Unfortunately, we don't seem to have a direct truncation, but the
extension can be legally split into two operations so we should
support that.
llvm-svn: 213357
Clang may well start emitting these soon, and while it may not be
directly relevant for OpenCL or GLSL, the instructions were just
sitting there waiting to be used.
llvm-svn: 213356
Since the result of a SETCC for X86 is 0 or -1 in each lane, we can
move unary operations, in this case [su]int_to_fp through the mask
operation and constant fold the operation away. Generally speaking:
UNARYOP(AND(VECTOR_CMP(x,y), constant))
--> AND(VECTOR_CMP(x,y), constant2)
where constant2 is UNARYOP(constant).
This implements the transform where UNARYOP is [su]int_to_fp.
For example, consider the simple function:
define <4 x float> @foo(<4 x float> %val, <4 x float> %test) nounwind {
%cmp = fcmp oeq <4 x float> %val, %test
%ext = zext <4 x i1> %cmp to <4 x i32>
%result = sitofp <4 x i32> %ext to <4 x float>
ret <4 x float> %result
}
Before this change, the SSE code is generated as:
LCPI0_0:
.long 1 ## 0x1
.long 1 ## 0x1
.long 1 ## 0x1
.long 1 ## 0x1
.section __TEXT,__text,regular,pure_instructions
.globl _foo
.align 4, 0x90
_foo: ## @foo
cmpeqps %xmm1, %xmm0
andps LCPI0_0(%rip), %xmm0
cvtdq2ps %xmm0, %xmm0
retq
After, the code is improved to:
LCPI0_0:
.long 1065353216 ## float 1.000000e+00
.long 1065353216 ## float 1.000000e+00
.long 1065353216 ## float 1.000000e+00
.long 1065353216 ## float 1.000000e+00
.section __TEXT,__text,regular,pure_instructions
.globl _foo
.align 4, 0x90
_foo: ## @foo
cmpeqps %xmm1, %xmm0
andps LCPI0_0(%rip), %xmm0
retq
The cvtdq2ps has been constant folded away and the floating point 1.0f
vector lanes are materialized directly via the ModRM operand of andps.
llvm-svn: 213342
Since the result of a SETCC for AArch64 is 0 or -1 in each lane, we can
move unary operations, in this case [su]int_to_fp through the mask
operation and constant fold the operation away. Generally speaking:
UNARYOP(AND(VECTOR_CMP(x,y), constant))
--> AND(VECTOR_CMP(x,y), constant2)
where constant2 is UNARYOP(constant).
This implements the transform where UNARYOP is [su]int_to_fp.
For example, consider the simple function:
define <4 x float> @foo(<4 x float> %val, <4 x float> %test) nounwind {
%cmp = fcmp oeq <4 x float> %val, %test
%ext = zext <4 x i1> %cmp to <4 x i32>
%result = sitofp <4 x i32> %ext to <4 x float>
ret <4 x float> %result
}
Before this change, the code is generated as:
fcmeq.4s v0, v0, v1
movi.4s v1, #0x1 // Integer splat value.
and.16b v0, v0, v1 // Mask lanes based on the comparison.
scvtf.4s v0, v0 // Convert each lane to f32.
ret
After, the code is improved to:
fcmeq.4s v0, v0, v1
fmov.4s v1, #1.00000000 // f32 splat value.
and.16b v0, v0, v1 // Mask lanes based on the comparison.
ret
The svvtf.4s has been constant folded away and the floating point 1.0f
vector lanes are materialized directly via fmov.4s.
Rather than do the folding manually in the target code, teach getNode()
in the generic SelectionDAG to handle folding constant operands of
vector [su]int_to_fp nodes. It is reasonable (as noted in a FIXME) to do
additional constant folding there as well, but I don't have test cases
for those operations, so leaving them for another time when it becomes
appropriate.
rdar://17693791
llvm-svn: 213341
Options struct and move the comment to inMips16HardFloat. Use the
fact that we now know whether or not we cared about soft float to
set the libcalls.
Accordingly rename mipsSEUsesSoftFloat to abiUsesSoftFloat and
propagate since it's no longer CPU specific.
llvm-svn: 213335
We now consider the FPOpFusion flag when determining whether
to fuse ops. We also explicitly emit add.rn when fusion is
disabled to prevent ptxas from fusing the operations on its
own.
llvm-svn: 213287
There are two parts here. First is to modify tablegen to adjust the encoding
type ENCODING_RM with the scaling factor.
