b56d22c2f6
Handle SETO* like we handle SET*, restoring behavior after Evan's setcc
...
change. This fixes PowerPC/fnegsel.ll.
llvm-svn: 28443
2006-05-24 00:06:44 +00:00
eb755fc1b3
Make PPC call lowering more aggressive, making the isel matching code simple
...
enough to be autogenerated.
llvm-svn: 28354
2006-05-17 19:00:46 +00:00
b1e9e37c58
Switch PPC over to a call-selection model where the lowering code creates
...
the copyto/fromregs instead of making the PPCISD::CALL selection code create
them. This vastly simplifies the selection code, and moves the ABI handling
parts into one place.
llvm-svn: 28346
2006-05-17 06:01:33 +00:00
b7552a88d6
3 changes, 2 of which are cleanup one of which changes codegen:
...
1. Rearrange code a bit so that the special case doesn't require indenting lots
of code.
2. Add comments describing PPC calling convention.
3. Only round up to 56-bytes of stack space for an outgoing call if the callee
is varargs. This saves a bit of stack space.
llvm-svn: 28342
2006-05-17 00:15:40 +00:00
f058f5aef1
implement passing/returning vector regs to calls, at least non-varargs calls.
...
llvm-svn: 28341
2006-05-16 23:54:25 +00:00
aa40ec1b32
Instead of implementing LowerCallTo directly, let the default impl produce an
...
ISD::CALL node, then custom lower that. This means that we only have to handle
LEGAL call operands/results, not every possible type. This allows us to
simplify the call code, shrinking it by about 1/3.
llvm-svn: 28339
2006-05-16 22:56:08 +00:00
26e2fcd8b1
Simplify the argument counting logic by only incrementing the index.
...
llvm-svn: 28335
2006-05-16 18:58:15 +00:00
76c47b50e7
Simplify the dead argument handling code.
...
llvm-svn: 28334
2006-05-16 18:54:32 +00:00
318f0d2122
Vector args passed in registers don't reserve stack space.
...
llvm-svn: 28333
2006-05-16 18:51:52 +00:00
4302e8fb67
Switch the PPC backend over to using FORMAL_ARGUMENTS for formal argument
...
handling. This makes the lower argument code significantly simpler (we
only need to handle legal argument types).
Incidentally, this also implements support for vector argument registers,
so long as they are not on the stack.
llvm-svn: 28331
2006-05-16 18:18:50 +00:00
d2ca9abf57
Fit in 80 cols
...
llvm-svn: 28311
2006-05-16 04:20:24 +00:00
ae48a894b1
Remove dead var, fix bad override.
...
llvm-svn: 28264
2006-05-12 21:09:57 +00:00
84b49d51be
Fix CodeGen/Generic/2006-04-28-Sign-extend-bool.ll
...
llvm-svn: 28017
2006-04-28 21:56:10 +00:00
4ca2ea5b43
JumpTable support! What this represents is working asm and jit support for
...
x86 and ppc for 100% dense switch statements when relocations are non-PIC.
This support will be extended and enhanced in the coming days to support
PIC, and less dense forms of jump tables.
llvm-svn: 27947
2006-04-22 18:53:45 +00:00
518834c67e
Fix a crash on:
...
void foo2(vector float *A, vector float *B) {
vector float C = (vector float)vec_cmpeq(*A, *B);
if (!vec_any_eq(*A, *B))
*B = (vector float){0,0,0,0};
*A = C;
}
llvm-svn: 27808
2006-04-18 18:28:22 +00:00
1e174c87c3
pretty print node name
...
llvm-svn: 27806
2006-04-18 18:05:58 +00:00
9754d142a4
Implement an important entry from README_ALTIVEC:
...
