MCPcopy Create free account
hub / github.com/NVIDIA/cutlass / implicit_gemm_size

Method implicit_gemm_size

python/cutlass_cppgen/shape.py:149–167  ·  view source on GitHub ↗
(self, kind: ConvKind)

Source from the content-addressed store, hash-verified

147 return Conv2DProblemSize_(self)
148
149 def implicit_gemm_size(self, kind: ConvKind):
150 if kind == ConvKind.Fprop:
151 return GemmCoord(
152 self.N * self.P * self.Q,
153 self.K,
154 self.R * self.S * self.C // self.groups
155 )
156 elif kind == ConvKind.Dgrad:
157 return GemmCoord(
158 self.N * self.H * self.W,
159 self.C,
160 self.R * self.S * self.K
161 )
162 elif kind == ConvKind.Wgrad:
163 return GemmCoord(
164 self.K,
165 self.R * self.S * self.C,
166 self.N * self.P * self.Q
167 )
168
169 @staticmethod
170 def from_sizes(input_size, weight_size):

Callers 1

runMethod · 0.80

Calls 1

GemmCoordClass · 0.70

Tested by

no test coverage detected