FZGPUModules 2.0
GPU-accelerated modular compression pipelines
Loading...
Searching...
No Matches
warp_op_params.h
Go to the documentation of this file.
1#pragma once
2
19#include <cstdint>
20
21namespace fz {
22namespace fused {
23namespace warp {
24
37constexpr uint32_t kMaxWarpElemsPerLane = 4; // block_size <= 128
38
42struct Lorenzo1DParams { float inv2eb; uint32_t epl; };
43
46 float inv2eb;
47 uint32_t dx, dy;
48 uint32_t tx, ty;
49 uint32_t ntx;
50};
51
54 float inv2eb;
55 uint32_t dx, dy, dz;
56 uint32_t tx, ty, tz;
57 uint32_t ntx, nty;
58};
59
60} // namespace warp
61} // namespace fused
62} // namespace fz
Definition dag.h:24
Definition warp_op_params.h:42
cuSZp3: linear-ABS quant + 2-D separable tiled Lorenzo (tz == 1).
Definition warp_op_params.h:45
uint32_t dy
field extents (x fastest)
Definition warp_op_params.h:47
uint32_t ty
tile extents (tx*ty == block_size == 64)
Definition warp_op_params.h:48
uint32_t ntx
number of tiles along x (= ceil(dx/tx))
Definition warp_op_params.h:49
cuSZp3: linear-ABS quant + 3-D separable tiled Lorenzo (tz > 1). PROTOTYPE.
Definition warp_op_params.h:53
uint32_t dz
field extents (x fastest)
Definition warp_op_params.h:55
uint32_t tz
tile extents (tx*ty*tz == block_size == 64)
Definition warp_op_params.h:56
uint32_t nty
tiles along x, y (= ceil(dx/tx), ceil(dy/ty))
Definition warp_op_params.h:57
constexpr uint32_t kMaxWarpElemsPerLane
Definition warp_op_params.h:37