If an altivec predicate compare is used immediately by a branch, don't
use a (serializing) MFCR instruction to read the CR6 register, which requires
a compare to get it back to CR's. Instead, just branch on CR6 directly. :)
For example, for:
void foo2(vector float *A, vector float *B) {
if (!vec_any_eq(*A, *B))
*B = (vector float){0,0,0,0};
}
We now generate:
_foo2:
mfspr r2, 256
oris r5, r2, 12288
mtspr 256, r5
lvx v2, 0, r4
lvx v3, 0, r3
vcmpeqfp. v2, v3, v2
bne cr6, LBB1_2 ; UnifiedReturnBlock
LBB1_1: ; cond_true
vxor v2, v2, v2
stvx v2, 0, r4
mtspr 256, r2
blr
LBB1_2: ; UnifiedReturnBlock
mtspr 256, r2
blr
instead of:
_foo2:
mfspr r2, 256
oris r5, r2, 12288
mtspr 256, r5
lvx v2, 0, r4
lvx v3, 0, r3
vcmpeqfp. v2, v3, v2
mfcr r3, 2
rlwinm r3, r3, 27, 31, 31
cmpwi cr0, r3, 0
beq cr0, LBB1_2 ; UnifiedReturnBlock
LBB1_1: ; cond_true
vxor v2, v2, v2
stvx v2, 0, r4
mtspr 256, r2
blr
LBB1_2: ; UnifiedReturnBlock
mtspr 256, r2
blr
This implements CodeGen/PowerPC/vec_br_cmp.ll.
llvm-svn: 27804
2006-04-18 17:59:36 +00:00
96d50487c9
Use vmladduhm to do v8i16 multiplies which is faster and simpler than doing
...
even/odd halves. Thanks to Nate telling me what's what.
llvm-svn: 27793
2006-04-18 04:28:57 +00:00
d6d82aa889
Implement v16i8 multiply with this code:
...
vmuloub v5, v3, v2
vmuleub v2, v3, v2
vperm v2, v2, v5, v4
This implements CodeGen/PowerPC/vec_mul.ll. With this, v16i8 multiplies are
6.79x faster than before.
Overall, UnitTests/Vector/multiplies.c is now 2.45x faster with LLVM than with
GCC.
Remove the 'integer multiplies' todo from the README file.
llvm-svn: 27792
2006-04-18 03:57:35 +00:00
7e439874cb
Lower v8i16 multiply into this code:
...
li r5, lo16(LCPI1_0)
lis r6, ha16(LCPI1_0)
lvx v4, r6, r5
vmulouh v5, v3, v2
vmuleuh v2, v3, v2
vperm v2, v2, v5, v4
where v4 is:
LCPI1_0: ; <16 x ubyte>
.byte 2
.byte 3
.byte 18
.byte 19
.byte 6
.byte 7
.byte 22
.byte 23
.byte 10
.byte 11
.byte 26
.byte 27
.byte 14
.byte 15
.byte 30
.byte 31
This is 5.07x faster on the G5 (measured) than lowering to scalar code +
loads/stores.
llvm-svn: 27789
2006-04-18 03:43:48 +00:00
a2cae1bb10
Custom lower v4i32 multiplies into a cute sequence, instead of having legalize
...
scalarize the sequence into 4 mullw's and a bunch of load/store traffic.
This speeds up v4i32 multiplies 4.1x (measured) on a G5. This implements
PowerPC/vec_mul.ll
llvm-svn: 27788
2006-04-18 03:24:30 +00:00
e54133cfba
Make sure to check splats of every constant we can, handle splat(31) by
...
being a bit more clever, add support for odd splats from -31 to -17.
llvm-svn: 27764
2006-04-17 18:09:22 +00:00
264c908e3a
Teach the ppc backend to use rol and vsldoi to generate splatted constants.
...
This implements vec_constants.ll:test_vsldoi and test_rol
llvm-svn: 27760
2006-04-17 17:55:10 +00:00
1b3806ace5
Make some code more general, adding support for constant formation of several
...
new patterns.
llvm-svn: 27754
2006-04-17 06:58:41 +00:00
f8dd76df5b
Learn how to make odd splatted constants in range [17,29]. This implements
...
