[PATCH v2] S/390: Allow LARL of literal pool entries

[PATCH v2] S/390: Allow LARL of literal pool entries

[Ilya Leoshkevich]

Bootstrapped and regtested on s390x-redhat-linux.

Changes since v1:
* Removed unnecessary gen_rtx_CONST call.
* UNSPEC_LTREF are now omitted for all relatively addressed pool
  entries, and not only those which occur in LARL-like patterns.

r265490 allowed the compiler to choose in a more flexible way whether to
use load or load-address-relative-long (LARL) instruction.  When it
chose LARL for literal pool references, the latter ones were rewritten
by pass_s390_early_mach to use UNSPEC_LTREF, which assumes base register
usage, which in turn is not compatible with LARL.  The end result was an
ICE because of unrecognizable insn.

UNSPEC_LTREF and friends are necessary in order to communicate the
dependency on the base register to pass_sched2.  When LARL is used, no
base register is necessary, so in such cases the rewrite must be
avoided.

gcc/ChangeLog:

2018-10-26  Ilya Leoshkevich  <iii@linux.ibm.com>

	PR target/87762
	* config/s390/predicates.md (larl_operand): Use
	s390_symbol_larl_p () to reduce code duplication.
	* config/s390/s390-protos.h (s390_symbol_larl_p): New function.
	* config/s390/s390.c (s390_symbol_larl_p): New function.
	(annotate_constant_pool_refs): Skip LARL operands.
	(find_constant_pool_ref): Handle non-annotated literal pool
	references, which are usable with LARL.
	(replace_constant_pool_ref): Skip LARL operands.
---
 gcc/config/s390/predicates.md |  9 ++-------
 gcc/config/s390/s390-protos.h |  1 +
 gcc/config/s390/s390.c        | 30 ++++++++++++++++++++++++++++++
 3 files changed, 33 insertions(+), 7 deletions(-)

diff --git a/gcc/config/s390/predicates.md b/gcc/config/s390/predicates.md
index 98a824e77b7..0e431302479 100644
--- a/gcc/config/s390/predicates.md
+++ b/gcc/config/s390/predicates.md
@@ -151,9 +151,7 @@
   if (GET_CODE (op) == LABEL_REF)
     return true;
   if (SYMBOL_REF_P (op))
-    return (!SYMBOL_FLAG_NOTALIGN2_P (op)
-	    && SYMBOL_REF_TLS_MODEL (op) == 0
-	    && s390_rel_address_ok_p (op));
+    return s390_symbol_larl_p (op);
 
   /* Everything else must have a CONST, so strip it.  */
   if (GET_CODE (op) != CONST)
@@ -176,10 +174,7 @@
   if (GET_CODE (op) == LABEL_REF)
     return true;
   if (SYMBOL_REF_P (op))
-    return (!SYMBOL_FLAG_NOTALIGN2_P (op)
-	    && SYMBOL_REF_TLS_MODEL (op) == 0
-	    && s390_rel_address_ok_p (op));
-
+    return s390_symbol_larl_p (op);
 
   /* Now we must have a @GOTENT offset or @PLT stub
      or an @INDNTPOFF TLS offset.  */
diff --git a/gcc/config/s390/s390-protos.h b/gcc/config/s390/s390-protos.h
index 96fa705f879..a5cd80fa446 100644
--- a/gcc/config/s390/s390-protos.h
+++ b/gcc/config/s390/s390-protos.h
@@ -157,6 +157,7 @@ extern void s390_indirect_branch_via_thunk (unsigned int regno,
 					    rtx comparison_operator,
 					    enum s390_indirect_branch_type type);
 extern void s390_indirect_branch_via_inline_thunk (rtx execute_target);
+extern bool s390_symbol_larl_p (rtx);
 #endif /* RTX_CODE */
 
 /* s390-c.c routines */
diff --git a/gcc/config/s390/s390.c b/gcc/config/s390/s390.c
index 29a829f48ea..bac215f647c 100644
--- a/gcc/config/s390/s390.c
+++ b/gcc/config/s390/s390.c
@@ -2816,6 +2816,16 @@ s390_decompose_constant_pool_ref (rtx *ref, rtx *disp, bool *is_ptr,
   return true;
 }
 
+/* Return true iff SYMBOL_REF X can be used with a LARL instruction. */
+
+bool
+s390_symbol_larl_p (rtx x)
+{
+  return (!SYMBOL_FLAG_NOTALIGN2_P (x)
+	  && SYMBOL_REF_TLS_MODEL (x) == 0
+	  && s390_rel_address_ok_p (x));
+}
+
 /* Decompose a RTL expression ADDR for a memory address into
    its components, returned in OUT.
 
