Commit Graph

12821 Commits

Author SHA1 Message Date
Jeff Cohen 55f63f1b53 Fix VC++ warning.
llvm-svn: 26228
2006-02-16 04:07:37 +00:00
Evan Cheng ae82498e81 Use movaps / movapd (instead of movss / movsd) to do FR32 / FR64 reg to reg
transfer.

According to the Intel P4 Optimization Manual:

Moves that write a portion of a register can introduce unwanted
dependences. The movsd reg, reg instruction writes only the bottom
64 bits of a register, not to all 128 bits. This introduces a dependence on
the preceding instruction that produces the upper 64 bits (even if those
bits are not longer wanted). The dependence inhibits register renaming,
and thereby reduces parallelism.

Not to mention movaps is shorter than movss.

llvm-svn: 26226
2006-02-16 01:50:02 +00:00
Chris Lattner ff42e81028 fix a bug where we unswitched the wrong way
llvm-svn: 26225
2006-02-16 01:24:41 +00:00
Evan Cheng 03c1e6f48e A bit more memset / memcpy optimization.
Turns them into calls to memset / memcpy if 1) buffer(s) are not DWORD aligned,
2) size is not known to be greater or equal to some minimum value (currently 128).

llvm-svn: 26224
2006-02-16 00:21:07 +00:00
Chris Lattner fdff0bb43e Implement trivial unswitching for switch stmts. This allows us to trivial
unswitch this loop on 2 before sweating to unswitch on 1/3.

void test4(int N, int i, int C, int*P, int*Q) {
  int j;
  for (j = 0; j < N; ++j) {
    switch (C) {                // general unswitching.
    default: P[i+j] = 0; break;
    case 1: Q[i+j] = 0; break;
    case 3: P[i+j] = Q[i+j]; break;
    case 2: break;              //  TRIVIAL UNSWITCH on C==2
    }
  }
}

llvm-svn: 26223
2006-02-15 22:52:05 +00:00
Evan Cheng 76a7775ce1 Remove an entry.
llvm-svn: 26222
2006-02-15 22:14:34 +00:00
Evan Cheng 93e4865d4b Remove an unused function parameter.
llvm-svn: 26221
2006-02-15 22:12:35 +00:00
Chris Lattner e5cb76d744 make "trivial" unswitching significantly more general. It can now handle
this for example:

  for (j = 0; j < N; ++j) {     // trivial unswitch
    if (C)
      P[i+j] = 0;
  }

turning it into the obvious code without bothering to duplicate an empty loop.

llvm-svn: 26220
2006-02-15 22:03:36 +00:00
Evan Cheng 6781b6e62e Turn a memcpy from string constant into a series of stores of constant values.
llvm-svn: 26219
2006-02-15 21:59:04 +00:00
Andrew Lenharth 47da60130a fix a bunch of alpha regressions. see bug 709
llvm-svn: 26218
2006-02-15 21:13:37 +00:00
Chris Lattner 6afb5587da new test
llvm-svn: 26217
2006-02-15 19:52:06 +00:00
Jim Laskey 2eea436192 Should not combine ISD::LOCATIONs until we have scheme to remove from
MachineDebugInfo tables.

llvm-svn: 26216
2006-02-15 19:34:44 +00:00
Chris Lattner 65152d80ec Checking the wrong value. This caused us to emit silly code like
Y = seteq bool X, true
instead of just using X :)

llvm-svn: 26215
2006-02-15 19:05:52 +00:00
Chris Lattner f20e61f003 Convert this over to work with the new makefiles
llvm-svn: 26206
2006-02-15 07:22:58 +00:00
Chris Lattner 6db414e8de Sparc actually *DOES* have a directive for emitting zeros. In fact, it requires
it, because this:

.bss
X:
.byte 0

results in the assembler warning: "initialization in bss segment".  Annoying.

llvm-svn: 26204
2006-02-15 07:07:14 +00:00
Chris Lattner 7edc87279c random lexer change to test the makefile updating stuff
llvm-svn: 26203
2006-02-15 07:02:59 +00:00
Chris Lattner a9d0b5800a Fix SingleSource/Regression/C/2004-08-12-InlinerAndAllocas.c on Sparc.
The ABI specifies that there is a register save area at the bottom of the
stack, which means the actual used pointer needs to be an offset from
the subtracted value.

llvm-svn: 26202
2006-02-15 06:41:34 +00:00
Evan Cheng 7a6c21ac26 Remove an entry.
llvm-svn: 26197
2006-02-15 01:56:48 +00:00
Evan Cheng 2d23c9f1ab Use .zerofill on x86/darwin.
llvm-svn: 26196
2006-02-15 01:56:23 +00:00
Evan Cheng e2038bdeee Lower memcpy with small constant size operand into a series of load / store
ops.

llvm-svn: 26195
2006-02-15 01:54:51 +00:00
Chris Lattner 01db04efb0 more refactoring, no functionality change.
llvm-svn: 26194
2006-02-15 01:44:42 +00:00
Evan Cheng aacc4c3b4c cvtsd2ss / cvtss2sd encoding bug.
llvm-svn: 26193
2006-02-15 00:31:03 +00:00
Evan Cheng 665c26ab40 movaps, movapd encoding bug.
llvm-svn: 26192
2006-02-15 00:11:37 +00:00
Chris Lattner b0cbe7106e pull some code out into a function
llvm-svn: 26191
2006-02-15 00:07:43 +00:00
Chris Lattner 9c5693fb2a Canonicalize inner loops before outer loops. Inner loop canonicalization
can provide work for the outer loop to canonicalize.

