store-merging, riscv: Consider widen_bswap_or_bitreverse in store-merging decisions...

On Fri, Jun 05, 2026 at 02:22:25PM +0200, Andreas Schwab wrote:
> This breaks riscv:
>
> ../../gcc/config/riscv/riscv.md:5297:14: error: bswapsi2 cannot FAIL
> ../../gcc/config/riscv/bitmanip.md:543:5: note: in expansion of macro 'FAIL'
>   543 |     FAIL;
>       |     ^~~~

wtw mentioned that on IRC.
In the way IFN_BSWAP/IFN_BITREVERSE are used currently (i.e. only used
for large/huge _BitInt or in the C++ FE, but not making it to expansion)
there is no reason why FAIL wouldn't be allowed (like before my changes),
though I find the riscv hack really ugly.
Instead of adding a workaround, this patch removes the riscv hack and
instead adjusts store-merging pass to take into account
widen_bswap_or_bitreverse; currently it already takes into account
expand_doubleword_bswap_or_bitreverse (on 32-bit word targets which have
bswapsi2 expanders expects __builtin_bswap64 to be cheap).  With this
patch, it expects __builtin_bswap32 to be cheap if bswapdi2 expander
is present, because widen_bswap_or_bitreverse will handle it as
(subreg:SI (lshiftrt:DI (bswap:DI (subreg:DI (arg:SI) 0)) (const_int 32)) low)

2026-06-05  Jakub Jelinek  <jakub@redhat.com>

	* gimple-ssa-store-merging.cc (maybe_optimize_vector_constructor):
	Also support bswap32 if bswapdi2 expander is present.  Add comment
	about bswap64 support on 32-bit targets with bswapsi2 expander.
	(pass_optimize_bswap::execute): Likewise.
	* config/riscv/bitmanip.md (bswapsi2): Change condition to disable
	the expander on TARGET_64BIT without TARGET_XTHEADBB, never FAIL.

Reviewed-by: Richard Sandiford <rdsandiford@googlemail.com>
Reviewed-by: Richard Biener <rguenth@suse.de>

diff --git a/gcc/config/riscv/bitmanip.md b/gcc/config/riscv/bitmanip.md
index 05559405fcc966e2f2c5bdbbf78f0e7bdd1d22af..bd2aaf0587ebe58530cfc8840d0d5e976a387e90 100644 (file)
--- a/gcc/config/riscv/bitmanip.md
+++ b/gcc/config/riscv/bitmanip.md
@@ -533,16 +533,7 @@
 (define_expand "bswapsi2"
   [(set (match_operand:SI 0 "register_operand")
        (bswap:SI (match_operand:SI 1 "register_operand")))]
-  "TARGET_ZBB || TARGET_ZBKB || TARGET_XTHEADBB"
-{
-  /* Expose bswapsi2 on TARGET_64BIT so that the gimple store
-     merging pass will create suitable bswap insns.  We can actually
-     just FAIL that case when generating RTL and let the generic code
-     handle it.  */
-  if (TARGET_64BIT && !TARGET_XTHEADBB)
-    FAIL;
-})
-
+  "(!TARGET_64BIT && (TARGET_ZBB || TARGET_ZBKB)) || TARGET_XTHEADBB")
 
 (define_insn "*bswap<mode>2"
   [(set (match_operand:X 0 "register_operand" "=r")

diff --git a/gcc/gimple-ssa-store-merging.cc b/gcc/gimple-ssa-store-merging.cc
index 96fd9fa73a35bb6330adca41c4deed0d015d9eac..2e5232fa65fff802e69126a679aca9e841d27811 100644 (file)
--- a/gcc/gimple-ssa-store-merging.cc
+++ b/gcc/gimple-ssa-store-merging.cc
@@ -1570,7 +1570,10 @@ maybe_optimize_vector_constructor (gimple *cur_stmt)
       break;
     case 32:
       if (builtin_decl_explicit_p (BUILT_IN_BSWAP32)
-         && optab_handler (bswap_optab, SImode) != CODE_FOR_nothing)
+         && (optab_handler (bswap_optab, SImode) != CODE_FOR_nothing
+             /* widen_bswap_or_bitreverse can implement 32-bit bswap
+                using bswapdi2 and shift.  */
+             || optab_handler (bswap_optab, DImode) != CODE_FOR_nothing))
        {
          load_type = uint32_type_node;
          fndecl = builtin_decl_explicit (BUILT_IN_BSWAP32);
@@ -1582,6 +1585,8 @@ maybe_optimize_vector_constructor (gimple *cur_stmt)
     case 64:
       if (builtin_decl_explicit_p (BUILT_IN_BSWAP64)
          && (optab_handler (bswap_optab, DImode) != CODE_FOR_nothing
+             /* expand_doubleword_bswap_or_bitreverse can use 2 bswapsi2
+                expanders on 32-bit targets.  */
              || (word_mode == SImode
                  && builtin_decl_explicit_p (BUILT_IN_BSWAP32)
                  && optab_handler (bswap_optab, SImode) != CODE_FOR_nothing)))
@@ -1631,9 +1636,15 @@ pass_optimize_bswap::execute (function *fun)
   tree bswap32_type = NULL_TREE, bswap64_type = NULL_TREE;
 
   bswap32_p = (builtin_decl_explicit_p (BUILT_IN_BSWAP32)
-              && optab_handler (bswap_optab, SImode) != CODE_FOR_nothing);
+              && (optab_handler (bswap_optab, SImode) != CODE_FOR_nothing
+                  /* widen_bswap_or_bitreverse can implement 32-bit bswap
+                     using bswapdi2 and shift.  */
+                  || (optab_handler (bswap_optab, DImode)
+                      != CODE_FOR_nothing)));
   bswap64_p = (builtin_decl_explicit_p (BUILT_IN_BSWAP64)
               && (optab_handler (bswap_optab, DImode) != CODE_FOR_nothing
+                  /* expand_doubleword_bswap_or_bitreverse can use 2 bswapsi2
+                     expanders on 32-bit targets.  */
                   || (bswap32_p && word_mode == SImode)));
 
   /* Determine the argument type of the builtins.  The code later on