BRIGFE fixes:

* brig-builtins.def: Treat HSAIL barrier builtins as
  setjmp/longjump style functions.

* brigfrontend/brig-to-generic.cc: Ensure per WI copies of
  private variables are aligned too.

* rt/workitems.c: Assume the host runtime allocates the work group
  memory.

From-SVN: r253160

6 files changed, 36 insertions, 29 deletions

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index e27d0ba..f0a5a3d 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,8 @@
+2017-09-25  Pekka Jaaskelainen <pekka@parmance.com>
+
+	* brig-builtins.def: Treat HSAIL barrier builtins as
+	setjmp/longjump style functions.
+
 2017-09-25  Richard Sandiford  <richard.sandiford@linaro.org>
 
 	* target.def (constant_alignment): New hook.
diff --git a/gcc/brig-builtins.def b/gcc/brig-builtins.def
index f525610..39f403e 100644
--- a/gcc/brig-builtins.def
+++ b/gcc/brig-builtins.def
@@ -240,7 +240,7 @@ DEF_HSAIL_BUILTIN (BUILT_IN_HSAIL_FRACT_F64, BRIG_OPCODE_FRACT,
 
 DEF_HSAIL_BUILTIN (BUILT_IN_HSAIL_BARRIER, BRIG_OPCODE_BARRIER,
 		  BRIG_TYPE_NONE, "__hsail_barrier", BT_FN_VOID_PTR,
-		  ATTR_NOTHROW_LIST)
+		  ATTR_RT_NOTHROW_LEAF_LIST)
 
 DEF_HSAIL_BUILTIN (BUILT_IN_HSAIL_INITFBAR, BRIG_OPCODE_INITFBAR,
 		  BRIG_TYPE_NONE, "__hsail_initfbar", BT_FN_VOID_UINT_PTR,
@@ -252,11 +252,11 @@ DEF_HSAIL_BUILTIN (BUILT_IN_HSAIL_JOINFBAR, BRIG_OPCODE_JOINFBAR,
 
 DEF_HSAIL_BUILTIN (BUILT_IN_HSAIL_WAITFBAR, BRIG_OPCODE_WAITFBAR,
 		  BRIG_TYPE_NONE, "__hsail_waitfbar", BT_FN_VOID_UINT_PTR,
-		  ATTR_NOTHROW_LIST)
+		  ATTR_RT_NOTHROW_LEAF_LIST)
 
 DEF_HSAIL_BUILTIN (BUILT_IN_HSAIL_ARRIVEFBAR, BRIG_OPCODE_ARRIVEFBAR,
 		  BRIG_TYPE_NONE, "__hsail_arrivefbar", BT_FN_VOID_UINT_PTR,
-		  ATTR_NOTHROW_LIST)
+		  ATTR_RT_NOTHROW_LEAF_LIST)
 
 DEF_HSAIL_BUILTIN (BUILT_IN_HSAIL_LEAVEFBAR, BRIG_OPCODE_LEAVEFBAR,
 		  BRIG_TYPE_NONE, "__hsail_leavefbar", BT_FN_VOID_UINT_PTR,
diff --git a/gcc/brig/ChangeLog b/gcc/brig/ChangeLog
index ebf31c4..69c57cb 100644
--- a/gcc/brig/ChangeLog
+++ b/gcc/brig/ChangeLog
@@ -1,3 +1,8 @@
+2017-05-13  Pekka Jääskeläinen  <pekka.jaaskelainen@parmance.com>
+
+	* brigfrontend/brig-to-generic.cc: Ensure per WI copies of
+	private variables are aligned too.
+
 2017-09-17  Thomas Schwinge  <thomas@codesourcery.com>
 
 	* Make-lang.in (GO_TEXI_FILES): Rename to...
diff --git a/gcc/brig/brigfrontend/brig-to-generic.cc b/gcc/brig/brigfrontend/brig-to-generic.cc
index 7559c05..2b1d94e 100644
--- a/gcc/brig/brigfrontend/brig-to-generic.cc
+++ b/gcc/brig/brigfrontend/brig-to-generic.cc
@@ -599,18 +599,34 @@ brig_to_generic::group_segment_size () const
   return m_next_group_offset;
 }
 
