The scalarization overhead was set deliberately high for MVE, whilst the
codegen was new. It helps protect us against the negative ramifications
of mixing scalar and vector instructions. This decreases that,
especially for floating point where the cost of extracting/inserting
lane elements can be low. For integer the cost is still fairly high due
to the cross-register-bank copy, but is no longer n^2 in the length of
the vector.
In general, this will decrease the cost of scalarizing floats and long
integer vectors. i64 increase in cost, having a high cost before and
after this patch. For floats this allows up to start doing things like
vectorizing fdiv instructions, even if they are scalarized.
Differential Revision: https://reviews.llvm.org/D98245
This adds the CostKind to getMVEVectorCostFactor, so that it can
automatically account for CodeSize costs, where it returns a cost of 1
not the MVEFactor used for Throughput/Latency. This helps simplify the
caller code and allows us to get the codesize cost more correct in more
cases.
We did not have specific costs for larger than legal truncates that were
not otherwise cheap (where they were next to stores, for example). As
MVE does not have a dedicated instruction for them (and we do not use
loads/stores yet), they should be expensive as they get expanded to a
series of lane moves.
Differential Revision: https://reviews.llvm.org/D94260
As with other targets, set the throughput cost of control-flow
instructions to free so that we don't miss out of vectorization
opportunities.
Differential Revision: https://reviews.llvm.org/D85283
This adds some default costs for fp extends and truncates, generally
costing them as 1 per lane. If the type is not legal then the cost will
include a call to an __aeabi_ function.
Some NEON code is also adjusted to make sure it applies to the expected
types, now that fp16 is a more common thing.
Differential Revision: https://reviews.llvm.org/D82458
This file has grown quite large and could do with being split up. This
splits away the load/store + cast tests into a separate file. Some
masked load/store + cast tests have been added too, along with some
extra load/store + fpcast tests.
Have BasicTTI call the base implementation so that both agree on the
default behaviour, which the default being a cost of '1'. This has
required an X86 specific implementation as it seems to be very
reliant on those instructions being free. Changes are also made to
AMDGPU so that their implementations distinguish between cost kinds,
so that the unrolling isn't affected. PowerPC also has its own
implementation to prevent changes to the reg-usage vectorizer test.
The cost model test changes now reflect that ret instructions are not
generally free.
Differential Revision: https://reviews.llvm.org/D79164
We need to use it to handle <16 x double> indirect indexes
in the AMDGPU BE.
The only visible change from adding it is in ARM cost model.
To me it looks reasonable. With doubling a vector size it
quadruples the cost up to the size 8 and then it did only
double it. Now it also quadruples, which seems a logical
progression to me.
Actual AMDGPU code is to follow, this is a common part, plus
load/store legalization in the AMDGPU BE not to break what
works now.
Differential Revision: https://reviews.llvm.org/D79952
Fix the assumption that all bitcasts of the same type sizes are free.
We now only assume that bitcasts between ints and ptrs of the same
size are free. This allows TTImpl to just call the concrete
implementation of getCastInstrCost.
Differential Revision: https://reviews.llvm.org/D78918
- Specifically check for sext/zext users which have 'long' form NEON
instructions.
- Add more entries to the table for sext/zexts so that we can report
more accurately the number of vmovls required for NEON.
- Pass the instruction to the pass implementation.
Differential Revision: https://reviews.llvm.org/D79561
If both the source and the destination need to be split then the two halves of the split operation are completely independent and don't need to be split or joined. So we don't need to assess a cost for the split or join.
Differential Revision: https://reviews.llvm.org/D79111
Previously, we just always returned 1. But that ignores that we have to do the operation for each subvector or scalar component.
Differential Revision: https://reviews.llvm.org/D78824
This adds some sext costs for MVE, taken from the length of assembly sequences
that we currently generate.
Differential Revision: https://reviews.llvm.org/D66010
llvm-svn: 369244
MVE also has some sext of loads, which will be free just as scalar
instructions are.
Differential Revision: https://reviews.llvm.org/D66008
llvm-svn: 369118
The MVE architecture has the idea of "beats", where a vector instruction can be
executed over several ticks of the architecture. This adds a similar system
into the Arm backend cost model, multiplying the cost of all vector
instructions by a factor.
This factor essentially becomes the expected difference between scalar code
and vector code, on average. MVE Vector instructions can also overlap so the a
true cost of them is often lower. But equally scalar instructions can in some
situations be dual issued, or have other optimisations such as unrolling or
make use of dsp instructions. The default is chosen as 2. This should not
prevent vectorisation is a most cases (as the vector instructions will still be
doing at least 4 times the work), but it will help prevent over vectorising in
cases where the benefits are less likely.
This adds things so far to the obvious places in ARMTargetTransformInfo, and
updates a few related costs like not treating float instructions as cost 2 just
because they are floats.
Differential Revision: https://reviews.llvm.org/D66005
llvm-svn: 368733
This teaches the cost model that the sext or zext of a load is going to be
free.
