public inbox for gcc-cvs@sourceware.org
help / color / mirror / Atom feed
* [gcc r10-10397] rs6000: Disable optimizing multiple xxsetaccz instructions into one xxsetaccz
@ 2022-01-15  6:00 Peter Bergner
  0 siblings, 0 replies; only message in thread
From: Peter Bergner @ 2022-01-15  6:00 UTC (permalink / raw)
  To: gcc-cvs

https://gcc.gnu.org/g:2875c871c9c1c75035df662e13f9d69cd338e821

commit r10-10397-g2875c871c9c1c75035df662e13f9d69cd338e821
Author: Peter Bergner <bergner@linux.ibm.com>
Date:   Tue Sep 14 10:47:18 2021 -0500

    rs6000: Disable optimizing multiple xxsetaccz instructions into one xxsetaccz
    
    Fwprop will happily optimize two xxsetaccz instructions into one xxsetaccz
    by propagating the results of the first to the uses of the second.
    We really don't want that to happen given the late priming/depriming of
    accumulators.  I fixed this by making the xxsetaccz source operand an
    unspec volatile.  I also removed the mma_xxsetaccz define_expand and
    define_insn_and_split and replaced it with a simple define_insn.
    
    Rather than a new test case, I was able to just modify the current test case
    to add another __builtin_mma_xxsetaccz call which shows the bad code gen
    with unpatched compilers.
    
    2021-09-14  Peter Bergner  <bergner@linux.ibm.com>
    
    gcc/
            * config/rs6000/mma.md (unspecv): Add UNSPECV_MMA_XXSETACCZ.
            (*movpxi): Remove 'O' alternative.
            (mma_xxsetaccz): Change to define_insn.  Use UNSPECV_MMA_XXSETACCZ.
            Add comment.
            * config/rs6000/rs6000.c (rs6000_rtx_costs): Handle UNSPEC_VOLATILE.
    
    gcc/testsuite/
            * gcc.target/powerpc/mma-builtin-6.c: Add second call to xxsetacc
            built-in.  Update instruction counts.
    
    (cherry picked from commit f80b9be083e0e7d49e7744b7e531b9aa52acd563)

Diff:
---
 gcc/config/rs6000/mma.md                         | 36 +++++++++++++-----------
 gcc/config/rs6000/rs6000.c                       |  8 ++++++
 gcc/testsuite/gcc.target/powerpc/mma-builtin-6.c | 14 +++++----
 3 files changed, 36 insertions(+), 22 deletions(-)

diff --git a/gcc/config/rs6000/mma.md b/gcc/config/rs6000/mma.md
index d9757160845..f39a5c8ba99 100644
--- a/gcc/config/rs6000/mma.md
+++ b/gcc/config/rs6000/mma.md
@@ -99,6 +99,10 @@
    UNSPEC_MMA_XXMTACC
   ])
 
+(define_c_enum "unspecv"
+  [UNSPECV_MMA_XXSETACCZ
+  ])
+
 ;; MMA instructions with 1 accumulator argument
 (define_int_iterator MMA_ACC		[UNSPEC_MMA_XXMFACC
 					 UNSPEC_MMA_XXMTACC])
@@ -325,26 +329,24 @@
 })
 
 (define_insn_and_split "*movpxi"
-  [(set (match_operand:PXI 0 "nonimmediate_operand" "=d,m,d,d")
-	(match_operand:PXI 1 "input_operand" "m,d,d,O"))]
+  [(set (match_operand:PXI 0 "nonimmediate_operand" "=d,m,d")
+	(match_operand:PXI 1 "input_operand" "m,d,d"))]
   "TARGET_MMA
    && (gpc_reg_operand (operands[0], PXImode)
        || gpc_reg_operand (operands[1], PXImode))"
   "@
    #
    #
-   #
-   xxsetaccz %A0"
-  "&& reload_completed
-   && !(fpr_reg_operand (operands[0], PXImode) && operands[1] == const0_rtx)"
+   #"
+  "&& reload_completed"
   [(const_int 0)]
 {
   rs6000_split_multireg_move (operands[0], operands[1]);
   DONE;
 }
-  [(set_attr "type" "vecload,vecstore,veclogical,mma")
-   (set_attr "length" "8,8,16,*")
-   (set_attr "max_prefixed_insns" "2,2,*,*")])
+  [(set_attr "type" "vecload,vecstore,veclogical")
+   (set_attr "length" "8,8,16")
+   (set_attr "max_prefixed_insns" "2,2,*")])
 
