fix perchanel fp16 load

intel · Aug 30, 2024 · af22c33 · af22c33
1 parent af20dba
commit af22c33
Show file tree

Hide file tree

Showing 3 changed files with 9 additions and 47 deletions.
diff --git a/include/common/core/memory.hpp b/include/common/core/memory.hpp
@@ -476,24 +476,9 @@ __XETLA_API xetla_vector<T, N> xetla_load_global(
         Y);
     return ret.xetla_format<T>();
   } else if constexpr (BlockWidth * sizeof(T) < sizeof(uint32_t)) {
-    constexpr auto scale_factor = sizeof(uint32_t) / sizeof(T);
-    xetla_vector<uint32_t, N> ret = xetla_load_global<
-        uint32_t,
-        BlockWidth,
-        BlockHeight,
-        NBlocks,
-        Transposed,
-        Transformed,
-        L1H,
-        L2H>(
-        reinterpret_cast<const uint32_t*>(Ptr),
-        SurfaceWidth,
-        SurfaceHeight,
-        SurfacePitch,
-        X / scale_factor,
-        Y);
-    return ret.xetla_format<T>().xetla_select<N, scale_factor>(
-        X % scale_factor);
+    xetla_vector<uint32_t, BlockHeight> byte_offsets =
+        xetla_vector_gen<uint32_t, BlockHeight>(0, SurfacePitch);
+    return xetla_load_global<T, N, BlockWidth, L1H, L2H>(Ptr, byte_offsets);
   } else {
     return __ESIMD_ENS::lsc_load_2d<
         T,

diff --git a/include/subgroup/tile/impl/load_xe.hpp b/include/subgroup/tile/impl/load_xe.hpp
@@ -236,10 +236,10 @@ tile_load(tile_t& tile, payload_t& payload) {
               reg_tmp
                   .xetla_format<
                       native_type_t<load_dtype>,
-                      block_size_x / scale_factor,
+                      ld_blk_width / scale_factor,
                       ld_blk_height>()
                   .xetla_select<
-                      block_size_x / scale_factor,
+                      ld_blk_width / scale_factor,
                       1,
                       ld_blk_size_y,
                       1>(0, 0);
@@ -297,9 +297,9 @@ tile_load(tile_t& tile, payload_t& payload) {
       //   xetla_tdescriptor tdesc = payload_row.row(j);
       auto reg_blk = tile.reg.xetla_select<remained_block_elems * arr_len, 1>(
           processed_elems + j * remained_block_elems);
-      // constexpr uint32_t ld_blk_height = (reg_transpose && trans)
-      //     ? detail::getNextPowerOf2<remained_ld_blk_size_y>()
-      //     : remained_ld_blk_size_y;
+      constexpr uint32_t ld_blk_height = (reg_transpose && trans)
+          ? detail::getNextPowerOf2<remained_ld_blk_size_y>()
+          : remained_ld_blk_size_y;
       constexpr uint32_t tmp_size = ld_blk_height * block_size_x * arr_len;
       xetla_vector<dtype, tmp_size> reg_tmp;
 #pragma unroll
@@ -311,7 +311,7 @@ tile_load(tile_t& tile, payload_t& payload) {
         reg_tmp.xetla_format<native_type_t<load_dtype>>() = xetla_load_global<
             native_type_t<load_dtype>,
             block_size_x / scale_factor,
-            ld_blk_height,
+            remained_ld_blk_size_y,
             arr_len,
             trans,
             mem_transform,
@@ -325,15 +325,6 @@ tile_load(tile_t& tile, payload_t& payload) {
             payload.offset_x + offset_x / scale_factor,
             payload.offset_y + num_block_y * block_size_y +
                 ii * remained_ld_blk_size_y);
-        //  xetla_tload_global<
-        // load_dtype,
-        // (ld_blk_height * block_size_x * arr_len / scale_factor),
-        // L1,
-        // L2,
-        // trans,
-        // mem_transform,
-        // arch_tag>(tdesc);
-
         if constexpr (reg_transpose && trans) {
           reg_blk.xetla_select<load_elems, 1>(ii * load_elems)
               .xetla_format<native_type_t<load_dtype>>() =

diff --git a/include/subgroup/tile/impl/payload_xe.hpp b/include/subgroup/tile/impl/payload_xe.hpp
@@ -1841,20 +1841,6 @@ struct prefetch_payload_t<
     return channel >= 32 ? 32 : channel >= 16 ? 16 : channel >= 8 ? 8 : 1;
   }
 
-  static constexpr uint32_t num_channel = select_channel(
-      std::min(mem_transpose ? block_size_x : block_size_y, max_channel));
-
-  static constexpr uint32_t max_channel =
-      max_prefetch_vec_len / (vector_size * sizeof(prefetch_dtype));
-
-  static constexpr uint32_t select_channel(const uint32_t channel) {
-    return (channel >= load_store_attr::max_channel_num)
-        ? load_store_attr::max_channel_num
-        : channel >= 16 ? 16
-        : channel >= 8  ? 8
-                        : 1;
-  }
-
   static constexpr uint32_t num_channel = select_channel(
       std::min(mem_transpose ? block_size_x : block_size_y, max_channel));