Confused about identifiers in PTX code

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi there, I used Decuda to dissemble a cubin file (see code below).

// Disassembling _Z4testPi

.entry _Z4testPi

{

.lmem 0

.smem 24

.reg 2

.bar 0

mov.b32 $r0, s[0x0010]

mov.u32 $r0, g[$r0]

mul24.lo.u32.u16.u16 $r1, $r0.lo, $r0.hi

mad24.lo.u32.u16.u16.u32 $r1, $r0.hi, $r0.lo, $r1

shl.u32 $r1, $r1, 0x00000010

set.ne.s32 $p0|$o127, $r0, c1[0x0000]

mad24.lo.u32.u16.u16.u32 $r1, $r0.lo, $r0.lo, $r1

@$p0.ne return

add.half.b32 $r1, $r0, $r1

mov.half.b32 $r0, s[0x0010]

add.b32 $r1, $r1, 0x00000001

mov.end.u32 g[$r0], $r1

#.constseg 1:0x0000 const

#{

#d.u32 0x00000001 // 0000

#}

For most parts of the code I’m ok, hiwever I am confused by these ‘identifiers’ used to describe memory or pointers: s

  • , g
  • and c1
  • . What these identifiers refer to: values in memory, pointers to memory ?. Also does anybody can help me with the exact functionality of the following three lines found in the code above:

    shl.u32 $r1, $r1, 0x00000010

    …

    set.ne.s32 $p0|$o127, $r0, c1[0x0000]

    …

    @$p0.ne return

    …

    Thanks.

    PS: the original kernel in CUDA is:

    __global__ void test(int *par) {

	int y;

	int x=*par;

	int z=*par * *par;

	if (x==1) {

		y=x+z+1;

		*par = y;

	}

}

    I was just trying to see the optimizations the compiler does, so the kernel may look a bit inefficient.