This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [patch] various OpenACC reduction enhancements - ME and nvptx changes


On Tue, 4 Dec 2018 16:55:04 +0100
Tom de Vries <tdevries@suse.de> wrote:

> On 04-12-18 13:29, Jakub Jelinek wrote:
> > On Fri, Jun 29, 2018 at 11:19:53AM -0700, Cesar Philippidis wrote:  
> >> The attached patch includes the nvptx and GCC ME reductions
> >> enhancements.
> >>
> >> Is this patch OK for trunk? It bootstrapped / regression tested
> >> cleanly for x86_64 with nvptx offloading.  
> > This is all OpenACC specific code not really shareable with OpenMP,
> > if Thomas (for middle-end) and Tom (for NVPTX backend) are ok with
> > it, it is ok for trunk.
> >   
> 
> Formatting needs to be fixed:
> ...
> There should be exactly one space between function name and
> parenthesis. 160:+      unsigned old_shift = DIM_SIZE(VECTOR);
> ...
> 
> Also, the updated patch does not address my comment about
> probabilities here
> ( https://gcc.gnu.org/ml/gcc-patches/2018-10/msg00325.html ): ...
> > +  /* Create the loop.  */
> > +  post_edge->flags ^= EDGE_TRUE_VALUE | EDGE_FALLTHRU;  
> 
> Edges need probabilities, as in nvptx_lockless_update,
> nvptx_lockfull_update and nvptx_goacc_reduction_init.
> ...

Something like the attached?

Tested alongside other revised patches in the series:

https://gcc.gnu.org/ml/gcc-patches/2018-12/msg00930.html
https://gcc.gnu.org/ml/gcc-patches/2018-12/msg00931.html

(except the lines adding edge probabilities, which I've
smoke-tested but haven't yet gone through a full test cycle).

Thanks,

Julian

ChangeLog

        gcc/
        * config/nvptx/nvptx.c (nvptx_propagate_unified): New.
        (nvptx_split_blocks): Call it for cond_uni insn.
        (nvptx_expand_cond_uni): New.
        (enum nvptx_builtins): Add NVPTX_BUILTIN_COND_UNI.
        (nvptx_init_builtins): Initialize it.
        (nvptx_expand_builtin):
        (nvptx_generate_vector_shuffle): Change integral SHIFT operand to
        tree BITS operand.
        (nvptx_vector_reduction): New.
        (nvptx_adjust_reduction_type): New.
        (nvptx_goacc_reduction_setup): Use it to adjust the type of ref_to_res.
        (nvptx_goacc_reduction_init): Don't update LHS if it doesn't exist.
        (nvptx_goacc_reduction_fini): Call nvptx_vector_reduction for vector.
        Use it to adjust the type of ref_to_res.
        (nvptx_goacc_reduction_teardown):
        * config/nvptx/nvptx.md (cond_uni): New pattern.

commit 401876d422c4fa7f02c1b899e81568eea6ad7531
Author: Julian Brown <julian@codesourcery.com>
Date:   Tue Dec 11 13:35:52 2018 -0800

    Various OpenACC reduction enhancements - ME and nvptx changes
    
    	gcc/
    	* config/nvptx/nvptx.c (nvptx_propagate_unified): New.
    	(nvptx_split_blocks): Call it for cond_uni insn.
    	(nvptx_expand_cond_uni): New.
    	(enum nvptx_builtins): Add NVPTX_BUILTIN_COND_UNI.
    	(nvptx_init_builtins): Initialize it.
    	(nvptx_expand_builtin):
    	(nvptx_generate_vector_shuffle): Change integral SHIFT operand to
    	tree BITS operand.
    	(nvptx_vector_reduction): New.
    	(nvptx_adjust_reduction_type): New.
    	(nvptx_goacc_reduction_setup): Use it to adjust the type of ref_to_res.
    	(nvptx_goacc_reduction_init): Don't update LHS if it doesn't exist.
    	(nvptx_goacc_reduction_fini): Call nvptx_vector_reduction for vector.
    	Use it to adjust the type of ref_to_res.
    	(nvptx_goacc_reduction_teardown):
    	* config/nvptx/nvptx.md (cond_uni): New pattern.

