From 2ff691f3f2184f94b452fad7bf9e9abc868c5261 Mon Sep 17 00:00:00 2001 From: Clement Lin Date: Wed, 12 Mar 2025 14:44:41 +0800 Subject: [PATCH] Utilize vectorized memory access 1998.24 GB/s -> 2673 GB/s --- example/ck_tile/99_toy_example/01_add/add.cpp | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/example/ck_tile/99_toy_example/01_add/add.cpp b/example/ck_tile/99_toy_example/01_add/add.cpp index 022df309b5..cbe019ba5e 100644 --- a/example/ck_tile/99_toy_example/01_add/add.cpp +++ b/example/ck_tile/99_toy_example/01_add/add.cpp @@ -48,9 +48,9 @@ bool run(const ck_tile::ArgParser& arg_parser) // 27xx -> 32xx, 1D block to 2D block using BlockWarps = ck_tile::sequence<1, 4>; - using BlockTile = ck_tile::sequence<1, 1024>; - using WarpTile = ck_tile::sequence<1, 256>; - using Vector = ck_tile::sequence<1, 4>; + using BlockTile = ck_tile::sequence<1, 2048>; + using WarpTile = ck_tile::sequence<1, 512>; + using Vector = ck_tile::sequence<1, 8>; constexpr ck_tile::index_t kBlockSize = 256; constexpr ck_tile::index_t kBlockPerCu = 1;