inline hipFloatComplex hipConjf(hipFloatComplex z) {
__HOST_DEVICE__ static inline hipFloatComplex hipConjf(hipFloatComplex z) {
hipFloatComplex ret;
ret.x = z.x;
ret.y = -z.y;
return ret;
}
__device__ __host__ static inline float hipCsqabsf(hipFloatComplex z) {
__HOST_DEVICE__ static inline float hipCsqabsf(hipFloatComplex z) {
return z.x * z.x + z.y * z.y;
}
__device__ __host__ static inline hipFloatComplex hipCaddf(hipFloatComplex p, hipFloatComplex q) {
__HOST_DEVICE__ static inline hipFloatComplex hipCaddf(hipFloatComplex p, hipFloatComplex q) {
return make_hipFloatComplex(p.x + q.x, p.y + q.y);
}
__device__ __host__ static inline hipFloatComplex hipCsubf(hipFloatComplex p, hipFloatComplex q) {
__HOST_DEVICE__ static inline hipFloatComplex hipCsubf(hipFloatComplex p, hipFloatComplex q) {
return make_hipFloatComplex(p.x - q.x, p.y - q.y);
}
__device__ __host__ static inline hipFloatComplex hipCmulf(hipFloatComplex p, hipFloatComplex q) {
__HOST_DEVICE__ static inline hipFloatComplex hipCmulf(hipFloatComplex p, hipFloatComplex q) {
return make_hipFloatComplex(p.x * q.x - p.y * q.y, p.y * q.x + p.x * q.y);
}
__device__ __host__ static inline hipFloatComplex hipCdivf(hipFloatComplex p, hipFloatComplex q) {
__HOST_DEVICE__ static inline hipFloatComplex hipCdivf(hipFloatComplex p, hipFloatComplex q) {
float sqabs = hipCsqabsf(q);
hipFloatComplex ret;
ret.x = (p.x * q.x + p.y * q.y) / sqabs;
@@ -167,46 +172,46 @@ __device__ __host__ static inline hipFloatComplex hipCdivf(hipFloatComplex p, hi
return ret;
}
__device__ __host__ static inline float hipCabsf(hipFloatComplex z) { return sqrtf(hipCsqabsf(z)); }
__HOST_DEVICE__ static inline float hipCabsf(hipFloatComplex z) { return sqrtf(hipCsqabsf(z)); }
typedef double2 hipDoubleComplex;
__device__ __host__ static inline double hipCreal(hipDoubleComplex z) { return z.x; }
__HOST_DEVICE__ static inline double hipCreal(hipDoubleComplex z) { return z.x; }
__device__ __host__ static inline double hipCimag(hipDoubleComplex z) { return z.y; }
__HOST_DEVICE__ static inline double hipCimag(hipDoubleComplex z) { return z.y; }
__device__ __host__ static inline hipDoubleComplex make_hipDoubleComplex(double a, double b) {
__HOST_DEVICE__ static inline hipDoubleComplex make_hipDoubleComplex(double a, double b) {
hipDoubleComplex z;
z.x = a;
z.y = b;
return z;
}
__device__ __host__ static inline hipDoubleComplex hipConj(hipDoubleComplex z) {
__HOST_DEVICE__ static inline hipDoubleComplex hipConj(hipDoubleComplex z) {
hipDoubleComplex ret;
ret.x = z.x;
ret.y = -z.y;
return ret;
}
__device__ __host__ static inline double hipCsqabs(hipDoubleComplex z) {
__HOST_DEVICE__ static inline double hipCsqabs(hipDoubleComplex z) {
return z.x * z.x + z.y * z.y;
}
__device__ __host__ static inline hipDoubleComplex hipCadd(hipDoubleComplex p, hipDoubleComplex q) {
__HOST_DEVICE__ static inline hipDoubleComplex hipCadd(hipDoubleComplex p, hipDoubleComplex q) {
return make_hipDoubleComplex(p.x + q.x, p.y + q.y);
}
__device__ __host__ static inline hipDoubleComplex hipCsub(hipDoubleComplex p, hipDoubleComplex q) {
__HOST_DEVICE__ static inline hipDoubleComplex hipCsub(hipDoubleComplex p, hipDoubleComplex q) {
return make_hipDoubleComplex(p.x - q.x, p.y - q.y);
}
__device__ __host__ static inline hipDoubleComplex hipCmul(hipDoubleComplex p, hipDoubleComplex q) {
__HOST_DEVICE__ static inline hipDoubleComplex hipCmul(hipDoubleComplex p, hipDoubleComplex q) {
return make_hipDoubleComplex(p.x * q.x - p.y * q.y, p.y * q.x + p.x * q.y);
}
__device__ __host__ static inline hipDoubleComplex hipCdiv(hipDoubleComplex p, hipDoubleComplex q) {
__HOST_DEVICE__ static inline hipDoubleComplex hipCdiv(hipDoubleComplex p, hipDoubleComplex q) {
double sqabs = hipCsqabs(q);
hipDoubleComplex ret;
ret.x = (p.x * q.x + p.y * q.y) / sqabs;
@@ -214,7 +219,7 @@ __device__ __host__ static inline hipDoubleComplex hipCdiv(hipDoubleComplex p, h
return ret;
}
__device__ __host__ static inline double hipCabs(hipDoubleComplex z) { return sqrt(hipCsqabs(z)); }
__HOST_DEVICE__ static inline double hipCabs(hipDoubleComplex z) { return sqrt(hipCsqabs(z)); }
#if __cplusplus
@@ -268,19 +273,19 @@ COMPLEX_SCALAR_PRODUCT(hipDoubleComplex, unsigned long long)
typedef hipFloatComplex hipComplex;
__device__ __host__ static inline hipComplex make_hipComplex(float x, float y) {
__HOST_DEVICE__ static inline hipComplex make_hipComplex(float x, float y) {
return make_hipFloatComplex(x, y);
}
__device__ __host__ static inline hipFloatComplex hipComplexDoubleToFloat(hipDoubleComplex z) {
__HOST_DEVICE__ static inline hipFloatComplex hipComplexDoubleToFloat(hipDoubleComplex z) {
return make_hipFloatComplex((float)z.x, (float)z.y);
}
__device__ __host__ static inline hipDoubleComplex hipComplexFloatToDouble(hipFloatComplex z) {
__HOST_DEVICE__ static inline hipDoubleComplex hipComplexFloatToDouble(hipFloatComplex z) {
return make_hipDoubleComplex((double)z.x, (double)z.y);
}
__device__ __host__ static inline hipComplex hipCfmaf(hipComplex p, hipComplex q, hipComplex r) {
__HOST_DEVICE__ static inline hipComplex hipCfmaf(hipComplex p, hipComplex q, hipComplex r) {
float real = (p.x * q.x) + r.x;
float imag = (q.x * p.y) + r.y;
@@ -290,7 +295,7 @@ __device__ __host__ static inline hipComplex hipCfmaf(hipComplex p, hipComplex q
return make_hipComplex(real, imag);
}
__device__ __host__ static inline hipDoubleComplex hipCfma(hipDoubleComplex p, hipDoubleComplex q,
__HOST_DEVICE__ static inline hipDoubleComplex hipCfma(hipDoubleComplex p, hipDoubleComplex q,
hipDoubleComplex r) {
double real = (p.x * q.x) + r.x;
double imag = (q.x * p.y) + r.y;