diff --git a/gcc/config/nvptx/nvptx.c b/gcc/config/nvptx/nvptx.c
index 9903a27..0023dad 100644
--- a/gcc/config/nvptx/nvptx.c
+++ b/gcc/config/nvptx/nvptx.c
@@ -2863,6 +2863,52 @@ nvptx_reorg_uniform_simt ()
     }
 }
 
+/* UNIFIED is a cond_uni insn.  Find the branch insn it affects, and
+   mark that as unified.  We expect to be in a single block.  */
+
+static void
+nvptx_propagate_unified (rtx_insn *unified)
+{
+  rtx_insn *probe = unified;
+  rtx cond_reg = SET_DEST (PATTERN (unified));
+  rtx pat = NULL_RTX;
+
+  /* Find the comparison.  (We could skip this and simply scan to he
+     blocks' terminating branch, if we didn't care for self
+     checking.)  */
+  for (;;)
+    {
+      probe = next_real_insn (probe);
+      if (!probe)
+	break;
+      pat = PATTERN (probe);
+
+      if (GET_CODE (pat) == SET
+	  && GET_RTX_CLASS (GET_CODE (SET_SRC (pat))) == RTX_COMPARE
+	  && XEXP (SET_SRC (pat), 0) == cond_reg)
+	break;
+      gcc_assert (NONJUMP_INSN_P (probe));
+    }
+  gcc_assert (pat);
+  rtx pred_reg = SET_DEST (pat);
+
+  /* Find the branch.  */
+  do
+    probe = NEXT_INSN (probe);
+  while (!JUMP_P (probe));
+
+  pat = PATTERN (probe);
+  rtx itec = XEXP (SET_SRC (pat), 0);
+  gcc_assert (XEXP (itec, 0) == pred_reg);
+
+  /* Mark the branch's condition as unified.  */
+  rtx unspec = gen_rtx_UNSPEC (BImode, gen_rtvec (1, pred_reg),
+			       UNSPEC_BR_UNIFIED);
+  bool ok = validate_change (probe, &XEXP (itec, 0), unspec, false);
+
+  gcc_assert (ok);
+}
+
 /* Loop structure of the function.  The entire function is described as
    a NULL loop.  */
 
@@ -2964,6 +3010,9 @@ nvptx_split_blocks (bb_insn_map_t *map)
 	    continue;
 	  switch (recog_memoized (insn))
 	    {
+	    case CODE_FOR_cond_uni:
+	      nvptx_propagate_unified (insn);
+	      /* FALLTHROUGH */
 	    default:
 	      seen_insn = true;
 	      continue;
@@ -5083,6 +5132,21 @@ nvptx_expand_cmp_swap (tree exp, rtx target,
   return target;
 }
 
+/* Expander for the compare unified builtin.  */
+
+static rtx
+nvptx_expand_cond_uni (tree exp, rtx target, machine_mode mode, int ignore)
+{
+  if (ignore)
+    return target;
+
+  rtx src = expand_expr (CALL_EXPR_ARG (exp, 0),
+			 NULL_RTX, mode, EXPAND_NORMAL);
+
+  emit_insn (gen_cond_uni (target, src));
+
+  return target;
+}
 
 /* Codes for all the NVPTX builtins.  */
 enum nvptx_builtins
@@ -5092,6 +5156,7 @@ enum nvptx_builtins
   NVPTX_BUILTIN_WORKER_ADDR,
   NVPTX_BUILTIN_CMP_SWAP,
   NVPTX_BUILTIN_CMP_SWAPLL,
+  NVPTX_BUILTIN_COND_UNI,
   NVPTX_BUILTIN_MAX
 };
 
