Merge pull request #515 from ROCm-Developer-Tools/hipclang-add-amdgcn-funcs
Add hipclang amdgcn functions
[ROCm/hip commit: fbbe2599dd]
此提交包含在：
Maneesh Gupta
GitHub
@@ -1,387 +0,0 @@
|
||||
/*
|
||||
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
|
||||
Permission is hereby granted, free of charge, to any person obtaining a copy
|
||||
of this software and associated documentation files (the "Software"), to deal
|
||||
in the Software without restriction, including without limitation the rights
|
||||
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
||||
copies of the Software, and to permit persons to whom the Software is
|
||||
furnished to do so, subject to the following conditions:
|
||||
The above copyright notice and this permission notice shall be included in
|
||||
all copies or substantial portions of the Software.
|
||||
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
||||
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
||||
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
||||
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
||||
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
||||
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
|
||||
THE SOFTWARE.
|
||||
*/
|
||||
|
||||
#include <hip/device_functions.h>
|
||||
#include <hc.hpp>
|
||||
#include <grid_launch.h>
|
||||
#include <hc_math.hpp>
|
||||
#include "device_util.h"
|
||||
|
||||
__device__ float __double2float_rd(double x) { return (double)x; }
|
||||
__device__ float __double2float_rn(double x) { return (double)x; }
|
||||
__device__ float __double2float_ru(double x) { return (double)x; }
|
||||
__device__ float __double2float_rz(double x) { return (double)x; }
|
||||
|
||||
|
||||
__device__ int __double2hiint(double x) {
|
||||
static_assert(sizeof(double) == 2 * sizeof(int), "");
|
||||
|
||||
int tmp[2];
|
||||
__builtin_memcpy(tmp, &x, sizeof(tmp));
|
||||
|
||||
return tmp[1];
|
||||
}
|
||||
__device__ int __double2loint(double x) {
|
||||
static_assert(sizeof(double) == 2 * sizeof(int), "");
|
||||
|
||||
int tmp[2];
|
||||
__builtin_memcpy(tmp, &x, sizeof(tmp));
|
||||
|
||||
return tmp[0];
|
||||
}
|
||||
|
||||
|
||||
__device__ int __double2int_rd(double x) { return (int)x; }
|
||||
__device__ int __double2int_rn(double x) { return (int)x; }
|
||||
__device__ int __double2int_ru(double x) { return (int)x; }
|
||||
__device__ int __double2int_rz(double x) { return (int)x; }
|
||||
|
||||
__device__ long long int __double2ll_rd(double x) { return (long long int)x; }
|
||||
__device__ long long int __double2ll_rn(double x) { return (long long int)x; }
|
||||
__device__ long long int __double2ll_ru(double x) { return (long long int)x; }
|
||||
__device__ long long int __double2ll_rz(double x) { return (long long int)x; }
|
||||
|
||||
|
||||
__device__ unsigned int __double2uint_rd(double x) { return (unsigned int)x; }
|
||||
__device__ unsigned int __double2uint_rn(double x) { return (unsigned int)x; }
|
||||
__device__ unsigned int __double2uint_ru(double x) { return (unsigned int)x; }
|
||||
__device__ unsigned int __double2uint_rz(double x) { return (unsigned int)x; }
|
||||
|
||||
__device__ unsigned long long int __double2ull_rd(double x) { return (unsigned long long int)x; }
|
||||