This fixes a case that breaks unswitching.

llvm-svn: 26189
2006-02-14 23:06:02 +00:00
Evan Cheng 0451499b3c Doh again!
llvm-svn: 26188
2006-02-14 23:05:54 +00:00
Chris Lattner cffbbee8d1 When splitting exit edges to canonicalize loops, make sure to put the new
block in the appropriate loop nest.

Third time is the charm, right?

llvm-svn: 26187
2006-02-14 22:34:08 +00:00
Chris Lattner e3c793a71a new note
llvm-svn: 26186
2006-02-14 22:19:54 +00:00
Chris Lattner b134520b86 If we have zero initialized data with external linkage, use .zerofill to
emit it (instead of .space), saving a bit of space in the .o file.

For example:
int foo[100];
int bar[100] = {};

when compiled with C++ or -fno-common results in shrinkage from 1160 to 360
bytes of space.  The X86 backend can also do this on darwin.

llvm-svn: 26185
2006-02-14 22:18:23 +00:00
Jim Laskey ebb50a61d2 Using wrong DW_FORM.
llvm-svn: 26184
2006-02-14 22:01:57 +00:00
Evan Cheng f84774ed46 Don't special case XS, XD prefixes.
llvm-svn: 26183
2006-02-14 21:52:51 +00:00
Evan Cheng fb7b5ef74b Bug fix: XS, XD prefixes were being emitted twice.
XMM registers were not being handled.

llvm-svn: 26182
2006-02-14 21:45:24 +00:00
Chris Lattner 84fb09eba4 Make sure that weak functions are aligned properly
llvm-svn: 26181
2006-02-14 20:42:33 +00:00
Evan Cheng 43b72f4421 Duh
llvm-svn: 26180
2006-02-14 20:37:37 +00:00
Evan Cheng ad8c20cd2b Remove -disable-x86-sse
llvm-svn: 26179
2006-02-14 20:30:14 +00:00
Chris Lattner 94f936d8ef add an assert
llvm-svn: 26178
2006-02-14 20:14:17 +00:00
Evan Cheng db2a7a736a Keep to < 80 cols
llvm-svn: 26177
2006-02-14 20:12:38 +00:00
Evan Cheng 038521ef76 Missed a break so memcpy cases fell through to memset. Doh.
llvm-svn: 26176
2006-02-14 19:45:56 +00:00
Evan Cheng d502610604 Fixed a build breakage.
llvm-svn: 26175
2006-02-14 09:11:59 +00:00
Evan Cheng 4b40a42653 Rename maxStoresPerMemSet to maxStoresPerMemset, etc.
llvm-svn: 26174
2006-02-14 08:38:30 +00:00
Evan Cheng f976d79f78 Add a entry.
llvm-svn: 26173
2006-02-14 08:25:32 +00:00
Evan Cheng 6a37456d73 Set maxStoresPerMemSet to 16. Ditto for maxStoresPerMemCpy and
maxStoresPerMemMove. Although the last one is not used.

llvm-svn: 26172
2006-02-14 08:25:08 +00:00
Evan Cheng 81fcea8aa2 Expand memset dst, c, size to a series of stores if size falls below the
target specific theshold, e.g. 16 for x86.

llvm-svn: 26171
2006-02-14 08:22:34 +00:00
Evan Cheng 40b6eb9973 Enable SSE (for the right subtargets)
llvm-svn: 26169
2006-02-14 08:07:58 +00:00
Chris Lattner 1784a9d267 now that libcalls don't suck, we can remove this hack
llvm-svn: 26164
2006-02-14 05:39:35 +00:00
Chris Lattner 0242688ad8 adjust to new style of handling lexer changes
llvm-svn: 26162
2006-02-14 05:14:46 +00:00
Chris Lattner 0b8ec1a132 Use statistics to keep track of what flavors of loops we are unswitching
llvm-svn: 26157
2006-02-14 01:01:41 +00:00
Chris Lattner 8e2ee7358f Fix a latent bug in the call sequence handling stuff. Some targets (e.g. x86)
create these nodes with flag results.  Remember that we legalized them.

llvm-svn: 26156
2006-02-14 00:55:02 +00:00
Chris Lattner 8b10ab3002 Implement Instcombine/and.ll:test34
llvm-svn: 26155
2006-02-13 23:07:23 +00:00
Chris Lattner 7d8522884b If any of the sign extended bits are demanded, the input sign bit is demanded
for a sign extension.

This fixes InstCombine/2006-02-13-DemandedMiscompile.ll and Ptrdist/bc.

llvm-svn: 26152
2006-02-13 22:41:07 +00:00