@@ -5129,6 +5194,7 @@ nvptx_init_builtins (void)
        (PTRVOID, ST, UINT, UINT, NULL_TREE));
   DEF (CMP_SWAP, "cmp_swap", (UINT, PTRVOID, UINT, UINT, NULL_TREE));
   DEF (CMP_SWAPLL, "cmp_swapll", (LLUINT, PTRVOID, LLUINT, LLUINT, NULL_TREE));
+  DEF (COND_UNI, "cond_uni", (integer_type_node, integer_type_node, NULL_TREE));
 
 #undef DEF
 #undef ST
@@ -5161,6 +5227,9 @@ nvptx_expand_builtin (tree exp, rtx target, rtx ARG_UNUSED (subtarget),
     case NVPTX_BUILTIN_CMP_SWAPLL:
       return nvptx_expand_cmp_swap (exp, target, mode, ignore);
 
+    case NVPTX_BUILTIN_COND_UNI:
+      return nvptx_expand_cond_uni (exp, target, mode, ignore);
+
     default: gcc_unreachable ();
     }
 }
@@ -5284,7 +5353,7 @@ nvptx_get_worker_red_addr (tree type, tree offset)
 
 static void
 nvptx_generate_vector_shuffle (location_t loc,
-			       tree dest_var, tree var, unsigned shift,
+			       tree dest_var, tree var, tree bits,
 			       gimple_seq *seq)
 {
   unsigned fn = NVPTX_BUILTIN_SHUFFLE;
@@ -5307,7 +5376,6 @@ nvptx_generate_vector_shuffle (location_t loc,
     }
   
   tree call = nvptx_builtin_decl (fn, true);
-  tree bits = build_int_cst (unsigned_type_node, shift);
   tree kind = build_int_cst (unsigned_type_node, SHUFFLE_DOWN);
   tree expr;
 
@@ -5583,6 +5651,128 @@ nvptx_reduction_update (location_t loc, gimple_stmt_iterator *gsi,
     return nvptx_lockfull_update (loc, gsi, ptr, var, op);
 }
 