__device__ unsigned long long int __double2ull_rn(double x) { return (unsigned long long int)x; }
|
||||
__device__ unsigned long long int __double2ull_ru(double x) { return (unsigned long long int)x; }
|
||||
__device__ unsigned long long int __double2ull_rz(double x) { return (unsigned long long int)x; }
|
||||
|
||||
__device__ long long int __double_as_longlong(double x) {
|
||||
static_assert(sizeof(long long) == sizeof(double), "");
|
||||
|
||||
long long tmp;
|
||||
__builtin_memcpy(&tmp, &x, sizeof(tmp));
|
||||
|
||||
return tmp;
|
||||
}
|
||||
|
||||
__device__ int __float2int_rd(float x) { return (int)__ocml_floor_f32(x); }
|
||||
__device__ int __float2int_rn(float x) { return (int)__ocml_rint_f32(x); }
|
||||
__device__ int __float2int_ru(float x) { return (int)__ocml_ceil_f32(x); }
|
||||
__device__ int __float2int_rz(float x) { return (int)__ocml_trunc_f32(x); }
|
||||
|
||||
__device__ long long int __float2ll_rd(float x) { return (long long int)x; }
|
||||
__device__ long long int __float2ll_rn(float x) { return (long long int)x; }
|
||||
__device__ long long int __float2ll_ru(float x) { return (long long int)x; }
|
||||
__device__ long long int __float2ll_rz(float x) { return (long long int)x; }
|
||||
|
||||
__device__ unsigned int __float2uint_rd(float x) { return (unsigned int)x; }
|
||||
__device__ unsigned int __float2uint_rn(float x) { return (unsigned int)x; }
|
||||
__device__ unsigned int __float2uint_ru(float x) { return (unsigned int)x; }
|
||||
__device__ unsigned int __float2uint_rz(float x) { return (unsigned int)x; }
|
||||
|
||||
__device__ unsigned long long int __float2ull_rd(float x) { return (unsigned long long int)x; }
|
||||
__device__ unsigned long long int __float2ull_rn(float x) { return (unsigned long long int)x; }
|
||||
__device__ unsigned long long int __float2ull_ru(float x) { return (unsigned long long int)x; }
|
||||
__device__ unsigned long long int __float2ull_rz(float x) { return (unsigned long long int)x; }
|
||||
|
||||
__device__ int __float_as_int(float x) {
|
||||
static_assert(sizeof(int) == sizeof(float), "");
|
||||
|
||||
int tmp;
|
||||
__builtin_memcpy(&tmp, &x, sizeof(tmp));
|
||||
|
||||
return tmp;
|
||||
}
|
||||
__device__ unsigned int __float_as_uint(float x) {
|
||||
static_assert(sizeof(unsigned int) == sizeof(float), "");
|
||||
|
||||
unsigned int tmp;
|
||||
__builtin_memcpy(&tmp, &x, sizeof(tmp));
|
||||
|
||||
return tmp;
|
||||
}
|
||||
__device__ double __hiloint2double(int32_t hi, int32_t lo) {
|
||||
static_assert(sizeof(double) == sizeof(uint64_t), "");
|
||||
|
||||
uint64_t tmp0 = (static_cast<uint64_t>(hi) << 32ull) | static_cast<uint32_t>(lo);
|
||||
double tmp1;
|
||||
__builtin_memcpy(&tmp1, &tmp0, sizeof(tmp0));
|
||||
|
||||
return tmp1;
|
||||
}
|
||||
__device__ double __int2double_rn(int x) { return (double)x; }
|
||||
|
||||
__device__ float __int2float_rd(int x) { return (float)x; }
|
||||
__device__ float __int2float_rn(int x) { return (float)x; }
|
||||
__device__ float __int2float_ru(int x) { return (float)x; }
|
||||
__device__ float __int2float_rz(int x) { return (float)x; }
|
||||
|
||||
__device__ float __int_as_float(int x) {
|
||||
static_assert(sizeof(float) == sizeof(int), "");
|
||||
|
||||
float tmp;
|
||||
