From 584ebaa74a838680e6ed1fa13ac266e88c30c071 Mon Sep 17 00:00:00 2001 From: Jonathan Date: Tue, 26 Jun 2018 13:20:39 -0700 Subject: exports and imports param data in new debug tool: WatchYourStep --- debug_tools/WatchYourStep/ptxjitplus/myPtx32.ptx | 33 ++++++++++++++++++++++++ 1 file changed, 33 insertions(+) create mode 100644 debug_tools/WatchYourStep/ptxjitplus/myPtx32.ptx (limited to 'debug_tools/WatchYourStep/ptxjitplus/myPtx32.ptx') diff --git a/debug_tools/WatchYourStep/ptxjitplus/myPtx32.ptx b/debug_tools/WatchYourStep/ptxjitplus/myPtx32.ptx new file mode 100644 index 0000000..28e8e4d --- /dev/null +++ b/debug_tools/WatchYourStep/ptxjitplus/myPtx32.ptx @@ -0,0 +1,33 @@ +// +// Generated by NVIDIA NVVM Compiler +// Compiler built on Sat Mar 30 20:51:33 2013 (1364691093) +// Cuda compilation tools, release 5.5, V5.5.0 +// + +.version 3.2 +.target sm_20 +.address_size 32 + +.visible .entry _Z8myKernelPi( + .param .u32 _Z8myKernelPi_param_0 +) +{ + .reg .s32 %r<9>; + + + ld.param.u32 %r1, [_Z8myKernelPi_param_0]; + cvta.to.global.u32 %r2, %r1; + .loc 1 3 1 + mov.u32 %r3, %ntid.x; + mov.u32 %r4, %ctaid.x; + mov.u32 %r5, %tid.x; + mad.lo.s32 %r6, %r3, %r4, %r5; + .loc 1 4 1 + shl.b32 %r7, %r6, 2; + add.s32 %r8, %r2, %r7; + st.global.u32 [%r8], %r6; + .loc 1 5 2 + ret; +} + + -- cgit v1.3