+/* Emit a vector-level reduction loop.  OLD_VAR is the incoming
+   variable to reduce (valid in each vector), OP is the reduction
+   operator.  Return the reduced value (an SSA var).
+
+   The code we generate looks like:
+      unsigned old_shift = DIM_SIZE (VECTOR);
+      do
+	{
+	  shift = PHI (old_shift, new_shift);
+	  var = PHI (old_var, new_var);
+	  new_shift = shift >> 1;
+	  other_var = VSHUFFLE (var, new_shift);
+	  new_var = var OP other_var;
+	  cond_var = builtin_cond_uni (new_shift);
+	}
+      while (cond_var > 1);
+
+  The builtin_cond_ini expands to a cond_uni instruction, which is
+  processed in nvptx_split_blocks to mark the loop's terminating
+  branch instruction.  */
+
+static tree
+nvptx_vector_reduction (location_t loc, gimple_stmt_iterator *gsi,
+			tree old_var, tree_code op)
+{
+  tree var_type = TREE_TYPE (old_var);
+
+  /*  Emit old_shift = DIM_SIZE (VECTOR) */
+  tree old_shift = make_ssa_name (integer_type_node);
+  tree dim = build_int_cst (integer_type_node, GOMP_DIM_VECTOR);
+  gcall *call = gimple_build_call_internal (IFN_GOACC_DIM_SIZE, 1, dim);
+  gimple_set_lhs (call, old_shift);
+  gimple_set_location (call, loc);
+  gsi_insert_before (gsi, call, GSI_SAME_STMT);
+
+  /* Split the block just after the init stmts.  */
+  basic_block pre_bb = gsi_bb (*gsi);
+  edge pre_edge = split_block (pre_bb, call);
+  basic_block loop_bb = pre_edge->dest;
+  pre_bb = pre_edge->src;
+  /* Reset the iterator.  */
+  *gsi = gsi_for_stmt (gsi_stmt (*gsi));
+
+  tree shift = make_ssa_name (integer_type_node);
+  tree new_shift = make_ssa_name (integer_type_node);
+  tree var = make_ssa_name (var_type);
+  tree other_var = make_ssa_name (var_type);
+  tree new_var = make_ssa_name (var_type);
+
+  /* Build and insert the loop body.  */
+  gimple_seq loop_seq = NULL;
+
+  /* new_shift = shift >> 1 */
+  tree shift_expr = fold_build2 (RSHIFT_EXPR, integer_type_node,
+				 shift, integer_one_node);
+  gimplify_assign (new_shift, shift_expr, &loop_seq);
+
+  /* other_var = shuffle (var, shift) */
+  nvptx_generate_vector_shuffle (loc, other_var, var, new_shift, &loop_seq);
+  /* new_var = var OP other_var */
+  tree red_expr = fold_build2 (op, var_type, var, other_var);
+  gimplify_assign (new_var, red_expr, &loop_seq);
+
+  /* Mark the iterator variable as unified.  */
+  tree cond_var = make_ssa_name (integer_type_node);
+  tree uni_fn = nvptx_builtin_decl (NVPTX_BUILTIN_COND_UNI, true);
+  tree uni_expr = build_call_expr_loc (loc, uni_fn, 1, new_shift);
+  gimplify_assign (cond_var,  uni_expr, &loop_seq);
+
+  gcond *cond = gimple_build_cond (LE_EXPR, cond_var, integer_one_node,
+				   NULL_TREE, NULL_TREE);
+  gimple_seq_add_stmt (&loop_seq, cond);
+
+  gsi_insert_seq_before (gsi, loop_seq, GSI_SAME_STMT);
+
+  /* Split the block just after the loop stmts.  */
+  edge post_edge = split_block (loop_bb, cond);
+  post_edge->probability = profile_probability::even ();
+  basic_block post_bb = post_edge->dest;
+  loop_bb = post_edge->src;
+  *gsi = gsi_for_stmt (gsi_stmt (*gsi));
+
+  /* Create the loop.  */
+  post_edge->flags ^= EDGE_TRUE_VALUE | EDGE_FALLTHRU;
+  edge loop_edge = make_edge (loop_bb, loop_bb, EDGE_FALSE_VALUE);
+  loop_edge->probability = profile_probability::even ();
+  set_immediate_dominator (CDI_DOMINATORS, loop_bb, pre_bb);
+  set_immediate_dominator (CDI_DOMINATORS, post_bb, loop_bb);
+
+  gphi *shift_phi = create_phi_node (shift, loop_bb);
+  add_phi_arg (shift_phi, old_shift, pre_edge, loc);
+  add_phi_arg (shift_phi, new_shift, loop_edge, loc);
+
+  gphi *var_phi = create_phi_node (var, loop_bb);
+  add_phi_arg (var_phi, old_var, pre_edge, loc);
+  add_phi_arg (var_phi, new_var, loop_edge, loc);
+
+  loop *loop = alloc_loop ();
+  loop->header = loop_bb;
+  loop->latch = loop_bb;
+  add_loop (loop, loop_bb->loop_father);
+
+  return new_var;
+}
+
+/* Dummy reduction vars that have GOMP_MAP_FIRSTPRIVATE_POINTER data
+   mappings gets retyped to (void *).  Adjust the type of VAR to TYPE
+   as appropriate.  */
+
+static tree
+nvptx_adjust_reduction_type (tree var, tree type, gimple_seq *seq)
+{
+  if (TREE_TYPE (TREE_TYPE (var)) == type)
+    return var;
+
+  tree ptype = build_pointer_type (type);
+  tree t = make_ssa_name (ptype);
+  tree expr = fold_build1 (NOP_EXPR, ptype, var);
+  gimple_seq_add_stmt (seq, gimple_build_assign (t, expr));
+  return t;
+}
+
 /* NVPTX implementation of GOACC_REDUCTION_SETUP.  */
 