 (define_expand "vsx_assemble_pair"
   [(match_operand:POI 0 "vsx_register_operand")
@@ -413,14 +415,16 @@
   "<acc> %A0"
   [(set_attr "type" "mma")])
 
-(define_expand "mma_xxsetaccz"
-  [(set (match_operand:PXI 0 "fpr_reg_operand")
-	(const_int 0))]
+;; We can't have integer constants in PXImode so we wrap this in an
+;; UNSPEC_VOLATILE.
+
+(define_insn "mma_xxsetaccz"
+  [(set (match_operand:PXI 0 "fpr_reg_operand" "=d")
+	(unspec_volatile:PXI [(const_int 0)]
+			     UNSPECV_MMA_XXSETACCZ))]
   "TARGET_MMA"
-{
-  emit_insn (gen_movpxi (operands[0], const0_rtx));
-  DONE;
-})
+  "xxsetaccz %A0"
+  [(set_attr "type" "mma")])
 
 (define_insn "mma_<vv>"
   [(set (match_operand:PXI 0 "fpr_reg_operand" "=&d")
diff --git a/gcc/config/rs6000/rs6000.c b/gcc/config/rs6000/rs6000.c
index 8ca7f2fe543..a9d1769350e 100644
--- a/gcc/config/rs6000/rs6000.c
+++ b/gcc/config/rs6000/rs6000.c
@@ -21353,6 +21353,14 @@ rs6000_rtx_costs (rtx x, machine_mode mode, int outer_code,
 	}
       break;
 
+    case UNSPEC_VOLATILE:
+      if (XINT (x, 1) == UNSPECV_MMA_XXSETACCZ)
+        {
+          *total = 0;
+          return true;
+        }
+      break;
+
     default:
       break;
     }
diff --git a/gcc/testsuite/gcc.target/powerpc/mma-builtin-6.c b/gcc/testsuite/gcc.target/powerpc/mma-builtin-6.c
index 0c6517211e3..715b28138e9 100644
--- a/gcc/testsuite/gcc.target/powerpc/mma-builtin-6.c
+++ b/gcc/testsuite/gcc.target/powerpc/mma-builtin-6.c
@@ -5,14 +5,16 @@
 void
 foo (__vector_quad *dst)
 {
-  __vector_quad acc;
-  __builtin_mma_xxsetaccz (&acc);
-  *dst = acc;
+  __vector_quad acc0, acc1;
+  __builtin_mma_xxsetaccz (&acc0);
+  __builtin_mma_xxsetaccz (&acc1);
+  dst[0] = acc0;
+  dst[1] = acc1;
 }
 
 /* { dg-final { scan-assembler-not {\mlxv\M} } } */
 /* { dg-final { scan-assembler-not {\mlxvp\M} } } */
 /* { dg-final { scan-assembler-not {\mxxmtacc\M} } } */
-/* { dg-final { scan-assembler-times {\mxxsetaccz\M} 1 } } */
-/* { dg-final { scan-assembler-times {\mxxmfacc\M} 1 } } */
-/* { dg-final { scan-assembler-times {\mstxvp\M} 2 } } */
+/* { dg-final { scan-assembler-times {\mxxsetaccz\M} 2 } } */
+/* { dg-final { scan-assembler-times {\mxxmfacc\M} 2 } } */
+/* { dg-final { scan-assembler-times {\mstxvp\M} 4 } } */


^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2022-01-15  6:00 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-01-15  6:00 [gcc r10-10397] rs6000: Disable optimizing multiple xxsetaccz instructions into one xxsetaccz Peter Bergner

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).