Differential Revision: https://reviews.llvm.org/D66006
llvm-svn: 368593
This puts some of the calls in ARMTargetTransformInfo.cpp behind hasNeon()
checks, now that we have MVE, and updates all the tests accordingly.
Differential Revision: https://reviews.llvm.org/D63447
llvm-svn: 368587
This adds a number of cost model tests for ARM, useful for MVE. It also re-jigs
some of the existing tests to make them easier to update and read.
llvm-svn: 368586
The cost model should not assume vector casts get completely scalarized, since
on targets that have vector support, the common case is a partial split up to
the legal vector size. So, when a vector cast gets split, the resulting casts
end up legal and cheap.
Instead of pessimistically assuming scalarization, base TTI can use the costs
the concrete TTI provides for the split vector, plus a fudge factor to account
for the cost of the split itself. This fudge factor is currently 1 by default,
except on AMDGPU where inserts and extracts are considered free.
Differential Revision: http://reviews.llvm.org/D21251
llvm-svn: 274642
Summary:
This change limits the minimum cost of an insert/extract
element operation to 2 in cases where this would result
in mixing of NEON and VFP code.
Reviewers: rengolin
Subscribers: mssimpso, aemerson, llvm-commits, rengolin
Differential Revision: http://reviews.llvm.org/D12030
llvm-svn: 245225
Cross-class copies being expensive is actually a trait of the microarchitecture, but as I haven't yet seen an example of a microarchitecture where they're cheap it seems best to just enable this by default, covering the non-mcpu build case.
llvm-svn: 217674
the legalization cost must be included to get an accurate
estimation of the total cost of the scalarized vector.
The inaccurate cost triggered unprofitable SLP vectorization on
32-bit X86.
Summary:
Include legalization overhead when computing scalarization cost
Reviewers: hfinkel, nadav
CC: chandlerc, rnk, llvm-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D2992
llvm-svn: 203509
Rather than just splitting the input type and hoping for the best, apply
a bit more cleverness. Just splitting the types until the source is
legal often leads to an illegal result time, which is then widened and a
scalarization step is introduced which leads to truly horrible code
generation. With the loop vectorizer, these sorts of operations are much
more common, and so it's worth extra effort to do them well.
Add a legalization hook for the operands of a TRUNCATE node, which will
be encountered after the result type has been legalized, but if the
operand type is still illegal. If simple splitting of both types
ends up with the result type of each half still being legal, just
do that (v16i16 -> v16i8 on ARM, for example). If, however, that would
result in an illegal result type (v8i32 -> v8i8 on ARM, for example),
we can get more clever with power-two vectors. Specifically,
split the input type, but also widen the result element size, then
concatenate the halves and truncate again. For example on ARM,
To perform a "%res = v8i8 trunc v8i32 %in" we transform to:
%inlo = v4i32 extract_subvector %in, 0
%inhi = v4i32 extract_subvector %in, 4
%lo16 = v4i16 trunc v4i32 %inlo
%hi16 = v4i16 trunc v4i32 %inhi
%in16 = v8i16 concat_vectors v4i16 %lo16, v4i16 %hi16
%res = v8i8 trunc v8i16 %in16
This allows instruction selection to generate three VMOVN instructions
instead of a sequences of moves, stores and loads.
Update the ARMTargetTransformInfo to take this improved legalization
into account.
Consider the simplified IR:
define <16 x i8> @test1(<16 x i32>* %ap) {
%a = load <16 x i32>* %ap
%tmp = trunc <16 x i32> %a to <16 x i8>
ret <16 x i8> %tmp
}
define <8 x i8> @test2(<8 x i32>* %ap) {
%a = load <8 x i32>* %ap
%tmp = trunc <8 x i32> %a to <8 x i8>
ret <8 x i8> %tmp
}
Previously, we would generate the truly hideous:
.syntax unified
.section __TEXT,__text,regular,pure_instructions
.globl _test1
.align 2
_test1: @ @test1
@ BB#0:
push {r7}
mov r7, sp
sub sp, sp, #20