 static void
@@ -5602,7 +5792,11 @@ nvptx_goacc_reduction_setup (gcall *call)
       tree ref_to_res = gimple_call_arg (call, 1);
 
       if (!integer_zerop (ref_to_res))
-	var = build_simple_mem_ref (ref_to_res);
+	{
+	  ref_to_res = nvptx_adjust_reduction_type (ref_to_res, TREE_TYPE (var),
+						    &seq);
+	  var = build_simple_mem_ref (ref_to_res);
+	}
     }
   
   if (level == GOMP_DIM_WORKER)
@@ -5702,7 +5896,11 @@ nvptx_goacc_reduction_init (gcall *call)
 	    init = var;
 	}
 
-      gimplify_assign (lhs, init, &seq);
+      /* The LHS may be NULL if a reduction variable on a parallel
+	 construct is initialized to some constant inside the parallel
+	 region.  */
+      if (lhs)
+	gimplify_assign (lhs, init, &seq);
     }
 
   pop_gimplify_context (NULL);
@@ -5727,22 +5925,7 @@ nvptx_goacc_reduction_fini (gcall *call)
   push_gimplify_context (true);
 
   if (level == GOMP_DIM_VECTOR)
-    {
-      /* Emit binary shuffle tree.  TODO. Emit this as an actual loop,
-	 but that requires a method of emitting a unified jump at the
-	 gimple level.  */
-      for (int shfl = PTX_VECTOR_LENGTH / 2; shfl > 0; shfl = shfl >> 1)
-	{
-	  tree other_var = make_ssa_name (TREE_TYPE (var));
-	  nvptx_generate_vector_shuffle (gimple_location (call),
-					 other_var, var, shfl, &seq);
-
-	  r = make_ssa_name (TREE_TYPE (var));
-	  gimplify_assign (r, fold_build2 (op, TREE_TYPE (var),
-					   var, other_var), &seq);
-	  var = r;
-	}
-    }
+    r = nvptx_vector_reduction (gimple_location (call), &gsi, var, op);
   else
     {
       tree accum = NULL_TREE;
@@ -5760,7 +5943,11 @@ nvptx_goacc_reduction_fini (gcall *call)
       else if (integer_zerop (ref_to_res))
 	r = var;
       else
-	accum = ref_to_res;
+	{
+	  ref_to_res = nvptx_adjust_reduction_type (ref_to_res, TREE_TYPE (var),
+						    &seq);
+	  accum = ref_to_res;
+	}
 
       if (accum)
 	{
@@ -5809,7 +5996,11 @@ nvptx_goacc_reduction_teardown (gcall *call)
       tree ref_to_res = gimple_call_arg (call, 1);
 
       if (!integer_zerop (ref_to_res))
-	gimplify_assign (build_simple_mem_ref (ref_to_res), var, &seq);
+	{
+	  ref_to_res = nvptx_adjust_reduction_type (ref_to_res, TREE_TYPE (var),
+						    &seq);
+	  gimplify_assign (build_simple_mem_ref (ref_to_res), var, &seq);
+	}
     }
 
   if (lhs)
diff --git a/gcc/config/nvptx/nvptx.md b/gcc/config/nvptx/nvptx.md
index ca00b1d..4f7d8ea 100644
--- a/gcc/config/nvptx/nvptx.md
+++ b/gcc/config/nvptx/nvptx.md
@@ -600,6 +600,13 @@
   "%J0\\tbra.uni\\t%l1;"
   [(set_attr "predicable" "false")])
 
+(define_insn "cond_uni"
+  [(set (match_operand:SI 0 "nvptx_register_operand" "=R")
+          (unspec:SI [(match_operand:SI 1 "nvptx_nonmemory_operand" "R")]
+		     UNSPEC_BR_UNIFIED))]
+  ""
+  "%.\\tmov%t0\\t%0, %1; // unified")
+
 (define_expand "cbranch<mode>4"
   [(set (pc)
 	(if_then_else (match_operator 0 "nvptx_comparison_operator"

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]