提交 4979f7e0 authored 作者: Arnaud Bergeron's avatar Arnaud Bergeron

Add f16 support for GpuIncSubtensor and GpuAdvancedIncSubtensor1_dev20

上级 26496654
...@@ -169,6 +169,10 @@ class GpuIncSubtensor(IncSubtensor): ...@@ -169,6 +169,10 @@ class GpuIncSubtensor(IncSubtensor):
The helper methods like do_type_checking, copy_of_x, etc. specialize The helper methods like do_type_checking, copy_of_x, etc. specialize
the c_code for this Op. the c_code for this Op.
""" """
@property
def _f16_ok(self):
return self.iadd_node.op._f16_ok
def c_headers(self): def c_headers(self):
return self.iadd_node.op.c_headers() return self.iadd_node.op.c_headers()
...@@ -325,7 +329,6 @@ class GpuIncSubtensor(IncSubtensor): ...@@ -325,7 +329,6 @@ class GpuIncSubtensor(IncSubtensor):
PyGpuArrayObject* src){ PyGpuArrayObject* src){
PyGpuArrayObject* ret = NULL; PyGpuArrayObject* ret = NULL;
""" % locals() """ % locals()
# def c_code(self, node, name, inputs, outputs, sub):
inputs = ["dst", "src"] inputs = ["dst", "src"]
outputs = ["ret"] outputs = ["ret"]
sub = {"fail": "return NULL;"} sub = {"fail": "return NULL;"}
...@@ -337,7 +340,6 @@ class GpuIncSubtensor(IncSubtensor): ...@@ -337,7 +340,6 @@ class GpuIncSubtensor(IncSubtensor):
return ret return ret
def add_to_zview(self, nodename, x, fail): def add_to_zview(self, nodename, x, fail):
# TODO
return """ return """
PyGpuArrayObject * add_result = inc_sub_iadd_%(nodename)s(zview, %(x)s); PyGpuArrayObject * add_result = inc_sub_iadd_%(nodename)s(zview, %(x)s);
...@@ -357,7 +359,7 @@ class GpuIncSubtensor(IncSubtensor): ...@@ -357,7 +359,7 @@ class GpuIncSubtensor(IncSubtensor):
elemwise_version = self.iadd_node.c_code_cache_version() elemwise_version = self.iadd_node.c_code_cache_version()
if not parent_version or not elemwise_version: if not parent_version or not elemwise_version:
return return
return parent_version + elemwise_version + (1,) return parent_version + elemwise_version + (2,)
class GpuAdvancedIncSubtensor1(HideC, tensor.AdvancedIncSubtensor1): class GpuAdvancedIncSubtensor1(HideC, tensor.AdvancedIncSubtensor1):
...@@ -391,6 +393,9 @@ class GpuAdvancedIncSubtensor1(HideC, tensor.AdvancedIncSubtensor1): ...@@ -391,6 +393,9 @@ class GpuAdvancedIncSubtensor1(HideC, tensor.AdvancedIncSubtensor1):
return gof.Apply(self, [x_, y_, ilist_], [x_.type()]) return gof.Apply(self, [x_, y_, ilist_], [x_.type()])
def getInplElemwiseAdditionKernel(self, a, b): def getInplElemwiseAdditionKernel(self, a, b):
if a.dtype == 'float16' or b.dtype == 'float16':
raise NotImplementedError('float16 is not supported by pygpu '
'elemwise')
a_arg = pygpu.tools.as_argument(a, 'a') a_arg = pygpu.tools.as_argument(a, 'a')
b_arg = pygpu.tools.as_argument(b, 'b') b_arg = pygpu.tools.as_argument(b, 'b')
args = [a_arg, b_arg] args = [a_arg, b_arg]
...@@ -452,10 +457,7 @@ class GpuAdvancedIncSubtensor1_dev20(GpuAdvancedIncSubtensor1): ...@@ -452,10 +457,7 @@ class GpuAdvancedIncSubtensor1_dev20(GpuAdvancedIncSubtensor1):
"""Implement AdvancedIncSubtensor1 on the gpu, but use function """Implement AdvancedIncSubtensor1 on the gpu, but use function
only avail on compute capability 2.0 and more recent. only avail on compute capability 2.0 and more recent.
""" """
_f16_ok = True
def __init__(self, inplace=False, set_instead_of_inc=False):
# The python implementation in the parent class is not applicable here
GpuAdvancedIncSubtensor1.__init__(self, inplace, set_instead_of_inc)
def make_node(self, x, y, ilist): def make_node(self, x, y, ilist):
"""It defer from GpuAdvancedIncSubtensor1 in that it make sure """It defer from GpuAdvancedIncSubtensor1 in that it make sure
...@@ -542,6 +544,30 @@ class GpuAdvancedIncSubtensor1_dev20(GpuAdvancedIncSubtensor1): ...@@ -542,6 +544,30 @@ class GpuAdvancedIncSubtensor1_dev20(GpuAdvancedIncSubtensor1):
itemsize_out = numpy.dtype(dtype_out).itemsize itemsize_out = numpy.dtype(dtype_out).itemsize
return """ return """
/*
* This is a version of atomicAdd that works for half-floats. It may
* read and write 2 bytes more than the size of the array if the array
* has an uneven number of elements. The actual value at that spot
* will not be modified.
*/
__device__ npy_float16 atomicAdd(npy_float16 *addr, npy_float16 val) {
npy_uint32 *base = (npy_uint32 *)((size_t)addr & ~2);
npy_uint32 old, assumed, sum, new_;
old = *base;
do {
assumed = old;
sum = __float2half_rn(
__half2float(val) +
__half2float((npy_float16)__byte_perm(old, 0,
((size_t)addr & 2) ? 0x4432 : 0x4410)));
new_ = __byte_perm(old, sum, ((size_t)addr & 2) ? 0x5410 : 0x3254);
old = atomicCAS(base, assumed, new_);
} while (assumed != old);
return (npy_float16)__byte_perm(old, 0,
((size_t)addr & 2) ? 0x4432 : 0x4410);
}
__global__ void k_vector_add_fast(int numRowsX, __global__ void k_vector_add_fast(int numRowsX,
int numColsX, int numColsX,
int stridesX0, int stridesX0,
......
Markdown 格式
0%
您添加了 0 到此讨论。请谨慎行事。
请先完成此评论的编辑!
注册 或者 后发表评论