mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-11 17:00:18 +00:00
refactored implicit gemm v1r3
This commit is contained in:
@@ -143,7 +143,7 @@ void device_convolution_implicit_gemm_v1_chwn_cyxk_khwn(InDesc,
|
||||
constexpr index_t WeiBlockCopyDataPerRead_K = 4;
|
||||
|
||||
constexpr index_t OutThreadCopyDataPerWrite_N = 2;
|
||||
#elif 0
|
||||
#elif 1
|
||||
// for 3x3, 34x34, v1r3, Pascal
|
||||
// for 3x3, 28x28, v1r3, Pascal
|
||||
// for 3x3, 14x14, v1r3, Pascal
|
||||
@@ -478,9 +478,9 @@ void device_convolution_implicit_gemm_v1_chwn_cyxk_khwn(InDesc,
|
||||
GridwiseConvolutionImplicitGemm_v1r1_chwn_cyxk_khwn
|
||||
#elif 0
|
||||
GridwiseConvolutionImplicitGemm_v1r2_chwn_cyxk_khwn
|
||||
#elif 0
|
||||
GridwiseConvolutionImplicitGemm_v1r3_chwn_cyxk_khwn
|
||||
#elif 1
|
||||
GridwiseConvolutionImplicitGemm_v1r3_chwn_cyxk_khwn
|
||||
#elif 0
|
||||
GridwiseConvolutionImplicitGemm_v1r3_chwn_cyxk_khwn_lds_double_buffer
|
||||
#endif
|
||||
<GridSize,
|
||||
|
||||
@@ -85,7 +85,7 @@ int main(int argc, char* argv[])
|
||||
|
||||
constexpr index_t HPad = 0;
|
||||
constexpr index_t WPad = 0;
|
||||
#elif 0
|
||||
#elif 1
|
||||
// 3x3, 34x34
|
||||
constexpr index_t N = 128;
|
||||
constexpr index_t C = 256;
|
||||
@@ -95,7 +95,7 @@ int main(int argc, char* argv[])
|
||||
constexpr index_t Y = 3;
|
||||
constexpr index_t X = 3;
|
||||
|
||||
using ConvStrides = Sequence<2, 2>;
|
||||
using ConvStrides = Sequence<1, 1>;
|
||||
using ConvDilations = Sequence<1, 1>;
|
||||
|
||||
constexpr index_t HPad = 0;
|
||||
@@ -519,19 +519,19 @@ int main(int argc, char* argv[])
|
||||
|
||||
#if 0
|
||||
device_convolution_direct_v2_nchw_kcyx_nkhw
|
||||
(in_nchw_desc, in_nchw, wei_kcyx_desc, wei_kcyx, out_nkhw_desc, out_nkhw_device, nrepeat);
|
||||
#elif 1
|
||||
device_convolution_implicit_gemm_v1_chwn_cyxk_khwn(
|
||||
in_nchw_desc, in_nchw, wei_kcyx_desc, wei_kcyx, out_nkhw_desc, out_nkhw_device, nrepeat);
|
||||
#elif 0
|
||||
device_convolution_implicit_gemm_v1_chwn_cyxk_khwn
|
||||
device_convolution_implicit_gemm_v1_nchw_cyxk_nkhw(
|
||||
in_nchw_desc, in_nchw, wei_kcyx_desc, wei_kcyx, out_nkhw_desc, out_nkhw_device, nrepeat);
|
||||
#elif 0
|
||||
device_convolution_implicit_gemm_v1_nchw_cyxk_nkhw
|
||||
in_nchw_desc, in_nchw, wei_kcyx_desc, wei_kcyx, out_nkhw_desc, out_nkhw_device, nrepeat);
|
||||
#elif 0
|
||||
device_convolution_implicit_gemm_v2_chwn_cyxk_khwn
|
||||
device_convolution_implicit_gemm_v2_chwn_cyxk_khwn(
|
||||
in_nchw_desc, in_nchw, wei_kcyx_desc, wei_kcyx, out_nkhw_desc, out_nkhw_device, nrepeat);
|
||||
#elif 0
|
||||
device_convolution_implicit_gemm_v3_nchw_cyxk_nkhw(
|
||||
in_nchw_desc, in_nchw, wei_kcyx_desc, wei_kcyx, out_nkhw_desc, out_nkhw_device, nrepeat);
|
||||
(in_nchw_desc, in_nchw, wei_kcyx_desc, wei_kcyx, out_nkhw_desc, out_nkhw_device, nrepeat);
|
||||
#elif 0
|
||||
device_convolution_implicit_gemm_v4r1_nchw_kcyx_nkhw(in_nchw_desc,
|
||||
in_nchw,
|
||||
|
||||
Reference in New Issue
Block a user