[Bug target/8603] [Alpha] s?addl pattern doesn't work
--- Comment #10 from uros at gcc dot gnu dot org 2009-08-14 07:41 --- Subject: Bug 8603 Author: uros Date: Fri Aug 14 07:41:17 2009 New Revision: 150735 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=150735 Log: Backport from mainline: 2009-08-11 Uros Bizjak ubiz...@gmail.com PR target/8603 * config/alpha/alpha.md (addsi3): Remove expander. (addsi3): Rename from *addsi3_internal insn pattern. (subsi3): Remove expander. (subsi3): Rename from *subsi3_internal insn pattern. Modified: branches/gcc-4_3-branch/gcc/ChangeLog branches/gcc-4_3-branch/gcc/config/alpha/alpha.md -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=8603
[Bug target/8603] [Alpha] s?addl pattern doesn't work
--- Comment #11 from ubizjak at gmail dot com 2009-08-14 08:11 --- Fixed. -- ubizjak at gmail dot com changed: What|Removed |Added Status|ASSIGNED|RESOLVED Known to fail|4.3.0 |4.3.4 4.4.1 Known to work|4.5.0 |4.3.5 4.4.2 4.5.0 Resolution||FIXED Target Milestone|--- |4.3.5 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=8603
[Bug target/8603] [Alpha] s?addl pattern doesn't work
--- Comment #9 from uros at gcc dot gnu dot org 2009-08-13 18:57 --- Subject: Bug 8603 Author: uros Date: Thu Aug 13 18:57:15 2009 New Revision: 150723 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=150723 Log: PR target/8603 * config/alpha/alpha.md (addsi3): Remove expander. (addsi3): Rename from *addsi3_internal insn pattern. (subsi3): Remove expander. (subsi3): Rename from *subsi3_internal insn pattern. Modified: branches/gcc-4_4-branch/gcc/ChangeLog branches/gcc-4_4-branch/gcc/config/alpha/alpha.md -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=8603
[Bug target/8603] [Alpha] s?addl pattern doesn't work
--- Comment #7 from ubizjak at gmail dot com 2009-08-11 06:46 --- (In reply to comment #6) Please add to gcc-4.3.x and gcc-4.4.x. OK, I will post the patch, thanks for the analysis. -- ubizjak at gmail dot com changed: What|Removed |Added AssignedTo|rask at gcc dot gnu dot org |ubizjak at gmail dot com http://gcc.gnu.org/bugzilla/show_bug.cgi?id=8603
[Bug target/8603] [Alpha] s?addl pattern doesn't work
--- Comment #8 from uros at gcc dot gnu dot org 2009-08-11 17:06 --- Subject: Bug 8603 Author: uros Date: Tue Aug 11 17:05:38 2009 New Revision: 150654 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=150654 Log: PR target/8603 * config/alpha/alpha.md (addsi3): Remove expander. (addsi3): Rename from *addsi3_internal insn pattern. (subsi3): Remove expander. (subsi3): Rename from *subsi3_internal insn pattern. Modified: trunk/gcc/ChangeLog trunk/gcc/config/alpha/alpha.md -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=8603
[Bug target/8603] [Alpha] s?addl pattern doesn't work
--- Comment #6 from mattst88 at gmail dot com 2009-08-11 02:38 ---
To show how worthwhile this trivial patch is -- the following table shows the
number of times s{4,8}{add,sub}l are used in building the Linux kernel
(2.6.31-rc5) with unpatched and patched gcc (4.3.4).
unpatched patched
s4addl 53 395
s8addl 79 132
s4subl 0 111
s8subl 0 35
This patch also causes gcc to produce exactly the same output as Compaq's C
compiler (this is a good thing!) for the two test cases given in the report.
