Added flag to disable src zero detection flag for eltwise unary/sfpu …

…ops in blackhole
tenstorrent · Feb 24, 2025 · a7c822c · a7c822c
1 parent 2456a22
commit a7c822c
Show file tree

Hide file tree

Showing 2 changed files with 12 additions and 5 deletions.
diff --git a/tt_metal/hw/ckernels/blackhole/metal/llk_api/llk_unpack_A_api.h b/tt_metal/hw/ckernels/blackhole/metal/llk_api/llk_unpack_A_api.h
@@ -10,26 +10,33 @@
  * LLK UNPACK A
  *************************************************************************/
 
-template <bool is_fp32_dest_acc_en = false, StochRndType stoch_rnd_mode = StochRndType::None>
+template <
+    bool is_fp32_dest_acc_en = false,
+    StochRndType stoch_rnd_mode = StochRndType::None,
+    bool disable_src_zero_flag = false>
 inline void llk_unpack_A_hw_configure(
     const llk_unpack_A_params_t* unpack_A_params, const int within_face_16x16_transpose = 0) {
     const uint32_t unpA_operand_id = get_operand_id(unpack_A_params->unpA_operand);
     const uint32_t unpA_num_faces = get_operand_num_faces(unpA_operand_id);
     const uint32_t unpA_face_r_dim = get_operand_face_r_dim(unpA_operand_id);
 
-    _llk_unpack_A_hw_configure_<is_fp32_dest_acc_en, stoch_rnd_mode>(
+    _llk_unpack_A_hw_configure_<is_fp32_dest_acc_en, stoch_rnd_mode, disable_src_zero_flag>(
         unpack_src_format[unpA_operand_id],
         unpack_dst_format[unpA_operand_id],
         unpA_face_r_dim,
         within_face_16x16_transpose,
         unpA_num_faces);
 }
 
-template <bool is_fp32_dest_acc_en = false, StochRndType stoch_rnd_mode = StochRndType::None>
+template <
+    bool is_fp32_dest_acc_en = false,
+    StochRndType stoch_rnd_mode = StochRndType::None,
+    bool disable_src_zero_flag = false>
 inline void llk_unpack_A_hw_configure_disaggregated(
     const std::uint32_t unpA_operand, const int within_face_16x16_transpose = 0) {
     const llk_unpack_A_params_t unpack_A_params = {.unpA_operand = unpA_operand};
-    llk_unpack_A_hw_configure<is_fp32_dest_acc_en, stoch_rnd_mode>(&unpack_A_params, within_face_16x16_transpose);
+    llk_unpack_A_hw_configure<is_fp32_dest_acc_en, stoch_rnd_mode, disable_src_zero_flag>(
+        &unpack_A_params, within_face_16x16_transpose);
 }
 
 template <

diff --git a/tt_metal/third_party/tt_llk b/tt_metal/third_party/tt_llk
+4 −1		tt_llk_blackhole/common/inc/cunpack_common.h
+2 −2		tt_llk_blackhole/llk_lib/llk_unpack_A.h