__builtin_memcpy(&tmp, &x, sizeof(tmp));
|
||||
|
||||
return tmp;
|
||||
}
|
||||
|
||||
__device__ double __ll2double_rd(long long int x) { return (double)x; }
|
||||
__device__ double __ll2double_rn(long long int x) { return (double)x; }
|
||||
__device__ double __ll2double_ru(long long int x) { return (double)x; }
|
||||
__device__ double __ll2double_rz(long long int x) { return (double)x; }
|
||||
|
||||
__device__ float __ll2float_rd(long long int x) { return (float)x; }
|
||||
__device__ float __ll2float_rn(long long int x) { return (float)x; }
|
||||
__device__ float __ll2float_ru(long long int x) { return (float)x; }
|
||||
__device__ float __ll2float_rz(long long int x) { return (float)x; }
|
||||
|
||||
__device__ double __longlong_as_double(long long int x) {
|
||||
static_assert(sizeof(double) == sizeof(long long), "");
|
||||
|
||||
double tmp;
|
||||
__builtin_memcpy(&tmp, &x, sizeof(tmp));
|
||||
|
||||
return x;
|
||||
}
|
||||
|
||||
__device__ double __uint2double_rn(int x) { return (double)x; }
|
||||
|
||||
__device__ float __uint2float_rd(unsigned int x) { return (float)x; }
|
||||
__device__ float __uint2float_rn(unsigned int x) { return (float)x; }
|
||||
__device__ float __uint2float_ru(unsigned int x) { return (float)x; }
|
||||
__device__ float __uint2float_rz(unsigned int x) { return (float)x; }
|
||||
|
||||
__device__ float __uint_as_float(unsigned int x) {
|
||||
static_assert(sizeof(float) == sizeof(unsigned int), "");
|
||||
|
||||
float tmp;
|
||||
__builtin_memcpy(&tmp, &x, sizeof(tmp));
|
||||
|
||||
return tmp;
|
||||
}
|
||||
|
||||
__device__ double __ull2double_rd(unsigned long long int x) { return (double)x; }
|
||||
__device__ double __ull2double_rn(unsigned long long int x) { return (double)x; }
|
||||
__device__ double __ull2double_ru(unsigned long long int x) { return (double)x; }
|
||||
__device__ double __ull2double_rz(unsigned long long int x) { return (double)x; }
|
||||
|
||||
__device__ float __ull2float_rd(unsigned long long int x) { return (float)x; }
|
||||
__device__ float __ull2float_rn(unsigned long long int x) { return (float)x; }
|
||||
__device__ float __ull2float_ru(unsigned long long int x) { return (float)x; }
|
||||
__device__ float __ull2float_rz(unsigned long long int x) { return (float)x; }
|
||||
|
||||
/*
|
||||
Integer Intrinsics
|
||||
*/
|
||||
|
||||
// integer intrinsic function __poc __clz __ffs __brev
|
||||
__device__ unsigned int __popc(unsigned int input) { return hc::__popcount_u32_b32(input); }
|
||||
|
||||
__device__ unsigned int __popcll(unsigned long long int input) {
|
||||
return hc::__popcount_u32_b64(input);
|
||||
}
|
||||
|
||||
__device__ unsigned int __clz(unsigned int input) {
|
||||
#ifdef NVCC_COMPAT
|
||||
return input == 0 ? 32 : hc::__firstbit_u32_u32(input);
|
||||
#else
|
||||
return hc::__firstbit_u32_u32(input);
|
||||
#endif
|
||||
}
|
||||
|
||||
__device__ unsigned int __clzll(unsigned long long int input) {
|
||||
#ifdef NVCC_COMPAT
|
||||
return input == 0 ? 64 : hc::__firstbit_u32_u64(input);
|
||||
#else
|
||||
return hc::__firstbit_u32_u64(input);
|
||||
#endif
|
||||
}
|
||||
|
||||
__device__ unsigned int __clz(int input) {
|
||||
#ifdef NVCC_COMPAT
|
||||
return input == 0 ? 32 : hc::__firstbit_u32_s32(input);
|