The second is to use the new encoding types to compute the correct
displacement in the decoder.
Fixes <rdar://problem/17608489>
llvm-svn: 213281
Passes the computed scaling factor in TSFlags rather than the old attributes.
Also removes the C++ version of computing the scaling factor (MemObjSize)
along with the asserts added by the previous patch.
No functional change.
llvm-svn: 213279
This does not actually move the logic yet but reimplements it in the Tablegen
language. Then asserts that the new implementation results in the same value.
The next patch will remove the assert and the temporary use of the TSFlags and
remove the C++ implementation.
The formula requires a limited form of the logical left and right operators.
I implemented these with the bit-extract/insert operator (i.e. blah{bits}).
No functional change.
llvm-svn: 213278
This also uses TSFlags to mark machine instructions that are surface/texture
accesses, as well as the vector width for surface operations. This is used
to simplify some of the switch statements that need to detect surface/texture
instructions
llvm-svn: 213256
Previously we asserted on this code. Currently compiler-rt doesn't
actually implement any of these new libcalls, but external help is
pretty much the only viable option for LLVM.
I've followed the much more generic "__truncST2" naming, as opposed to
the odd name for f32 -> f16 truncation. This can obviously be changed
later, or overridden by any targets that need to.
llvm-svn: 213252
x86 has no native ability to extend an f16 to f64, but the same result
is obtained if we expand it into two separate extensions: f16 -> f32
-> f64.
Unfortunately the same is not true for truncate, so that still results
in a compilation failure.
llvm-svn: 213251
This makes the two intrinsics @llvm.convert.from.f16 and
@llvm.convert.to.f16 accept types other than simple "float". This is
only strictly needed for the truncate operation, since otherwise
double rounding occurs and there's no way to represent the strict IEEE
conversion. However, for symmetry we allow larger types in the extend
too.
During legalization, we can expand an "fp16_to_double" operation into
two extends for convenience, but abort when the truncate isn't legal. A new
libcall is probably needed here.
Even after this commit, various target tweaks are needed to actually use the
extended intrinsics. I've put these into separate commits for clarity, so there
are no actual tests of f64 conversion here.
llvm-svn: 213248
Memory barrier __builtin_arm_[dmb, dsb, isb] intrinsics are required to
implement their corresponding ACLE and MSVC intrinsics.
This patch ports ARM dmb, dsb, isb intrinsic to AArch64.
Differential Revision: http://reviews.llvm.org/D4520
llvm-svn: 213247
Summary:
Generally speaking, mips-* vs mips64-* should not be used to make decisions
about the content or format of the ELF. This should be based on the ABI
and CPU in use. For example, `mips-linux-gnu-clang -mips64r2 -mabi=64`
should produce an ELF64 as should `mips64-linux-gnu-clang -mabi=64`.
Conversely, `mips64-linux-gnu-clang -mabi=n32` should produce an ELF32 as
should `mips-linux-gnu-clang -mips64r2 -mabi=n32`.
This patch fixes the e_flags but leaves the ELF32 vs ELF64 issue for now
since there is no apparent way to base this decision on the ABI and CPU.
Differential Revision: http://reviews.llvm.org/D4539
llvm-svn: 213244
Summary:
The cpr1_size field describes the minimum register width to run the program
rather than the size of the registers on the target. MIPS32r6 was acting
as if -mfp64 has been given because it starts off with 64-bit FPU registers.
Differential Revision: http://reviews.llvm.org/D4538
llvm-svn: 213243
Summary:
These options are not implemented yet but we act as if they are always
given.
The integrated assembler is driven by the clang driver so the e_flag test
cases should match the e_flags emitted by GCC+GAS rather than GAS
by itself.
Differential Revision: http://reviews.llvm.org/D4536
llvm-svn: 213242
We were not considering the stated alignment on vector loads/stores,
leading us to generate vector instructions even when we do not have
sufficient alignment.
