|
- //
- // 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_Mul_Maximum_fusion_1759270382508631768_kernel0
-
- .visible .entry Fused_Mul_Maximum_fusion_1759270382508631768_kernel0(
- .param .u64 Fused_Mul_Maximum_fusion_1759270382508631768_kernel0_param_0,
- .param .u64 Fused_Mul_Maximum_fusion_1759270382508631768_kernel0_param_1,
- .param .u64 Fused_Mul_Maximum_fusion_1759270382508631768_kernel0_param_2
- )
- {
- .reg .f32 %f<17>;
- .reg .b32 %r<6>;
- .reg .b64 %rd<11>;
-
-
- ld.param.u64 %rd1, [Fused_Mul_Maximum_fusion_1759270382508631768_kernel0_param_0];
- ld.param.u64 %rd2, [Fused_Mul_Maximum_fusion_1759270382508631768_kernel0_param_1];
- ld.param.u64 %rd3, [Fused_Mul_Maximum_fusion_1759270382508631768_kernel0_param_2];
- cvta.to.global.u64 %rd4, %rd1;
- mov.u32 %r1, %ctaid.x;
- shl.b32 %r2, %r1, 10;
- mov.u32 %r3, %tid.x;
- shl.b32 %r4, %r3, 2;
- add.s32 %r5, %r4, %r2;
- mul.wide.s32 %rd5, %r5, 4;
- add.s64 %rd6, %rd4, %rd5;
- ld.global.nc.v4.f32 {%f1, %f2, %f3, %f4}, [%rd6];
- cvta.to.global.u64 %rd7, %rd2;
- cvta.to.global.u64 %rd8, %rd3;
- add.s64 %rd9, %rd8, %rd5;
- mul.f32 %f9, %f4, 0f3E4CCCCD;
- max.f32 %f10, %f9, %f4;
- mul.f32 %f11, %f3, 0f3E4CCCCD;
- max.f32 %f12, %f11, %f3;
- mul.f32 %f13, %f2, 0f3E4CCCCD;
- max.f32 %f14, %f13, %f2;
- mul.f32 %f15, %f1, 0f3E4CCCCD;
- max.f32 %f16, %f15, %f1;
- st.global.v4.f32 [%rd9], {%f16, %f14, %f12, %f10};
- add.s64 %rd10, %rd7, %rd5;
- st.global.v4.f32 [%rd10], {%f15, %f13, %f11, %f9};
- ret;
- }
-
|