For example --
Test case:
UP1500 gcc-tests # cat s_addl.c
int f(int x, int y) { return 4 * x + y; }
int g(int x) { return 3 * x; }
Results with unpatched gcc-4.3.x
UP1500 gcc-tests # gcc-unpatched -O3 -mcpu=ev67 -c s_addl.c
UP1500 gcc-tests # objdump -d s_addl.o
s_addl.o: file format elf64-alpha
Disassembly of section .text:
f:
0: 40 04 11 42 s4addq a0,a1,v0
4: 00 00 e0 43 sextl v0,v0 -- unnecessary
8: 01 80 fa 6b ret
c: 00 00 fe 2f unop
0010 g:
10: 60 05 10 42 s4subq a0,a0,v0
14: 00 00 e0 43 sextl v0,v0 -- unnecessary
18: 01 80 fa 6b ret
1c: 00 00 fe 2f unop
Results with patched gcc-4.3.x
UP1500 gcc-tests # gcc-patched -O3 -mcpu=ev67 -c s_addl.c
UP1500 gcc-tests # objdump -d s_addl.o
s_addl.o: file format elf64-alpha
Disassembly of section .text:
f:
0: 40 00 11 42 s4addl a0,a1,v0
4: 01 80 fa 6b ret
8: 1f 04 ff 47 nop
c: 00 00 fe 2f unop
0010 g:
10: 60 01 10 42 s4subl a0,a0,v0
14: 01 80 fa 6b ret
18: 1f 04 ff 47 nop
1c: 00 00 fe 2f unop
Results with Compaq C compiler (what we're trying to replicate)
UP1500 gcc-tests # ccc -fast -host -c s_addl.c
UP1500 gcc-tests # objdump -d s_addl.o
s_addl.o: file format elf64-alpha
Disassembly of section .text:
f:
0: 40 00 11 42 s4addl a0,a1,v0
4: 01 80 fa 6b ret
8: 00 00 fe 2f unop
c: 00 00 fe 2f unop
0010 g:
10: 60 01 10 42 s4subl a0,a0,v0
14: 01 80 fa 6b ret
Please add to gcc-4.3.x and gcc-4.4.x.
--
mattst88 at gmail dot com changed:
What|Removed |Added
CC||mattst88 at gmail dot com
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=8603
[Bug target/8603] [Alpha] s?addl pattern doesn't work
--- Comment #5 from ubizjak at gmail dot com 2009-01-07 17:56 --- Following patch that changes addsi3 and subsi3 expander constraint fixes this problem: --cut here-- Index: alpha.md === --- alpha.md(revision 143157) +++ alpha.md(working copy) @@ -261,7 +261,7 @@ [(set (match_operand:SI 0 register_operand ) (plus:SI (match_operand:SI 1 reg_or_0_operand ) (match_operand:SI 2 add_operand )))] - ! optimize + ) (define_insn *addsi_internal @@ -622,7 +622,7 @@ [(set (match_operand:SI 0 register_operand ) (minus:SI (match_operand:SI 1 reg_or_0_operand ) (match_operand:SI 2 reg_or_8bit_operand )))] - ! optimize + ) (define_insn *subsi_internal --cut here-- With the patch (gcc -O2): f: s4addl $16,$17,$0 ret $31,($26),1 g: s4subl $16,$16,$0 ret $31,($26),1 This regression was introduced by: Sat Feb 23 08:42:47 2002 Richard Kenner ken...@vlsi1.ultra.nyu.edu * expr.c (store_expr): When converting expression to promoted equivalent type, allow using SUBREG_REG of TARGET as the target of the expansion of EXP. * loop.c (basic_induction_var, case SUBREG): Always look inside. * config/alpha/alpha.c (rtx_equiv_function_matters): Delete decl. (alpha_emit_set_const): Handle SImode when can't make new pseudos. (alpha_emit_set_const_1, alpha_sa_mask): Use no_new_pseudos. * config/alpha/alpha.md (addsi3, subsi3): Don't use if optimizing. And the comment above addsi3 says: ;; Don't say we have addsi3 if optimizing. This generates better code. We ;; have the anonymous addsi3 pattern below in case combine wants to make it. So, is this still true? Can somebody benchmark this patch? We can perhaps look at csibe numbers. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=8603
[Bug target/8603] [Alpha] s?addl pattern doesn't work
--- Comment #4 from rask at gcc dot gnu dot org 2007-11-14 19:10 --- For f(), combine wants a pattern to match (set (reg:DI 76) (sign_extend:DI (subreg:SI (plus:DI (subreg:DI (mult:SI (reg:SI 16 $16 [ x ]) (const_int 4 [0x4])) 0) (reg:DI 17 $17 [ y ])) 0))) but the closest one is (define_insn *saddl_se [(set (match_operand:DI 0 register_operand =r,r) (sign_extend:DI (plus:SI (mult:SI (match_operand:SI 1 reg_not_elim_operand r,r) (match_operand:SI 2 const48_operand I,I)) (match_operand:SI 3 sext_add_operand rI,O] and similarily for g() where the *ssubl_se pattern doesn't match. I wonder where the (subreg:DI (mult:SI ...)) part comes from. That can't be right. -- rask at gcc dot gnu dot org changed: What|Removed |Added AssignedTo|unassigned at gcc dot gnu |rask at gcc dot gnu dot org |dot org | Status|NEW |ASSIGNED Known to fail||4.3.0 Last reconfirmed|2005-09-07 17:37:27 |2007-11-14 19:10:09 date|| http://gcc.gnu.org/bugzilla/show_bug.cgi?id=8603
