Skip to content

CUDA 9.0 deprecated non-sync shuffle instructions #136

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
ProfFan wants to merge 2 commits into
base: master
Choose a base branch
from
Open

CUDA 9.0 deprecated non-sync shuffle instructions #136

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
03c9a54
Fix CUDA deprecation warning
ProfFan Oct 17, 2018
0bd857a
CUDA 9.0 deprecated non-sync shuffle
ProfFan Oct 17, 2018
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
20 changes: 20 additions & 0 deletions API/src/util.t
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -522,6 +522,7 @@ __syncthreads = cudalib.nvvm_barrier0


local __shfl_down
local __shfl_down_sync

if opt_float == float then

Expand All @@ -537,6 +538,14 @@ if opt_float == float then
ret = terralib.asm(float,"shfl.down.b32 $0, $1, $2, $3;","=f,f,r,r", true, v, delta, c)
return ret;
end

terra __shfl_down_sync(mask : uint, v : float, delta : uint, width : int)
var ret : float;
var c : int;
c = ((warpSize-width) << 8) or 0x1F;
ret = terralib.asm(float,"shfl.sync.down.b32 $0, $1, $2, $3, $4;","=f,f,r,r,r", true, v, delta, c, mask)
return ret;
end
else
struct ULLDouble {
union {
Expand Down Expand Up @@ -606,6 +615,17 @@ else
ret.u2.y = terralib.asm(uint32,"shfl.down.b32 $0, $1, $2, $3;","=f,f,r,r", true, init.u2.y, delta, c)
return ret.d;
end

terra __shfl_down_sync(mask: uint, v : double, delta : uint, width : int)
var ret : uint2Double;
var init : uint2Double;
init.d = v
var c : int;
c = ((warpSize-width) << 8) or 0x1F;
ret.u2.x = terralib.asm(uint32,"shfl.sync.down.b32 $0, $1, $2, $3, $4;","=f,f,r,r,r", true, init.u2.x, delta, c, mask)
ret.u2.y = terralib.asm(uint32,"shfl.sync.down.b32 $0, $1, $2, $3, $4;","=f,f,r,r,r", true, init.u2.y, delta, c, mask)
return ret.d;
end
end

-- Using the "Kepler Shuffle", see http://devblogs.nvidia.com/parallelforall/faster-parallel-reductions-kepler/
Expand Down
2 changes: 1 addition & 1 deletion examples/arap_mesh_deformation/src/WarpingSolverUtil.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,7 +12,7 @@
__inline__ __device__ float warpReduce(float val) {
int offset = 32 >> 1;
while (offset > 0) {
val = val + __shfl_down(val, offset, 32);
val = val + __shfl_down_sync(0xffffffff, val, offset, 32);
offset = offset >> 1;
}
return val;
Expand Down
2 changes: 1 addition & 1 deletion examples/image_warping/src/WarpingSolverUtil.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,7 +12,7 @@
__inline__ __device__ float warpReduce(float val) {
int offset = 32 >> 1;
while (offset > 0) {
val = val + __shfl_down(val, offset, 32);
val = val + __shfl_down_sync(0xffffffff, val, offset, 32);
offset = offset >> 1;
}
return val;
Expand Down
2 changes: 1 addition & 1 deletion examples/poisson_image_editing/src/PatchSolverWarpingUtil.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -105,7 +105,7 @@ __inline__ __device__ void loadPatchToCache(volatile float* cache, float* data,
__inline__ __device__ float warpReduce(float val) {
int offset = 32 >> 1;
while (offset > 0) {
val = val + __shfl_down(val, offset, 32);
val = val + __shfl_down_sync(0xffffffff, val, offset, 32);
offset = offset >> 1;
}
return val;
Expand Down
2 changes: 1 addition & 1 deletion examples/poisson_image_editing/src/WarpingSolverUtil.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,7 +12,7 @@
__inline__ __device__ float warpReduce(float val) {
int offset = 32 >> 1;
while (offset > 0) {
val = val + __shfl_down(val, offset, 32);
val = val + __shfl_down_sync(0xffffffff, val, offset, 32);
offset = offset >> 1;
}
return val;
Expand Down
2 changes: 1 addition & 1 deletion examples/shape_from_shading/src/SFSSolverUtil.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,7 +18,7 @@ __inline__ __device__ bool IsValidPoint(float d)
__inline__ __device__ float warpReduce(float val) {
int offset = 32 >> 1;
while (offset > 0) {
val = val + __shfl_down(val, offset, 32);
val = val + __shfl_down_sync(0xffffffff, val, offset, 32);
offset = offset >> 1;
}
return val;
Expand Down
2 changes: 1 addition & 1 deletion examples/volumetric_mesh_deformation/src/WarpingSolverUtil.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,7 +12,7 @@
__inline__ __device__ float warpReduce(float val) {
int offset = 32 >> 1;
while (offset > 0) {
val = val + __shfl_down(val, offset, 32);
val = val + __shfl_down_sync(0xffffffff, val, offset, 32);
offset = offset >> 1;
}
return val;
Expand Down