||||
#else
|
||||
return hc::__firstbit_u32_s32(input);
|
||||
#endif
|
||||
}
|
||||
|
||||
__device__ unsigned int __clzll(long long int input) {
|
||||
#ifdef NVCC_COMPAT
|
||||
return input == 0 ? 64 : hc::__firstbit_u32_s64(input);
|
||||
#else
|
||||
return hc::__firstbit_u32_s64(input);
|
||||
#endif
|
||||
}
|
||||
|
||||
__device__ unsigned int __ffs(unsigned int input) {
|
||||
#ifdef NVCC_COMPAT
|
||||
return hc::__lastbit_u32_u32(input) + 1;
|
||||
#else
|
||||
return hc::__lastbit_u32_u32(input);
|
||||
#endif
|
||||
}
|
||||
|
||||
__device__ unsigned int __ffsll(unsigned long long int input) {
|
||||
#ifdef NVCC_COMPAT
|
||||
return hc::__lastbit_u32_u64(input) + 1;
|
||||
#else
|
||||
return hc::__lastbit_u32_u64(input);
|
||||
#endif
|
||||
}
|
||||
|
||||
__device__ unsigned int __ffs(int input) {
|
||||
#ifdef NVCC_COMPAT
|
||||
return hc::__lastbit_u32_s32(input) + 1;
|
||||
#else
|
||||
return hc::__lastbit_u32_s32(input);
|
||||
#endif
|
||||
}
|
||||
|
||||
__device__ unsigned int __ffsll(long long int input) {
|
||||
#ifdef NVCC_COMPAT
|
||||
return hc::__lastbit_u32_s64(input) + 1;
|
||||
#else
|
||||
return hc::__lastbit_u32_s64(input);
|
||||
#endif
|
||||
}
|
||||
|
||||
__device__ unsigned int __brev(unsigned int input) { return hc::__bitrev_b32(input); }
|
||||
|
||||
__device__ unsigned long long int __brevll(unsigned long long int input) {
|
||||
return hc::__bitrev_b64(input);
|
||||
}
|
||||
|
||||
struct ucharHolder {
|
||||
union {
|
||||
unsigned char c[4];
|
||||
unsigned int ui;
|
||||
};
|
||||
} __attribute__((aligned(4)));
|
||||
|
||||
struct uchar2Holder {
|
||||
union {
|
||||
unsigned int ui[2];
|
||||
unsigned char c[8];
|
||||
};
|
||||
} __attribute__((aligned(8)));
|
||||
|
||||
__device__ unsigned int __byte_perm(unsigned int x, unsigned int y, unsigned int s) {
|
||||
struct uchar2Holder cHoldVal;
|
||||
struct ucharHolder cHoldKey;
|
||||
struct ucharHolder cHoldOut;
|
||||
cHoldKey.ui = s;
|
||||
cHoldVal.ui[0] = x;
|
||||
cHoldVal.ui[1] = y;
|
||||
cHoldOut.c[0] = cHoldVal.c[cHoldKey.c[0]];
|
||||
cHoldOut.c[1] = cHoldVal.c[cHoldKey.c[1]];
|
||||
cHoldOut.c[2] = cHoldVal.c[cHoldKey.c[2]];
|
||||
cHoldOut.c[3] = cHoldVal.c[cHoldKey.c[3]];
|
||||
return cHoldOut.ui;
|
||||
}
|
||||
|
||||
__device__ long long __mul64hi(long long int x, long long int y) {
|
||||
ulong x0 = (ulong)x & 0xffffffffUL;
|
||||
long x1 = x >> 32;
|
||||
ulong y0 = (ulong)y & 0xffffffffUL;
|
||||
long y1 = y >> 32;
|
||||
ulong z0 = x0*y0;
|
||||
long t = x1*y0 + (z0 >> 32);
|
||||
long z1 = t & 0xffffffffL;
|
||||
long z2 = t >> 32;
|
||||
z1 = x0*y1 + z1;
|
||||
return x1*y1 + z2 + (z1 >> 32);
|
||||
}
|
||||
|
||||
__device__ unsigned long long __umul64hi(unsigned long long int x, unsigned long long int y) {
|
||||
ulong x0 = x & 0xffffffffUL;
|
||||
ulong x1 = x >> 32;
|
||||
ulong y0 = y & 0xffffffffUL;
|
||||
ulong y1 = y >> 32;
|
||||
ulong z0 = x0*y0;
|
||||
ulong t = x1*y0 + (z0 >> 32);
|
||||
ulong z1 = t & 0xffffffffUL;
|
||||
ulong z2 = t >> 32;
|
||||
z1 = x0*y1 + z1;
|
||||
return x1*y1 + z2 + (z1 >> 32);
|
||||
}
|
||||
|
||||
/*
|
||||
HIP specific device functions
|
||||
*/
|
||||
|
||||
__device__ unsigned __hip_ds_bpermute(int index, unsigned src) {
|
||||
