https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83565
--- Comment #21 from Sergei Trofimovich <slyfox at inbox dot ru> ---
(In reply to Jim Wilson from comment #20)
> If nonzero_bits1 isn't changed, then we may need to remove the
> WORD_REGISTER_OPERATIONS definition in the ia64 port. Unfortunately, I
> don't have access to ia64 hardware, so I can't easily check to see what will
> happen if this is done.
I've dropped WORD_REGISTER_OPERATIONS locally as:
--- gcc/config/ia64/ia64.h (revision 256001)
+++ gcc/config/ia64/ia64.h (working copy)
@@ -1538,11 +1538,6 @@
#define CASE_VECTOR_PC_RELATIVE 1
-/* Define this macro if operations between registers with integral mode
smaller
- than a word are always performed on the entire register. */
-
-#define WORD_REGISTER_OPERATIONS 1
-
and ran 'make check' on clean master compiler and master compiler with this
change.
1. The original test case works as expected after removal:
# gcc-ia64-no-word-register/host-ia64-unknown-linux-gnu/gcc/xgcc
-Bgcc-ia64-no-word-register/host-ia64-unknown-linux-gnu/gcc /a.c -o /a -O1 &&
/a
off>>: 7fffffff
# gcc-ia64-clean/host-ia64-unknown-linux-gnu/gcc/xgcc
-Bgcc-ia64-clean/host-ia64-unknown-linux-gnu/gcc /a.c -o /a -O1 && /a
off>>: ffffffff
2. Result of 'make check' tests:
clean:
=== gcc Summary ===
# of expected passes 104484
# of unexpected failures 524
# of unexpected successes 36
# of expected failures 528
# of unresolved testcases 2
# of unsupported tests 2093
Executing on host: /root/gcc-ia64-clean/host-ia64-unknown-linux-gnu/gcc/xgcc -v
(timeout = 300)
spawn /root/gcc-ia64-clean/host-ia64-unknown-linux-gnu/gcc/xgcc -v
no-word-register:
=== gcc Summary ===
# of expected passes 104484
# of unexpected failures 524
# of unexpected successes 36
# of expected failures 528
# of unresolved testcases 2
# of unsupported tests 2093
Executing on host:
/root/gcc-ia64-no-word-register/host-ia64-unknown-linux-gnu/gcc/xgcc -v
(timeout = 300)
spawn /root/gcc-ia64-no-word-register/host-ia64-unknown-linux-gnu/gcc/xgcc -v
Or in diff form:
--- clean.log 2017-12-28 00:19:11.897405845 +0000
+++ nwr.log 2017-12-28 00:19:24.497403922 +0000
@@ -10,2 +10,2 @@
-Executing on host: /root/gcc-ia64-clean/host-ia64-unknown-linux-gnu/gcc/xgcc
-v (timeout = 300)
-spawn /root/gcc-ia64-clean/host-ia64-unknown-linux-gnu/gcc/xgcc -v
+Executing on host:
/root/gcc-ia64-no-word-register/host-ia64-unknown-linux-gnu/gcc/xgcc -v
(timeout = 300)
+spawn /root/gcc-ia64-no-word-register/host-ia64-unknown-linux-gnu/gcc/xgcc -v
@@ -13 +13 @@
-COLLECT_GCC=/root/gcc-ia64-clean/host-ia64-unknown-linux-gnu/gcc/xgcc
+COLLECT_GCC=/root/gcc-ia64-no-word-register/host-ia64-unknown-linux-gnu/gcc/xgcc
@@ -18 +18 @@
-/root/gcc-ia64-clean/host-ia64-unknown-linux-gnu/gcc/xgcc version 8.0.0
20171226 (experimental) (GCC)
+/root/gcc-ia64-no-word-register/host-ia64-unknown-linux-gnu/gcc/xgcc version
8.0.0 20171226 (experimental) (GCC)
@@ -20 +20 @@
-runtest completed at Wed Dec 27 17:05:35 2017
+runtest completed at Wed Dec 27 13:14:17 2017
Surprisingly (or maybe not so surprisingly) no effect!
Assembly diff is as expected:
--- /a-clean.s 2017-12-28 00:22:22.945376700 +0000
+++ /a-nwr.s 2017-12-28 00:22:56.665371555 +0000
@@ -23,2 +23,2 @@
- dep.z r14 = r15, 8, 24
- ld4.acq r8 = [r12]
+ dep.z r8 = r15, 8, 24
+ ld4.acq r16 = [r12]
@@ -27 +27 @@
- mix4.r r14 = r14, r14
+ mix4.r r8 = r8, r8
@@ -29 +29 @@
- shr.u r14 = r14, 31
+ shr.u r8 = r8, 31
@@ -31 +31 @@
- xor r8 = r8, r14
+ xor r8 = r16, r8
@@ -33 +33 @@
- extr.u r8 = r8, 1, 32
+ extr.u r8 = r8, 1, 31
