[committed] [PR104637] LRA: Split hard regs as many as possible on one subpass

[committed] [PR104637] LRA: Split hard regs as many as possible on one subpass

[Vladimir Makarov]

[-- Attachment #1: Type: text/plain, Size: 164 bytes --]

The following patch fixes

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104637

The patch was successfully bootstrapped and tested on x86-64, aarch64, 
and ppc64.


[-- Attachment #2: pr104637.patch --]
[-- Type: text/x-patch, Size: 4797 bytes --]

commit ec1b9ba2d7913fe5e9deacc8e55e7539262f5124
Author: Vladimir N. Makarov <vmakarov@redhat.com>
Date:   Mon Feb 28 16:43:50 2022 -0500

    [PR104637] LRA: Split hard regs as many as possible on one subpass
    
    LRA hard reg split subpass is a small subpass used as the last
    resort for LRA when it can not assign a hard reg to a reload
    pseudo by other ways (e.g. by spilling non-reload pseudos).  For
    simplicity the subpass works on one split base (as each split
    changes pseudo live range info).  In this case it results in
    reaching maximal possible number of subpasses.  The patch
    implements as many non-overlapping hard reg splits
    splits as possible on each subpass.
    
    gcc/ChangeLog:
    
            PR rtl-optimization/104637
            * lra-assigns.cc (lra_split_hard_reg_for): Split hard regs as many
            as possible on one subpass.
    
    gcc/testsuite/ChangeLog:
    
            PR rtl-optimization/104637
            * gcc.target/i386/pr104637.c: New.

diff --git a/gcc/lra-assigns.cc b/gcc/lra-assigns.cc
index c1d40ea2a14..ab3a6e6e9cc 100644
--- a/gcc/lra-assigns.cc
+++ b/gcc/lra-assigns.cc
@@ -1774,8 +1774,8 @@ lra_split_hard_reg_for (void)
      iterations.  Either it's an asm and something is wrong with the
      constraints, or we have run out of spill registers; error out in
      either case.  */
-  bool asm_p = false;
-  bitmap_head failed_reload_insns, failed_reload_pseudos;
+  bool asm_p = false, spill_p = false;
+  bitmap_head failed_reload_insns, failed_reload_pseudos, over_split_insns;
   
   if (lra_dump_file != NULL)
     fprintf (lra_dump_file,
@@ -1786,6 +1786,7 @@ lra_split_hard_reg_for (void)
   bitmap_ior (&non_reload_pseudos, &lra_inheritance_pseudos, &lra_split_regs);
   bitmap_ior_into (&non_reload_pseudos, &lra_subreg_reload_pseudos);
   bitmap_ior_into (&non_reload_pseudos, &lra_optional_reload_pseudos);
+  bitmap_initialize (&over_split_insns, &reg_obstack);
   for (i = lra_constraint_new_regno_start; i < max_regno; i++)
     if (reg_renumber[i] < 0 && lra_reg_info[i].nrefs != 0
 	&& (rclass = lra_get_allocno_class (i)) != NO_REGS
@@ -1793,14 +1794,41 @@ lra_split_hard_reg_for (void)
       {
 	if (! find_reload_regno_insns (i, first, last))
 	  continue;
-	if (BLOCK_FOR_INSN (first) == BLOCK_FOR_INSN (last)
-	    && spill_hard_reg_in_range (i, rclass, first, last))
+	if (BLOCK_FOR_INSN (first) == BLOCK_FOR_INSN (last))
 	  {
-	    bitmap_clear (&failed_reload_pseudos);
-	    return true;
+	    /* Check that we are not trying to split over the same insn
+	       requiring reloads to avoid splitting the same hard reg twice or
+	       more.  If we need several hard regs splitting over the same insn
+	       it can be finished on the next iterations.
+
+	       The following loop iteration number is small as we split hard
+	       reg in a very small range.  */
+	    for (insn = first;
+		 insn != NEXT_INSN (last);
+		 insn = NEXT_INSN (insn))
+	      if (bitmap_bit_p (&over_split_insns, INSN_UID (insn)))
+		break;
+	    if (insn != NEXT_INSN (last)
+		|| !spill_hard_reg_in_range (i, rclass, first, last))
+	      {
+		bitmap_set_bit (&failed_reload_pseudos, i);
+	      }
+	    else
+	      {
+		for (insn = first;
+		     insn != NEXT_INSN (last);
+		     insn = NEXT_INSN (insn))
+		  bitmap_set_bit (&over_split_insns, INSN_UID (insn));
+		spill_p = true;
+	      }
 	  }
-	bitmap_set_bit (&failed_reload_pseudos, i);
       }
+  bitmap_clear (&over_split_insns);
+  if (spill_p)
+    {
+      bitmap_clear (&failed_reload_pseudos);
+      return true;
+    }
   bitmap_clear (&non_reload_pseudos);
   bitmap_initialize (&failed_reload_insns, &reg_obstack);
   EXECUTE_IF_SET_IN_BITMAP (&failed_reload_pseudos, 0, u, bi)
diff --git a/gcc/testsuite/gcc.target/i386/pr104637.c b/gcc/testsuite/gcc.target/i386/pr104637.c
new file mode 100644
index 00000000000..65e8635d55e
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr104637.c
@@ -0,0 +1,22 @@
+/* { dg-do compile } */
+/* { dg-options "-Og -fno-forward-propagate -mavx" } */
+
+typedef short __attribute__((__vector_size__ (64))) U;
+typedef unsigned long long __attribute__((__vector_size__ (32))) V;
+typedef long double __attribute__((__vector_size__ (64))) F;
+
+int i;
+U u;
+F f;
+
+void
+foo (char a, char b, _Complex char c, V v)
+{
+  u = (U) { u[0] / 0, u[1] / 0, u[2] / 0, u[3] / 0, u[4] / 0, u[5] / 0, u[6] / 0, u[7] / 0,
+	    u[8] / 0, u[0] / 0, u[9] / 0, u[10] / 0, u[11] / 0, u[12] / 0, u[13] / 0, u[14] / 0, u[15] / 0,
+	    u[16] / 0, u[17] / 0, u[18] / 0, u[19] / 0, u[20] / 0, u[21] / 0, u[22] / 0, u[23] / 0,
+	    u[24] / 0, u[25] / 0, u[26] / 0, u[27] / 0, u[28] / 0, u[29] / 0, u[30] / 0, u[31] / 0 };
+  c += i;
+  f = (F) { v[0], v[1], v[2], v[3] };
+  i = (char) (__imag__ c + i);
+}