return hc::__amdgcn_ds_bpermute(index, src);
|
||||
}
|
||||
|
||||
__device__ float __hip_ds_bpermutef(int index, float src) {
|
||||
return hc::__amdgcn_ds_bpermute(index, src);
|
||||
}
|
||||
|
||||
__device__ unsigned __hip_ds_permute(int index, unsigned src) {
|
||||
return hc::__amdgcn_ds_permute(index, src);
|
||||
}
|
||||
|
||||
__device__ float __hip_ds_permutef(int index, float src) {
|
||||
return hc::__amdgcn_ds_permute(index, src);
|
||||
}
|
||||
|
||||
__device__ unsigned __hip_ds_swizzle(unsigned int src, int pattern) {
|
||||
return hc::__amdgcn_ds_swizzle(src, pattern);
|
||||
}
|
||||
|
||||
__device__ float __hip_ds_swizzlef(float src, int pattern) {
|
||||
return hc::__amdgcn_ds_swizzle(src, pattern);
|
||||
}
|
||||
|
||||
__device__ int __hip_move_dpp(int src, int dpp_ctrl, int row_mask, int bank_mask, bool bound_ctrl) {
|
||||
return hc::__amdgcn_move_dpp(src, dpp_ctrl, row_mask, bank_mask, bound_ctrl);
|
||||
}
|
||||
|
||||
#define MASK1 0x00ff00ff
|
||||
#define MASK2 0xff00ff00
|
||||
|
||||
__device__ char4 __hip_hc_add8pk(char4 in1, char4 in2) {
|
||||
char4 out;
|
||||
unsigned one1 = in1.a & MASK1;
|
||||
unsigned one2 = in2.a & MASK1;
|
||||
out.a = (one1 + one2) & MASK1;
|
||||
one1 = in1.a & MASK2;
|
||||
one2 = in2.a & MASK2;
|
||||
out.a = out.a | ((one1 + one2) & MASK2);
|
||||
return out;
|
||||
}
|
||||
|
||||
__device__ char4 __hip_hc_sub8pk(char4 in1, char4 in2) {
|
||||
char4 out;
|
||||
unsigned one1 = in1.a & MASK1;
|
||||
unsigned one2 = in2.a & MASK1;
|
||||
out.a = (one1 - one2) & MASK1;
|
||||
one1 = in1.a & MASK2;
|
||||
one2 = in2.a & MASK2;
|
||||
out.a = out.a | ((one1 - one2) & MASK2);
|
||||
return out;
|
||||
}
|
||||
|
||||
__device__ char4 __hip_hc_mul8pk(char4 in1, char4 in2) {
|
||||
char4 out;
|
||||
unsigned one1 = in1.a & MASK1;
|
||||
unsigned one2 = in2.a & MASK1;
|
||||
out.a = (one1 * one2) & MASK1;
|
||||
one1 = in1.a & MASK2;
|
||||
one2 = in2.a & MASK2;
|
||||
out.a = out.a | ((one1 * one2) & MASK2);
|
||||
return out;
|
||||
}
|
||||
@@ -144,70 +144,7 @@ __device__ void* __hip_hc_memset(void* dst, uint8_t val, size_t size) {
|
||||
return dst;
|
||||
}
|
||||
|
||||
__device__ long long int clock64() { return (long long int)hc::__cycle_u64(); };
|
||||
__device__ clock_t clock() { return (clock_t)hc::__cycle_u64(); };
|
||||
|
||||
// abort
|
||||
__device__ void abort() { return hc::abort(); }
|
||||
|
||||
// warp vote function __all __any __ballot
|
||||
__device__ int __all(int input) { return hc::__all(input); }
|
||||
|
||||
|
||||
__device__ int __any(int input) {
|
||||
#ifdef NVCC_COMPAT
|
||||
if (hc::__any(input) != 0)
|
||||
return 1;
|
||||
else
|
||||
return 0;
|
||||
#else
|
||||
return hc::__any(input);
|
||||
#endif
|
||||
}
|
||||
|
||||
__device__ unsigned long long int __ballot(int input) { return hc::__ballot(input); }
|
||||
|
||||
// warp shuffle functions
|
||||
__device__ int __shfl(int input, int lane, int width) { return hc::__shfl(input, lane, width); }
|
||||
|
||||
__device__ int __shfl_up(int input, unsigned int lane_delta, int width) {
|
||||
return hc::__shfl_up(input, lane_delta, width);
|
||||
}
|
||||
|
||||
__device__ int __shfl_down(int input, unsigned int lane_delta, int width) {
|
||||
return hc::__shfl_down(input, lane_delta, width);