@@ -8120,6 +8130,10 @@ annotate_constant_pool_refs (rtx *x)
   int i, j;
   const char *fmt;
 
+  /* Skip LARL operands, because they don't require a base register.  */
+  if (larl_operand (*x, VOIDmode))
+    return;
+
   gcc_assert (GET_CODE (*x) != SYMBOL_REF
 	      || !CONSTANT_POOL_ADDRESS_P (*x));
 
@@ -8223,6 +8237,18 @@ find_constant_pool_ref (rtx x, rtx *ref)
       && XINT (x, 1) == UNSPECV_POOL_ENTRY)
     return;
 
+  if (SYMBOL_REF_P (x)
+      && CONSTANT_POOL_ADDRESS_P (x)
+      && s390_symbol_larl_p (x))
+    {
+      if (*ref == NULL_RTX)
+	*ref = x;
+      else
+	gcc_assert (*ref == x);
+
+      return;
+    }
+
   gcc_assert (GET_CODE (x) != SYMBOL_REF
 	      || !CONSTANT_POOL_ADDRESS_P (x));
 
@@ -8264,6 +8290,10 @@ replace_constant_pool_ref (rtx *x, rtx ref, rtx offset)
   int i, j;
   const char *fmt;
 
+  /* Skip LARL operands, because they don't require a base register.  */
+  if (larl_operand (*x, VOIDmode))
+    return;
+
   gcc_assert (*x != ref);
 
   if (GET_CODE (*x) == UNSPEC
-- 
2.19.0

Re: [PATCH v2] S/390: Allow LARL of literal pool entries

[Ulrich Weigand]

Ilya Leoshkevich wrote:

> @@ -8223,6 +8237,18 @@ find_constant_pool_ref (rtx x, rtx *ref)
>        && XINT (x, 1) == UNSPECV_POOL_ENTRY)
>      return;
>  
> +  if (SYMBOL_REF_P (x)
> +      && CONSTANT_POOL_ADDRESS_P (x)
> +      && s390_symbol_larl_p (x))
> +    {
> +      if (*ref == NULL_RTX)
> +	*ref = x;
> +      else
> +	gcc_assert (*ref == x);
> +
> +      return;
> +    }

This definitely looks wrong.  If we haven't annotated the address,
it should *not* be found by find_constant_pool_ref, since we are
not going to replace it!  That was the whole point of not annotating
it in the first place ...

Bye,
Ulrich

-- 
  Dr. Ulrich Weigand
  GNU/Linux compilers and toolchain
  Ulrich.Weigand@de.ibm.com

Re: [PATCH v2] S/390: Allow LARL of literal pool entries

[Ilya Leoshkevich]

> Am 30.10.2018 um 18:22 schrieb Ulrich Weigand <uweigand@de.ibm.com>:
> 
> Ilya Leoshkevich wrote:
> 
>> @@ -8223,6 +8237,18 @@ find_constant_pool_ref (rtx x, rtx *ref)
>>       && XINT (x, 1) == UNSPECV_POOL_ENTRY)
>>     return;
>> 
>> +  if (SYMBOL_REF_P (x)
>> +      && CONSTANT_POOL_ADDRESS_P (x)
>> +      && s390_symbol_larl_p (x))
>> +    {
>> +      if (*ref == NULL_RTX)
>> +	*ref = x;
>> +      else
>> +	gcc_assert (*ref == x);
>> +
>> +      return;
>> +    }
> 
> This definitely looks wrong.  If we haven't annotated the address,
> it should *not* be found by find_constant_pool_ref, since we are
> not going to replace it!  That was the whole point of not annotating
> it in the first place …

There are two use cases for find_constant_pool_ref ().  One is indeed
replacing annotated references.  The other (in s390_mainpool_start ()
and s390_chunkify_start ()) is creating pool entries.  So I've decided
to let it find unannotated references for the second use case.

This impacts the first use case as well, that's why I have also changed
replace_constant_pool_ref () to ignore unannotated references.

Re: [PATCH v2] S/390: Allow LARL of literal pool entries

[Ulrich Weigand]

Ilya Leoshkevich wrote:
> Am 30.10.2018 um 18:22 schrieb Ulrich Weigand <uweigand@de.ibm.com>:
> > This definitely looks wrong.  If we haven't annotated the address,
> > it should *not* be found by find_constant_pool_ref, since we are
> > not going to replace it!  That was the whole point of not annotating
> > it in the first place ...
> 
> There are two use cases for find_constant_pool_ref ().  One is indeed
> replacing annotated references.  The other (in s390_mainpool_start ()
> and s390_chunkify_start ()) is creating pool entries.  So I've decided
> to let it find unannotated references for the second use case.

OK, but if we access the constant via relative address, we don't need
to copy it into the back-end managed pool either; the relative address
can just refer the constant in the default pool maintained by the
middle end.

Bye,
Ulrich

-- 
  Dr. Ulrich Weigand
  GNU/Linux compilers and toolchain
  Ulrich.Weigand@de.ibm.com

Re: [PATCH v2] S/390: Allow LARL of literal pool entries

[Ilya Leoshkevich]

> Am 31.10.2018 um 10:59 schrieb Ulrich Weigand <uweigand@de.ibm.com>:
> 
> Ilya Leoshkevich wrote:
>> Am 30.10.2018 um 18:22 schrieb Ulrich Weigand <uweigand@de.ibm.com>:
>>> This definitely looks wrong.  If we haven't annotated the address,
>>> it should *not* be found by find_constant_pool_ref, since we are
>>> not going to replace it!  That was the whole point of not annotating
>>> it in the first place ...
>> 
>> There are two use cases for find_constant_pool_ref ().  One is indeed
>> replacing annotated references.  The other (in s390_mainpool_start ()
>> and s390_chunkify_start ()) is creating pool entries.  So I've decided
>> to let it find unannotated references for the second use case.
> 
> OK, but if we access the constant via relative address, we don't need
> to copy it into the back-end managed pool either; the relative address
> can just refer the constant in the default pool maintained by the
> middle end.

Wouldn’t that prevent constant merging in case the same constant is
used with both relative and base-register addressing?