PowerPC/vec_constants.ll:test_29.
llvm-svn: 27752
2006-04-17 06:07:44 +00:00
2a099c04c1
Pull some code out into a helper function.
...
Effeciently codegen even splats in the range [-32,30].
This allows us to codegen <30,30,30,30> as:
vspltisw v0, 15
vadduwm v2, v0, v0
instead of as a cp load.
llvm-svn: 27750
2006-04-17 06:00:21 +00:00
071ad01ceb
Implement a TODO: for any shuffle that can be viewed as a v4[if]32 shuffle,
...
if it can be implemented in 3 or fewer discrete altivec instructions, codegen
it as such. This implements Regression/CodeGen/PowerPC/vec_perf_shuffle.ll
llvm-svn: 27748
2006-04-17 05:28:54 +00:00
06a21ba96b
Implement a TODO: have the legalizer canonicalize a bunch of operations to
...
one type (v4i32) so that we don't have to write patterns for each type, and
so that more CSE opportunities are exposed.
llvm-svn: 27731
2006-04-16 01:37:57 +00:00
fa5aa396c2
Make the BUILD_VECTOR lowering code much more aggressive w.r.t constant vectors.
...
Remove some done items from the todo list.
llvm-svn: 27729
2006-04-16 01:01:29 +00:00
24acbe46c0
Fix a crash when faced with a shuffle vector that has an undef in its mask.
...
llvm-svn: 27726
2006-04-15 23:48:05 +00:00
559c8ba466
Allow undef in a shuffle mask
...
llvm-svn: 27714
2006-04-14 23:19:08 +00:00
4211ca9108
Move the rest of the PPCTargetLowering::LowerOperation cases out into
...
separate functions, for simplicity and code clarity.
llvm-svn: 27693
2006-04-14 06:01:58 +00:00
19e9055eb5
Pull the VECTOR_SHUFFLE and BUILD_VECTOR lowering code out into separate
...
functions, which makes the code much cleaner :)
llvm-svn: 27692
2006-04-14 05:19:18 +00:00
883fb053bd
Force non-darwin targets to use a static relo model. This fixes PR734,
...
tested by CodeGen/Generic/vector.ll
llvm-svn: 27657
2006-04-13 17:10:48 +00:00
147e50e1c5
Add a new way to match vector constants, which make it easier to bang bits of
...
different types.
Codegen spltw(0x7FFFFFFF) and spltw(0x80000000) without a constant pool load,
implementing PowerPC/vec_constants.ll:test1. This compiles:
typedef float vf __attribute__ ((vector_size (16)));
typedef int vi __attribute__ ((vector_size (16)));
void test(vi *P1, vi *P2, vf *P3) {
*P1 &= (vi){0x80000000,0x80000000,0x80000000,0x80000000};
*P2 &= (vi){0x7FFFFFFF,0x7FFFFFFF,0x7FFFFFFF,0x7FFFFFFF};
*P3 = vec_abs((vector float)*P3);
}
to:
_test:
mfspr r2, 256
oris r6, r2, 49152
mtspr 256, r6
vspltisw v0, -1
vslw v0, v0, v0
lvx v1, 0, r3
vand v1, v1, v0
stvx v1, 0, r3
lvx v1, 0, r4
vandc v1, v1, v0
stvx v1, 0, r4
lvx v1, 0, r5
vandc v0, v1, v0
stvx v0, 0, r5
mtspr 256, r2
blr
instead of (with two constant pool entries):
_test:
mfspr r2, 256
oris r6, r2, 49152
mtspr 256, r6
li r6, lo16(LCPI1_0)
lis r7, ha16(LCPI1_0)
li r8, lo16(LCPI1_1)
lis r9, ha16(LCPI1_1)
lvx v0, r7, r6
lvx v1, 0, r3
vand v0, v1, v0
stvx v0, 0, r3
lvx v0, r9, r8
lvx v1, 0, r4
vand v1, v1, v0
stvx v1, 0, r4
lvx v1, 0, r5
vand v0, v1, v0
stvx v0, 0, r5
mtspr 256, r2
blr
GCC produces (with 2 cp entries):
_test:
mfspr r0,256
stw r0,-4(r1)
oris r0,r0,0xc00c
mtspr 256,r0
lis r2,ha16(LC0)
lis r9,ha16(LC1)
la r2,lo16(LC0)(r2)
lvx v0,0,r3
lvx v1,0,r5
la r9,lo16(LC1)(r9)
lwz r12,-4(r1)
lvx v12,0,r2
lvx v13,0,r9
vand v0,v0,v12
stvx v0,0,r3
vspltisw v0,-1
vslw v12,v0,v0
vandc v1,v1,v12
stvx v1,0,r5
lvx v0,0,r4
vand v0,v0,v13
stvx v0,0,r4
mtspr 256,r12
blr
llvm-svn: 27624
2006-04-12 19:07:14 +00:00
74cf9ff761
Rename get_VSPLI_elt -> get_VSPLTI_elt
...