|
||||
}
|
||||
|
||||
__device__ int __shfl_xor(int input, int lane_mask, int width) {
|
||||
return hc::__shfl_xor(input, lane_mask, width);
|
||||
}
|
||||
|
||||
__device__ float __shfl(float input, int lane, int width) { return hc::__shfl(input, lane, width); }
|
||||
|
||||
__device__ float __shfl_up(float input, unsigned int lane_delta, int width) {
|
||||
return hc::__shfl_up(input, lane_delta, width);
|
||||
}
|
||||
|
||||
__device__ float __shfl_down(float input, unsigned int lane_delta, int width) {
|
||||
return hc::__shfl_down(input, lane_delta, width);
|
||||
}
|
||||
|
||||
__device__ float __shfl_xor(float input, int lane_mask, int width) {
|
||||
return hc::__shfl_xor(input, lane_mask, width);
|
||||
}
|
||||
|
||||
__host__ __device__ int min(int arg1, int arg2) {
|
||||
return (int)(hc::precise_math::fmin((float)arg1, (float)arg2));
|
||||
}
|
||||
__host__ __device__ int max(int arg1, int arg2) {
|
||||
return (int)(hc::precise_math::fmax((float)arg1, (float)arg2));
|
||||
}
|
||||
|
||||
__device__ void* __get_dynamicgroupbaseptr() {
|
||||
return hc::get_dynamic_group_segment_base_pointer();
|
||||
}
|
||||
|
||||
__host__ void* __get_dynamicgroupbaseptr() { return nullptr; }
|
||||
|
||||
|
||||
__device__ void __threadfence_system(void) { std::atomic_thread_fence(std::memory_order_seq_cst); }
|
||||
@@ -125,7 +125,6 @@ __device__ double __hip_fast_dsqrt_rd(double x);
|
||||
__device__ double __hip_fast_dsqrt_rn(double x);
|
||||
__device__ double __hip_fast_dsqrt_ru(double x);
|
||||
__device__ double __hip_fast_dsqrt_rz(double x);
|
||||
__device__ void __threadfence_system(void);
|
||||
|
||||
float __hip_host_j0f(float x);
|
||||
double __hip_host_j0(double x);
|
||||
|
||||
@@ -1,30 +0,0 @@
|
||||
target datalayout = "e-p:32:32-p1:64:64-p2:64:64-p3:32:32-p4:64:64-p5:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64"
|
||||
target triple = "amdgcn--amdhsa"
|
||||
|
||||
define i32 @__hip_hc_ir_mul24_int(i32 %a, i32 %b) #1 {
|
||||
%1 = tail call i32 asm sideeffect "v_mul_i32_i24 $0, $1, $2","=v,v,v"(i32 %a, i32 %b)
|
||||
ret i32 %1
|
||||
}
|
||||
|
||||
define i32 @__hip_hc_ir_umul24_int(i32 %a, i32 %b) #1 {
|
||||
%1 = tail call i32 asm sideeffect "v_mul_u32_u24 $0, $1, $2","=v,v,v"(i32 %a, i32 %b)
|
||||
ret i32 %1
|
||||
}
|
||||
|
||||
define i32 @__hip_hc_ir_mulhi_int(i32 %a, i32 %b) #1 {
|
||||
%1 = tail call i32 asm sideeffect "v_mul_hi_i32 $0, $1, $2","=v,v,v"(i32 %a, i32 %b)
|
||||
ret i32 %1
|
||||
}
|
||||
|
||||
define i32 @__hip_hc_ir_umulhi_int(i32 %a, i32 %b) #1 {
|
||||
%1 = tail call i32 asm sideeffect "v_mul_hi_u32 $0, $1, $2","=v,v,v"(i32 %a, i32 %b)
|
||||
ret i32 %1
|
||||
}
|
||||
|
||||
define i32 @__hip_hc_ir_usad_int(i32 %a, i32 %b, i32 %c) #1 {
|
||||
%1 = tail call i32 asm sideeffect "v_sad_u32 $0, $1, $2, $3","=v,v,v,v"(i32 %a, i32 %b, i32 %c)
|
||||
ret i32 %1
|
||||
}
|
||||
|
||||
attributes #1 = { alwaysinline nounwind }
|
||||
|
||||
@@ -1,83 +0,0 @@
|
||||
/*
|
||||
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
|
||||
|
||||
Permission is hereby granted, free of charge, to any person obtaining a copy
|
||||
of this software and associated documentation files (the "Software"), to deal
|
||||
in the Software without restriction, including without limitation the rights
