Compression Level is interpreted by each PMD differently. However, lower numbers give faster compression at the expense of compression ratio, while higher numbers may give better compression ratios but are likely slower. The level affects the block size, which affects performance, the bigger the block, the faster the compression is. The problem was that higher levels caused bigger blocks: size = min_block_size - 1 + level. the solution is to reverse the above: size = max_block_size + 1 - level. Fixes: 39a2c8715f8f ("compress/mlx5: add transformation operations") Cc: matan@nvidia.com Cc: stable@dpdk.org Signed-off-by: Raja Zidane <rzidane@nvidia.com> --- drivers/compress/mlx5/mlx5_compress.c | 17 ++++++++++++----- 1 file changed, 12 insertions(+), 5 deletions(-) diff --git a/drivers/compress/mlx5/mlx5_compress.c b/drivers/compress/mlx5/mlx5_compress.c index 5c2b9dc859..883e720ec1 100644 --- a/drivers/compress/mlx5/mlx5_compress.c +++ b/drivers/compress/mlx5/mlx5_compress.c @@ -316,12 +316,19 @@ mlx5_compress_xform_create(struct rte_compressdev *dev, size /= MLX5_GGA_COMP_WIN_SIZE_UNITS; xfrm->gga_ctrl1 += RTE_MIN(rte_log2_u32(size), MLX5_COMP_MAX_WIN_SIZE_CONF) << - WQE_GGA_COMP_WIN_SIZE_OFFSET; - if (xform->compress.level == RTE_COMP_LEVEL_PMD_DEFAULT) + WQE_GGA_COMP_WIN_SIZE_OFFSET; + switch (xform->compress.level) { + case RTE_COMP_LEVEL_PMD_DEFAULT: size = MLX5_GGA_COMP_LOG_BLOCK_SIZE_MAX; - else - size = priv->min_block_size - 1 + - xform->compress.level; + break; + case RTE_COMP_LEVEL_MAX: + size = priv->min_block_size; + break; + default: + size = RTE_MAX(MLX5_GGA_COMP_LOG_BLOCK_SIZE_MAX + + 1 - xform->compress.level, + priv->min_block_size); + } xfrm->gga_ctrl1 += RTE_MIN(size, MLX5_GGA_COMP_LOG_BLOCK_SIZE_MAX) << WQE_GGA_COMP_BLOCK_SIZE_OFFSET; -- 2.27.0
> Compression Level is interpreted by each PMD differently.
> However, lower numbers give faster compression
> at the expense of compression ratio, while higher numbers
> may give better compression ratios but are likely slower.
> The level affects the block size, which affects performance,
> the bigger the block, the faster the compression is.
>
> The problem was that higher levels caused bigger blocks:
> size = min_block_size - 1 + level.
>
> the solution is to reverse the above:
> size = max_block_size + 1 - level.
>
> Fixes: 39a2c8715f8f ("compress/mlx5: add transformation operations")
> Cc: matan@nvidia.com
> Cc: stable@dpdk.org
>
> Signed-off-by: Raja Zidane <rzidane@nvidia.com>
> ---
This patch is not acked, hence not applied for RC3.
@Thomas Monjalon can you pick it directly if it gets acked?
From: Raja Zidane
> Compression Level is interpreted by each PMD differently.
> However, lower numbers give faster compression at the expense of
> compression ratio, while higher numbers may give better compression ratios
> but are likely slower.
> The level affects the block size, which affects performance, the bigger the
> block, the faster the compression is.
>
> The problem was that higher levels caused bigger blocks:
> size = min_block_size - 1 + level.
>
> the solution is to reverse the above:
> size = max_block_size + 1 - level.
>
> Fixes: 39a2c8715f8f ("compress/mlx5: add transformation operations")
> Cc: matan@nvidia.com
> Cc: stable@dpdk.org
>
> Signed-off-by: Raja Zidane <rzidane@nvidia.com>
Congrats on your first patch, Raja!
Acked-by: Matan Azrad <matan@nvidia.com>
01/08/2021 08:13, Matan Azrad: > From: Raja Zidane > > Compression Level is interpreted by each PMD differently. > > However, lower numbers give faster compression at the expense of > > compression ratio, while higher numbers may give better compression ratios > > but are likely slower. > > The level affects the block size, which affects performance, the bigger the > > block, the faster the compression is. > > > > The problem was that higher levels caused bigger blocks: > > size = min_block_size - 1 + level. > > > > the solution is to reverse the above: > > size = max_block_size + 1 - level. > > > > Fixes: 39a2c8715f8f ("compress/mlx5: add transformation operations") > > Cc: matan@nvidia.com > > Cc: stable@dpdk.org > > > > Signed-off-by: Raja Zidane <rzidane@nvidia.com> > > Congrats on your first patch, Raja! The explanation above is very clear, thank you and congratulations! > Acked-by: Matan Azrad <matan@nvidia.com> Applied, thanks.