@@ -134,6 +134,7 @@ TVM_DLL const Op& large_uint_imm();
134134 * (i.e., round(x.1) = x and round (x.5) = x+1)
135135 */
136136TVM_DLL const Op& q_multiply_shift ();
137+ TVM_DLL const Op& q_multiply_shift_per_axis ();
137138
138139/* !
139140 * \brief Returns the address of an element in the buffer (see pseudocode below).
@@ -499,6 +500,11 @@ TVM_DLL const Op& tvm_call_trace_packed_lowered();
499500 */
500501TVM_DLL const Op& tvm_storage_sync ();
501502
503+ /* !
504+ * \brief Marker where a transform should replace generated kernel initialization.
505+ */
506+ TVM_DLL const Op& tvm_kernel_replace_point ();
507+
502508/* !
503509 * \brief See pseudo code
504510 *
@@ -912,6 +918,11 @@ TVM_DLL const Op& cuda_atomic_add();
912918 */
913919TVM_DLL const Op& cuda_thread_fence ();
914920
921+ /* !
922+ * \brief tvm intrinsic for cuda warpgroup sync instruction
923+ */
924+ TVM_DLL const Op& cuda_warpgroup_sync ();
925+
915926/* !
916927 * \brief Warp-level butterfly shuffle-XOR reduction.
917928 *
@@ -952,6 +963,11 @@ TVM_DLL const Op& cuda_cta_sync();
952963 */
953964TVM_DLL const Op& cuda_grid_sync ();
954965
966+ /* !
967+ * \brief tvm intrinsic for cuda cluster-wide sync instruction
968+ */
969+ TVM_DLL const Op& cuda_cluster_sync ();
970+
955971/* !
956972 * \brief tvm intrinsic that returns ``cooperative_groups::thread_rank()``
957973 * for the enclosing CTA (linear thread index within the block).
@@ -1053,25 +1069,19 @@ TVM_DLL const Op& ptx_reduce3_max_f32();
10531069 */
10541070TVM_DLL const Op& ptx_reduce3_min_f32 ();
10551071
1056- /* !
1057- * \brief tvm intrinsic for PTX packed add instruction (sm_100a+)
1058- */
1059- TVM_DLL const Op& ptx_add_packed_f32x2 ();
1060-
1061- /* !
1062- * \brief tvm intrinsic for PTX packed subtract instruction (sm_100a+)
1063- */
1064- TVM_DLL const Op& ptx_sub_packed_f32x2 ();
1065-
1066- /* !
1067- * \brief tvm intrinsic for PTX packed multiply instruction (sm_100a+)
1068- */
1069- TVM_DLL const Op& ptx_mul_packed_f32x2 ();
1070-
1071- /* !
1072- * \brief tvm intrinsic for PTX packed FMA instruction (sm_100a+)
1073- */
1074- TVM_DLL const Op& ptx_fma_packed_f32x2 ();
1072+ TVM_DLL const Op& ptx_add_f32 ();
1073+ TVM_DLL const Op& ptx_add_f32x2 ();
1074+ TVM_DLL const Op& ptx_add_f64 ();
1075+ TVM_DLL const Op& ptx_sub_f32 ();
1076+ TVM_DLL const Op& ptx_sub_f32x2 ();
1077+ TVM_DLL const Op& ptx_sub_f64 ();
1078+ TVM_DLL const Op& ptx_mul_f32 ();
1079+ TVM_DLL const Op& ptx_mul_f32x2 ();
1080+ TVM_DLL const Op& ptx_mul_f64 ();
1081+ TVM_DLL const Op& ptx_fma_f32 ();
1082+ TVM_DLL const Op& ptx_fma_f32x2 ();
1083+ TVM_DLL const Op& ptx_fma_f64 ();
1084+ TVM_DLL const Op& ptx_max_f32 ();
10751085
10761086} // namespace builtin
10771087} // namespace tirx
0 commit comments