[pushed] aarch64: Fix SMSTART/SMSTOP save/restore for BE

Message ID mpth6kpkf3r.fsf@arm.com
State Unresolved
Headers
Series [pushed] aarch64: Fix SMSTART/SMSTOP save/restore for BE |

Checks

Context Check Description
snail/gcc-patch-check warning Git am fail log

Commit Message

Richard Sandiford Dec. 10, 2023, 7:54 p.m. UTC
  VNx16QI (the SVE register byte mode) is the only SVE mode for which
LD1 and LDR result in the same register layout for big-endian.  It is
therefore the only mode for which we allow LDR and STR to be used for
big-endian SVE moves.

The SME support sometimes needs to use LDR and STR to save and restore
Z register contents around an SMSTART/SMSTOP SM.  It therefore needs
to use VNx16QI regardless of the type of value that is stored in the
Z registers.

Tested on aarch64-linux-gnu and aarch64_be-elf, pushed to trunk.

Richard


gcc/
	PR target/112930
	* config/aarch64/aarch64.cc (aarch64_sme_mode_switch_regs::add_reg):
	Force specific SVE modes for single registers as well as structures.
---
 gcc/config/aarch64/aarch64.cc | 15 +++++++++------
 1 file changed, 9 insertions(+), 6 deletions(-)
  

Patch

diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc
index 5cffdabc62e..2a64053f675 100644
--- a/gcc/config/aarch64/aarch64.cc
+++ b/gcc/config/aarch64/aarch64.cc
@@ -4956,14 +4956,17 @@  aarch64_sme_mode_switch_regs::add_reg (machine_mode mode, unsigned int regno)
   gcc_assert ((vec_flags & VEC_STRUCT) || end_regno == regno + 1);
   for (; regno < end_regno; regno++)
     {
+      /* Force the mode of SVE saves and restores even for single registers.
+	 This is necessary because big-endian targets only allow LDR Z and
+	 STR Z to be used with byte modes.  */
       machine_mode submode = mode;
-      if (vec_flags & VEC_STRUCT)
+      if (vec_flags & VEC_SVE_PRED)
+	submode = VNx16BImode;
+      else if (vec_flags & VEC_SVE_DATA)
+	submode = SVE_BYTE_MODE;
+      else if (vec_flags & VEC_STRUCT)
 	{
-	  if (vec_flags & VEC_SVE_PRED)
-	    submode = VNx16BImode;
-	  else if (vec_flags & VEC_SVE_DATA)
-	    submode = SVE_BYTE_MODE;
-	  else if (vec_flags & VEC_PARTIAL)
+	  if (vec_flags & VEC_PARTIAL)
 	    submode = V8QImode;
 	  else
 	    submode = V16QImode;