|
- //
- // Generated by NVIDIA NVVM Compiler
- //
- // Compiler Build ID: CL-26907403
- // Cuda compilation tools, release 10.1, V10.1.243
- // Based on LLVM 3.4svn
- //
-
- .version 6.4
- .target sm_60
- .address_size 64
-
- // .globl Fused_BiasAdd_3555234060867906657_kernel0
- // _ZZ41Fused_BiasAdd_3555234060867906657_kernel0E14input_0_shared has been demoted
-
- .visible .entry Fused_BiasAdd_3555234060867906657_kernel0(
- .param .u64 Fused_BiasAdd_3555234060867906657_kernel0_param_0,
- .param .u64 Fused_BiasAdd_3555234060867906657_kernel0_param_1,
- .param .u64 Fused_BiasAdd_3555234060867906657_kernel0_param_2
- )
- {
- .reg .pred %p<2>;
- .reg .f32 %f<5>;
- .reg .b32 %r<5>;
- .reg .b64 %rd<12>;
- // demoted variable
- .shared .align 4 .b8 _ZZ41Fused_BiasAdd_3555234060867906657_kernel0E14input_0_shared[4];
-
- ld.param.u64 %rd3, [Fused_BiasAdd_3555234060867906657_kernel0_param_0];
- ld.param.u64 %rd4, [Fused_BiasAdd_3555234060867906657_kernel0_param_1];
- ld.param.u64 %rd5, [Fused_BiasAdd_3555234060867906657_kernel0_param_2];
- mov.u32 %r1, %tid.x;
- setp.ne.s32 %p1, %r1, 0;
- @%p1 bra BB0_2;
-
- mov.u32 %r3, %ctaid.x;
- cvta.to.global.u64 %rd6, %rd4;
- mul.wide.s32 %rd7, %r3, 4;
- add.s64 %rd8, %rd6, %rd7;
- ld.global.nc.f32 %f1, [%rd8];
- st.shared.f32 [_ZZ41Fused_BiasAdd_3555234060867906657_kernel0E14input_0_shared], %f1;
-
- BB0_2:
- cvta.to.global.u64 %rd1, %rd5;
- cvta.to.global.u64 %rd2, %rd3;
- mov.u32 %r2, %ctaid.x;
- bar.sync 0;
- mad.lo.s32 %r4, %r2, 961, %r1;
- mul.wide.s32 %rd9, %r4, 4;
- add.s64 %rd10, %rd2, %rd9;
- ld.shared.f32 %f2, [_ZZ41Fused_BiasAdd_3555234060867906657_kernel0E14input_0_shared];
- ld.global.nc.f32 %f3, [%rd10];
- add.f32 %f4, %f3, %f2;
- add.s64 %rd11, %rd1, %rd9;
- st.global.f32 [%rd11], %f4;
- bar.sync 0;
- ret;
- }
-
|