spcompress GPU Virtual ISA NVIDIA
Data Movement and Conversion Instructions
spcompress.elemsize.idxsize.spfactor.num mdata, cdata, data, spdesc;
The spcompress instruction copies elements from a dense vector data to form the structured sparse compressed vector cdata.
Encoding
PTX is a virtual instruction set. It has no single, stable native
binary encoding - the compiler lowers this instruction to different native machine
code depending on the selected NVIDIA target architecture (compute capability).
This page intentionally shows no bit-diagram; see the target/version requirements
below for what governs how this instruction compiles.
Syntax Forms
One mnemonic covers many type / state-space / scope / modifier combinations - each row below is an independently valid form.
| Syntax | Data Types | State Space(s) | Modifiers | Min. Target | Description |
|---|---|---|---|---|---|
| spcompress.elemsize.idxsize.spfactor.num mdata, cdata, data, spdesc; | sm_107a | The spcompress instruction copies elements from a dense vector data to form the structured sparse compressed vector cdata. The resulting metadata is written to the vector operand mdata. (see the official PTX ISA docs for the full description) |
Operands
-
mdata
Operand -
cdata
Operand -
data
Operand -
spdesc
Operand
At a Glance
Related
Reference
NVIDIA PTX ISA
Description
The spcompress instruction copies elements from a dense vector data to
form the structured sparse compressed vector cdata. The resulting metadata is written to
the vector operand mdata.
The compression factor is specified using qualifier .spfactor. The only allowed value
for .spfactor is .sp::2:4 resulting in 2:4 compression for a dense input vector.
In this case, 2 elements per 4 group of elements from data would be selected and copied
into cdata. The output vector mdata stores the indices for the chosen 2 elements per
group of 4 elements. The criterion for selection of these elements is specified by a field
from sparsity descriptor spdesc. (see the official PTX ISA docs for the full description)
Examples
.reg .b32 mdata_<2>, cdata_<4>, data_<8>, spdesc;
mov.b32 spdesc, 0x5; // MAXABS op, .s8 datatype
spcompress.b8.b2.sp::2:4.x4 {mdata_0},
{cdata_0, cdata_1, cdata_2, cdata_3},
{data_0, data_1, data_2, data_3, data_4, data_5, data_6, data_7},
spdesc;Reproduced from NVIDIA's official PTX ISA documentation for technical accuracy.
Sources
-
Parallel Thread Execution ISA ↗
- NVIDIA Corporation, Chapter 9 - Instruction Set
Deep-linked directly to this instruction's section.