-/* Appends a new group variable to the current kernel's private segment.  */
+/* Appends a new variable to the current kernel's private segment.  */
 
 void
 brig_to_generic::append_private_variable (const std::string &name,
 					  size_t size, size_t alignment)
 {
+  /* We need to take care of two cases of alignment with private
+     variables because of the layout where the same variable for
+     each work-item is laid out in successive addresses.
+
+     1) Ensure the first work-item's variable is in an aligned
+     offset:  */
   size_t align_padding = m_next_private_offset % alignment == 0 ?
     0 : (alignment - m_next_private_offset % alignment);
+
+  /* 2) Each successive per-work-item copy should be aligned.
+     If the variable has wider alignment than size then we need
+     to add extra padding to ensure it.  The padding must be
+     included in the size to allow per-work-item offset computation
+     to find their own aligned copy.  */
+
+  size_t per_var_padding = size % alignment == 0 ?
+    0 : (alignment - size % alignment);
+  m_private_data_sizes[name] = size + per_var_padding;
+
   m_next_private_offset += align_padding;
   m_private_offsets[name] = m_next_private_offset;
-  m_next_private_offset += size;
-  m_private_data_sizes[name] = size + align_padding;
+  m_next_private_offset += size + per_var_padding;
 }
 
 size_t
diff --git a/libhsail-rt/ChangeLog b/libhsail-rt/ChangeLog
index 31ffff6..bf86278 100644
--- a/libhsail-rt/ChangeLog
+++ b/libhsail-rt/ChangeLog
@@ -1,3 +1,7 @@
+2017-09-25  Pekka Jääskeläinen  <pekka.jaaskelainen@parmance.com>
+
+	* rt/workitems.c: Assume the host runtime allocates the work group
+	memory.
 2017-05-03  Pekka Jääskeläinen  <pekka.jaaskelainen@parmance.com>
 
 	* rt/workitems.c: Removed a leftover comment.
diff --git a/libhsail-rt/rt/workitems.c b/libhsail-rt/rt/workitems.c
index e2c2373..ed1185a 100644
--- a/libhsail-rt/rt/workitems.c
+++ b/libhsail-rt/rt/workitems.c
@@ -318,14 +318,6 @@ phsa_spawn_work_items (PHSAKernelLaunchData *context, void *group_base_ptr)
   hsa_kernel_dispatch_packet_t *dp = context->dp;
   size_t x, y, z;
 
-  /* TO DO: host-side memory management of group and private segment
-     memory.  Agents in general are less likely to support efficient dynamic mem
-     allocation.  */
-  if (dp->group_segment_size > 0
-      && posix_memalign (&group_base_ptr, PRIVATE_SEGMENT_ALIGN,
-			 dp->group_segment_size) != 0)
-    phsa_fatal_error (3);
-
   context->group_segment_start_addr = (size_t) group_base_ptr;
 
   /* HSA seems to allow the WG size to be larger than the grid size.  We need to
@@ -371,9 +363,6 @@ phsa_spawn_work_items (PHSAKernelLaunchData *context, void *group_base_ptr)
 
   phsa_execute_wi_gang (context, group_base_ptr, sat_wg_size_x, sat_wg_size_y,
 			sat_wg_size_z);
-
-  if (dp->group_segment_size > 0)
-    free (group_base_ptr);
 }
 #endif
 
@@ -390,14 +379,6 @@ phsa_execute_work_groups (PHSAKernelLaunchData *context, void *group_base_ptr)
   hsa_kernel_dispatch_packet_t *dp = context->dp;
   size_t x, y, z, wg_x, wg_y, wg_z;
 
-  /* TODO: host-side memory management of group and private segment
-     memory.  Agents in general are less likely to support efficient dynamic mem
-     allocation.  */
-  if (dp->group_segment_size > 0
-      && posix_memalign (&group_base_ptr, GROUP_SEGMENT_ALIGN,
-			 dp->group_segment_size) != 0)
-    phsa_fatal_error (3);
-
   context->group_segment_start_addr = (size_t) group_base_ptr;
 
   /* HSA seems to allow the WG size to be larger than the grid size.  We need
@@ -509,10 +490,6 @@ phsa_execute_work_groups (PHSAKernelLaunchData *context, void *group_base_ptr)
   printf ("### %lu WIs executed in %lu s (%lu WIs / s)\n", wi_total,
 	  (uint64_t) spent_time_sec, (uint64_t) wis_per_sec);
 #endif
-
-  if (dp->group_segment_size > 0)
-    free (group_base_ptr);
-
   free (private_base_ptr);
   private_base_ptr = NULL;
 }