bic sp, sp, #7
add r1, r0, #48
add r2, r0, #32
vld1.64 {d24, d25}, [r0:128]
vld1.64 {d16, d17}, [r1:128]
vld1.64 {d18, d19}, [r2:128]
add r1, r0, #16
vmovn.i32 d22, q8
vld1.64 {d16, d17}, [r1:128]
vmovn.i32 d20, q9
vmovn.i32 d18, q12
vmov.u16 r0, d22[3]
strb r0, [sp, #15]
vmov.u16 r0, d22[2]
strb r0, [sp, #14]
vmov.u16 r0, d22[1]
strb r0, [sp, #13]
vmov.u16 r0, d22[0]
vmovn.i32 d16, q8
strb r0, [sp, #12]
vmov.u16 r0, d20[3]
strb r0, [sp, #11]
vmov.u16 r0, d20[2]
strb r0, [sp, #10]
vmov.u16 r0, d20[1]
strb r0, [sp, #9]
vmov.u16 r0, d20[0]
strb r0, [sp, #8]
vmov.u16 r0, d18[3]
strb r0, [sp, #3]
vmov.u16 r0, d18[2]
strb r0, [sp, #2]
vmov.u16 r0, d18[1]
strb r0, [sp, #1]
vmov.u16 r0, d18[0]
strb r0, [sp]
vmov.u16 r0, d16[3]
strb r0, [sp, #7]
vmov.u16 r0, d16[2]
strb r0, [sp, #6]
vmov.u16 r0, d16[1]
strb r0, [sp, #5]
vmov.u16 r0, d16[0]
strb r0, [sp, #4]
vldmia sp, {d16, d17}
vmov r0, r1, d16
vmov r2, r3, d17
mov sp, r7
pop {r7}
bx lr
.globl _test2
.align 2
_test2: @ @test2
@ BB#0:
push {r7}
mov r7, sp
sub sp, sp, #12
bic sp, sp, #7
vld1.64 {d16, d17}, [r0:128]
add r0, r0, #16
vld1.64 {d20, d21}, [r0:128]
vmovn.i32 d18, q8
vmov.u16 r0, d18[3]
vmovn.i32 d16, q10
strb r0, [sp, #3]
vmov.u16 r0, d18[2]
strb r0, [sp, #2]
vmov.u16 r0, d18[1]
strb r0, [sp, #1]
vmov.u16 r0, d18[0]
strb r0, [sp]
vmov.u16 r0, d16[3]
strb r0, [sp, #7]
vmov.u16 r0, d16[2]
strb r0, [sp, #6]
vmov.u16 r0, d16[1]
strb r0, [sp, #5]
vmov.u16 r0, d16[0]
strb r0, [sp, #4]
ldm sp, {r0, r1}
mov sp, r7
pop {r7}
bx lr
Now, however, we generate the much more straightforward:
.syntax unified
.section __TEXT,__text,regular,pure_instructions
.globl _test1
.align 2
_test1: @ @test1
@ BB#0:
add r1, r0, #48
add r2, r0, #32
vld1.64 {d20, d21}, [r0:128]
vld1.64 {d16, d17}, [r1:128]
add r1, r0, #16
vld1.64 {d18, d19}, [r2:128]
vld1.64 {d22, d23}, [r1:128]
vmovn.i32 d17, q8
vmovn.i32 d16, q9
vmovn.i32 d18, q10
vmovn.i32 d19, q11
vmovn.i16 d17, q8
vmovn.i16 d16, q9
vmov r0, r1, d16
vmov r2, r3, d17
bx lr
.globl _test2
.align 2
_test2: @ @test2
@ BB#0:
vld1.64 {d16, d17}, [r0:128]
add r0, r0, #16
vld1.64 {d18, d19}, [r0:128]
vmovn.i32 d16, q8
vmovn.i32 d17, q9
vmovn.i16 d16, q8
vmov r0, r1, d16
bx lr
llvm-svn: 179989
The ARM backend currently has poor codegen for long sext/zext
operations, such as v8i8 -> v8i32. This patch addresses this
by performing a custom expansion in ARMISelLowering. It also
adds/changes the cost of such lowering in ARMTTI.
This partially addresses PR14867.
Patch by Pete Couperus
llvm-svn: 177380
The default logic marks them as too expensive.
For example, before this patch we estimated:
cost of 16 for instruction: %r = uitofp <4 x i16> %v0 to <4 x float>
While this translates to:
vmovl.u16 q8, d16
vcvt.f32.u32 q8, q8
All other costs are left to the values assigned by the fallback logic. Theses
costs are mostly reasonable in the sense that they get progressively more
expensive as the instruction sequences emitted get longer.
radar://13445992
llvm-svn: 177334
Fix cost of some "cheap" cast instructions. Before this patch we used to
estimate for example:
cost of 16 for instruction: %r = fptoui <4 x float> %v0 to <4 x i16>
While we would emit:
vcvt.s32.f32 q8, q8
vmovn.i32 d16, q8
vuzp.8 d16, d17
All other costs are left to the values assigned by the fallback logic. Theses
costs are mostly reasonable in the sense that they get progressively more
expensive as the instruction sequences emitted get longer.
radar://13434072
llvm-svn: 177333
Increase the cost of v8/v16-i8 to v8/v16-i32 casts and truncates as the backend
currently lowers those using stack accesses.
This was responsible for a significant degradation on
MultiSource/Benchmarks/Trimaran/enc-pc1/enc-pc1
where we vectorize one loop to a vector factor of 16. After this patch we select
a vector factor of 4 which will generate reasonable code.
unsigned char cle[32];
void test(short c) {
unsigned short compte;
for (compte = 0; compte <= 31; compte++) {
cle[compte] = cle[compte] ^ c;
}
}
radar://13220512
llvm-svn: 176898
Simon Pilgrim