|
||||
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
||||
copies of the Software, and to permit persons to whom the Software is
|
||||
furnished to do so, subject to the following conditions:
|
||||
|
||||
The above copyright notice and this permission notice shall be included in
|
||||
all copies or substantial portions of the Software.
|
||||
|
||||
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
||||
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
||||
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
||||
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
||||
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
||||
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
|
||||
THE SOFTWARE.
|
||||
*/
|
||||
|
||||
#include "hip/hcc_detail/hip_ldg.h"
|
||||
#include "hip/hcc_detail/hip_vector_types.h"
|
||||
|
||||
__device__ char __ldg(const char* ptr) { return *ptr; }
|
||||
|
||||
__device__ char2 __ldg(const char2* ptr) { return *ptr; }
|
||||
|
||||
__device__ char4 __ldg(const char4* ptr) { return *ptr; }
|
||||
|
||||
__device__ signed char __ldg(const signed char* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ unsigned char __ldg(const unsigned char* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ short __ldg(const short* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ short2 __ldg(const short2* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ short4 __ldg(const short4* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ unsigned short __ldg(const unsigned short* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ int __ldg(const int* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ int2 __ldg(const int2* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ int4 __ldg(const int4* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ unsigned int __ldg(const unsigned int* ptr) { return ptr[0]; }
|
||||
|
||||
|
||||
__device__ long __ldg(const long* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ unsigned long __ldg(const unsigned long* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ long long __ldg(const long long* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ longlong2 __ldg(const longlong2* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ unsigned long long __ldg(const unsigned long long* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ uchar2 __ldg(const uchar2* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ uchar4 __ldg(const uchar4* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ ushort2 __ldg(const ushort2* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ uint2 __ldg(const uint2* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ uint4 __ldg(const uint4* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ ulonglong2 __ldg(const ulonglong2* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ float __ldg(const float* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ float2 __ldg(const float2* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ float4 __ldg(const float4* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ double __ldg(const double* ptr) { return ptr[0]; }
|
||||
|
||||
__device__ double2 __ldg(const double2* ptr) { return ptr[0]; }
|
||||
新增問題並參考
封鎖使用者