Now, for IR like:
%1 = load <4 x float>, <4 x float>* %ptr, align 4
we will generate correct, conservative PTX like:
ld.f32 ... [%ptr]
ld.f32 ... [%ptr+4]
ld.f32 ... [%ptr+8]
ld.f32 ... [%ptr+12]
Or if we have an alignment of 8 (for example), we can
generate code like:
ld.v2.f32 ... [%ptr]
ld.v2.f32 ... [%ptr+8]
llvm-svn: 213186
It turns out that in most cases (the main exception being i1-related
types) once these operations are formed we cannot separate them and
the targets end up having to deal with them whether they want to or
not.
This is not a good situation, and a more reasonable default can be
formed by ackowledging this and having targets leave them as Legal.
Only x86 seems to be affected (other targets don't even try marking
the operation Expand).
Mostly there's no visible change here yet, but it will be useful to
have truly expanded EXTLOADS for MVT::f16 softening support.
llvm-svn: 213162
Summary:
A few instructions (mostly cvt.d.w and similar) are causing problems with
-mfp64 and -mno-odd-spreg and it looks like fixing it properly may
take several weeks. In the meantime, let's disable the odd-numbered
double-precision registers so that the generated code is at least valid.
The problem is that instructions like cvt.d.w read from the 32-bit low
subregister of a double-precision FPU register. This often leads to the compiler
to inserting moves to transfer a GPR32 to a FGR32 using mtc1. Such moves
violate the rules against 32-bit writes to odd-numbered FPU registers imposed
by -mno-odd-spreg. By disabling the odd-numbered double-precision registers, it
becomes impossible for the 32-bit low subregister to be odd-numbered.
This fixes numerous test-suite failures when compiling for the FP64A ABI
('-mfp64 -mno-odd-spreg'). There is no LLVM test case because it's difficult to
test that odd-numbered FPU registers are not allocatable. Instead, we depend on
the assembler (GAS and -fintegrated-as) raising errors when the rules are
violated.
Differential Revision: http://reviews.llvm.org/D4532
llvm-svn: 213160
Before this change, method 'isShuffleMaskLegal' didn't know that shuffles
implementing a 'movhlps' operation were perfectly legal for SSE targets.
This patch adds the missing check for 'isMOVHLPSMask' inside method
'isShuffleMaskLegal' to fix the problem.
The reason why it is important to do this is because the DAGCombiner
conservatively avoids combining a pair of shuffles if the resulting shuffle
node has an illegal mask. Before this patch, shuffles with a MOVHLPS mask were
wrongly considered not to be legal. This was the root cause of some poor-code
generation bugs.
llvm-svn: 213137
There exists a helper function to abstract away the various differences
between ConstantVector, ConstantDataVector, ConstantAggregateZero, etc.
Use it to simplify X86WindowsTargetObjectFile::getSectionForConstant.
llvm-svn: 213104
Refactoring; no functional changes intended
Removed PostRAScheduler bits from subtargets (X86, ARM).
Added PostRAScheduler bit to MCSchedModel class.
This bit is set by a CPU's scheduling model (if it exists).
Removed enablePostRAScheduler() function from TargetSubtargetInfo and subclasses.
Fixed the existing enablePostMachineScheduler() method to use the MCSchedModel (was just returning false!).
Added methods to TargetSubtargetInfo to allow overrides for AntiDepBreakMode, CriticalPathRCs, and OptLevel for PostRAScheduling.
Added enablePostRAScheduler() function to PostRAScheduler class which queries the subtarget for the above values.
Preserved existing scheduler behavior for ARM, MIPS, PPC, and X86:
a. ARM overrides the CPU's postRA settings by enabling postRA for any non-Thumb or Thumb2 subtarget.
b. MIPS overrides the CPU's postRA settings by enabling postRA for everything.
c. PPC overrides the CPU's postRA settings by enabling postRA for everything.
d. X86 is the only target that actually has postRA specified via sched model info.
Differential Revision: http://reviews.llvm.org/D4217
llvm-svn: 213101
The coalescer is very aggressive at propagating constraints on the register classes, and the register allocator doesn’t know how to split sub-registers later to recover. This patch provides an escape valve for targets that encounter this problem to limit coalescing.
This patch also implements such for ARM to lower register pressure when using lots of large register classes. This works around PR18825.
llvm-svn: 213078
v2: use ffbh/l if available
v3: Rebase on top of Matt's SI patches
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Reviewed-by: Tom Stellard <tom@stellard.net>
llvm-svn: 213072
Summary: Previously all the test cases set it after initialization with '.module fp=xx'.