[committed] testsuite: Fix up pr104637 testcase [PR104637]

[Jakub Jelinek]

Hi!

On Mon, Feb 28, 2022 at 04:49:08PM -0500, Vladimir Makarov via Gcc-patches wrote:
>             PR rtl-optimization/104637
>             * gcc.target/i386/pr104637.c: New.

This testcase FAILs everywhere for 3 reasons:
1) the testcase can't work on ia32, where sizeof (long double) == 12
   and as it is not a power of 2, we disallow creating vectors with such
   elements, -mx32 and -m64 are fine
2) the testcase emits a lot of -Wdiv-by-zero warnings, I've just added
   -Wno-div-by-zero to dg-options
3) my fault, when tweaking the testcase I've missed 33 initializers of
   a 32 element vector which didn't change anything on the ICE, but is
   still reported

This patch fixes all of it, tested with
RUNTESTFLAGS='--target_board=unix\{-m32,-m64\} i386.exp=pr104637.c'
both without the LRA fix where it ICEs and with it where it passes
everywhere.  Committed to trunk as obvious.

2022-03-02  Jakub Jelinek  <jakub@redhat.com>

	PR rtl-optimization/104637
	* gcc.target/i386/pr104637.c: Don't run on ia32.  Add -Wno-div-by-zero
	to dg-options.
	(foo): Remove extraneous initializer.

--- gcc/testsuite/gcc.target/i386/pr104637.c.jj	2022-03-01 19:42:39.362262443 +0100
+++ gcc/testsuite/gcc.target/i386/pr104637.c	2022-03-02 10:52:43.943775712 +0100
@@ -1,5 +1,5 @@
-/* { dg-do compile } */
-/* { dg-options "-Og -fno-forward-propagate -mavx" } */
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-options "-Og -fno-forward-propagate -mavx -Wno-div-by-zero" } */
 
 typedef short __attribute__((__vector_size__ (64))) U;
 typedef unsigned long long __attribute__((__vector_size__ (32))) V;
@@ -13,7 +13,7 @@ void
 foo (char a, char b, _Complex char c, V v)
 {
   u = (U) { u[0] / 0, u[1] / 0, u[2] / 0, u[3] / 0, u[4] / 0, u[5] / 0, u[6] / 0, u[7] / 0,
-	    u[8] / 0, u[0] / 0, u[9] / 0, u[10] / 0, u[11] / 0, u[12] / 0, u[13] / 0, u[14] / 0, u[15] / 0,
+	    u[8] / 0, u[9] / 0, u[10] / 0, u[11] / 0, u[12] / 0, u[13] / 0, u[14] / 0, u[15] / 0,
 	    u[16] / 0, u[17] / 0, u[18] / 0, u[19] / 0, u[20] / 0, u[21] / 0, u[22] / 0, u[23] / 0,
 	    u[24] / 0, u[25] / 0, u[26] / 0, u[27] / 0, u[28] / 0, u[29] / 0, u[30] / 0, u[31] / 0 };
   c += i;


	Jakub