Files
rocm-systems/rocrtst/suites/performance/kernels/test_kernel.hsail
T
Chris Freehill 8161ebb915 Refactored performance test code
Commented and flattened binary search sample.

Change-Id: Ib783292207c956d16003195924a3bcfbbde5039f
2017-05-11 14:45:45 -04:00

54 rader
1.3 KiB
Plaintext
Executable File

module &m:1:0:$full:$large:$default;
extension "amd:gcn";
extension "IMAGE";
decl prog function &abort()();
prog kernel &__OpenCL_vec_assign_kernel(
kernarg_u64 %buf,
kernarg_u32 %num)
{
pragma "AMD RTI", "ARGSTART:__OpenCL_vec_assign_kernel";
pragma "AMD RTI", "version:3:1:104";
pragma "AMD RTI", "device:generic";
pragma "AMD RTI", "uniqueid:1024";
pragma "AMD RTI", "function:1:0";
pragma "AMD RTI", "memory:64bitABI";
pragma "AMD RTI", "uavid:8";
pragma "AMD RTI", "privateid:8";
pragma "AMD RTI", "ARGEND:__OpenCL_vec_assign_kernel";
@__OpenCL_vec_assign_kernel_entry:
// BB#0: // %entry
ld_kernarg_align(8)_width(all)_u64 $d0, [%buf];
ld_global_u32 $s1, [$d0];
ld_kernarg_align(4)_width(all)_u32 $s0, [%num];
cmp_ge_b1_s32 $c0, $s1, $s0;
cbr_b1 $c0, @BB0_4;
// BB#1: // %while.body.lr.ph
workitemabsid_u32 $s1, 0;
cmp_eq_b1_s32 $c0, $s1, 0;
cbr_b1 $c0, @BB0_2;
@BB0_3:
// %while.cond.backedge
ld_global_u32 $s1, [$d0];
cmp_lt_b1_s32 $c0, $s1, $s0;
cbr_b1 $c0, @BB0_3;
br @BB0_4;
@BB0_2:
// %while.cond.backedge.us
ld_global_u32 $s1, [$d0];
add_u32 $s1, $s1, 1;
st_global_u32 $s1, [$d0];
ld_global_u32 $s1, [$d0];
cmp_lt_b1_s32 $c0, $s1, $s0;
cbr_b1 $c0, @BB0_2;
@BB0_4:
// %while.end
ret;
};