scan/reduce updates

Change-Id: Ic53def1aec575e4a05f8be1b492a4df574763a46
ROCm · Jun 13, 2018 · a23be48 · a23be48
1 parent 4a0848c
commit a23be48
Show file tree

Hide file tree

Showing 5 changed files with 113 additions and 166 deletions.
diff --git a/irif/inc/irif.h b/irif/inc/irif.h
@@ -309,6 +309,9 @@ extern uint __llvm_amdgcn_ds_swizzle(uint, uint) __asm("llvm.amdgcn.ds.swizzle")
 // llvm.amdgcn.mov.dpp.i32 <src> <dpp_ctrl> <row_mask> <bank_mask> <bound_ctrl>
 extern uint __llvm_amdgcn_mov_dpp_i32(uint, uint, uint, uint, bool) __asm("llvm.amdgcn.mov.dpp.i32");
 
+// llvm.amdgcn.update.dpp.i32 <old> <src> <dpp_ctrl> <row_mask> <bank_mask> <bound_ctrl>
+extern uint __llvm_amdgcn_update_dpp_i32(uint, uint, uint, uint, uint, bool) __asm("llvm.amdgcn.update.dpp.i32");
+
 // Operand bits: [0..3]=VM_CNT, [4..6]=EXP_CNT (Export), [8..11]=LGKM_CNT (LDS, GDS, Konstant, Message)
 extern void __llvm_amdgcn_s_waitcnt(int) __asm("llvm.amdgcn.s.waitcnt");
 

diff --git a/ockl/inc/ockl.h b/ockl/inc/ockl.h
@@ -133,6 +133,7 @@ DECL_CONST_OCKL_BINARY_U64(mul_hi)
 DECL_CONST_OCKL_BINARY_I32(mul24)
 DECL_CONST_OCKL_BINARY_U32(mul24)
 
+DECL_OCKL_NULLARY_U32(lane)
 DECL_OCKL_NULLARY_U32(activelane)
 
 

diff --git a/ockl/src/lane.cl b/ockl/src/lane.cl
@@ -0,0 +1,16 @@
+/*===--------------------------------------------------------------------------
+ *                   ROCm Device Libraries
+ *
+ * This file is distributed under the University of Illinois Open Source
+ * License. See LICENSE.TXT for details.
+ *===------------------------------------------------------------------------*/
+
+#include "irif.h"
+#include "ockl.h"
+
+__attribute__((always_inline)) uint
+OCKL_MANGLE_U32(lane)(void)
+{
+    return __llvm_amdgcn_mbcnt_hi(~0u, __llvm_amdgcn_mbcnt_lo(~0u, 0u));
+}
+