Canonicalize BUILD_VECTOR's that match VSPLTI's into a single type for each
form, eliminating a bunch of Pat patterns in the .td file and allowing us to
CSE stuff more aggressively. This implements
PowerPC/buildvec_canonicalize.ll:VSPLTI
llvm-svn: 27614
2006-04-12 17:37:20 +00:00
e318a7574e
Ensure that zero vectors are always v4i32, which forces them to CSE with
...
each other. This implements CodeGen/PowerPC/vxor-canonicalize.ll
llvm-svn: 27609
2006-04-12 16:53:28 +00:00
e4db08a2f1
Vector function results go into V2 according to GCC. The darwin ABI doc
...
doesn't say where they go :-/
llvm-svn: 27579
2006-04-11 01:38:39 +00:00
92533cfb4a
Move some return-handling code from lowerarguments to the ISD::RET handling stuff.
...
No functionality change.
llvm-svn: 27577
2006-04-11 01:21:43 +00:00
3a68f3c3ca
properly mark vector selects as expanded to select_cc
...
llvm-svn: 27544
2006-04-08 22:59:15 +00:00
0a3d1bbca4
Add VRRC select support
...
llvm-svn: 27543
2006-04-08 22:45:08 +00:00
d9e80f4516
Implement PowerPC/CodeGen/vec_splat.ll:spltish to use vsplish instead of a
...
constant pool load.
llvm-svn: 27538
2006-04-08 07:14:26 +00:00
d71a1f946d
Change the interface to the predicate that determines if vsplti* can be used.
...
No functionality changes.
llvm-svn: 27536
2006-04-08 06:46:53 +00:00
466841ddc7
Make sure to return the result in the right type.
...
llvm-svn: 27469
2006-04-06 23:12:19 +00:00
a4bbfaed5c
Match vpku[hw]um(x,x).
...
Convert vsldoi(x,x) to work the same way other (x,x) cases work.
llvm-svn: 27467
2006-04-06 22:28:36 +00:00
f38e033270
Add support for matching vmrg(x,x) patterns
...
llvm-svn: 27463
2006-04-06 22:02:42 +00:00
d1dcb52093
Pattern match vmrg* instructions, which are now lowered by the CFE into shuffles.
...
llvm-svn: 27457
2006-04-06 21:11:54 +00:00
1d33819194
Support pattern matching vsldoi(x,y) and vsldoi(x,x), which allows the f.e. to
...
lower it and LLVM to have one fewer intrinsic. This implements
CodeGen/PowerPC/vec_shuffle.ll
llvm-svn: 27450
2006-04-06 18:26:28 +00:00
e8b83b4206
Compile the vpkuhum/vpkuwum intrinsics into vpkuhum/vpkuwum instead of into
...
vperm with a perm mask lvx'd from the constant pool.
llvm-svn: 27448
2006-04-06 17:23:16 +00:00
39cc717c65
Fix CodeGen/PowerPC/2006-04-05-splat-ish.ll
...
llvm-svn: 27439
2006-04-05 17:39:25 +00:00
Chris Lattner