Differential Revision: http://reviews.llvm.org/D4489
llvm-svn: 213071
Fixes a gcc warning caused by a typo. A redundant assignment operation was
accidentally used as the third operand of a conditional expression.
No functional change intended.
llvm-svn: 213061
This implements the FastLowerCall hook, which is based on the DoSelectCall
function. The implementation is very similar, but the target-independent call
lowering part has been factored out.
This should also enable patchpoint intrinsic lowering for FastISel on X86.
Related to <rdar://problem/17427052>.
llvm-svn: 213049
Revert "[FastISel][X86] Implement the FastLowerIntrinsicCall hook."
Revert "[FastISel][X86] Implement the FastLowerCall hook."
This reverts commit r213035, r213036, and r213037 to make the
buildbots happy again.
llvm-svn: 213048
The constant pool entry code for WinCOFF assumed that vector constants
would be formed using ConstantDataVector, it did not expect to see a
ConstantVector. Furthermore, it did not expect undef as one of the
elements of the vector.
ConstantVectors should be handled like ConstantDataVectors, treat Undef
as zero.
llvm-svn: 213038
This implements the FastLowerCall hook, which is based on the DoSelectCall
function. The implementation is very similar, but the target-independent call
lowering part has been factored out.
This should also enable patchpoint intrinsic lowering for FastISel on X86.
Related to <rdar://problem/17427052>.
llvm-svn: 213035
This helps avoid redundant instructions to unpack, and repack
the vectors. Ideally we could recognize that pattern and eliminate
it. Currently v4i8 and other small element type vectors are scalarized,
so this has the added bonus of avoiding that.
llvm-svn: 213031
No functional change.
The offsets for the other bitfields are specified symbolically. I need to
increase the size for one of the earlier fields which is easier after this
cleanup.
Why these bits are relative to VEXShift is a bit strange but that is for
another cleanup.
I made sure that the values for the enums are unchanged after this change.
llvm-svn: 213011
COFF lacks a feature that other object file formats support: mergeable
sections.
To work around this, MSVC sticks constant pool entries in special COMDAT
sections so that each constant is in it's own section. This permits
unused constants to be dropped and it also allows duplicate constants in
different translation units to get merged together.
This fixes PR20262.
Differential Revision: http://reviews.llvm.org/D4482
llvm-svn: 213006
We would emit a libcall for a 64-bit atomic on x86 after SVN r212119. This was
due to the misuse of hasCmpxchg16 to indicate if cmpxchg8b was supported on a
32-bit target. They were added at different times and would result in the
border condition being mishandled.
This fixes the border case to emit the cmpxchg8b instruction for 64-bit atomic
operations on x86 at the cost of restoring a long-standing bug in the codegen.
We emit a cmpxchg8b on all x86 targets even where the CPU does not support this
instruction (pre-Pentium CPUs). Although this bug should be fixed, this was
present prior to SVN r212119 and this change, so this is not really introducing
a regression.
llvm-svn: 212956
We construct a temporary "atomicrmw xchg" instruction when lowering atomic
stores for widths that aren't supported natively. This isn't on the top-level
worklist though, so it won't be removed automatically and we have to do it
ourselves once that itself has been lowered.
Thanks Saleem for pointing this out!
llvm-svn: 212948
Summary:
.bss, .text, and .data are at least 16-byte aligned.
.reginfo is 4-byte aligned and has a 24-byte EntrySize.
.MIPS.abiflags has an 24-byte EntrySize.
.MIPS.options is 8-byte aligned and has 1-byte EntrySize.
Using a 1-byte EntrySize for .MIPS.options seems strange because the
records are neither 1-byte long nor fixed-length but this matches the value
that GAS emits.
Differential Revision: http://reviews.llvm.org/D4487
llvm-svn: 212939
Summary:
This is because the FP64A the hardware will redirect 32-bit reads/writes
from/to odd-numbered registers to the upper 32-bits of the corresponding
even register. In effect, simulating FR=0 mode when FR=0 mode is not
available.
Unfortunately, we have to make the decision to avoid mfc1/mtc1 before
register allocation so we currently do this for even registers too.
FPXX has a similar requirement on 32-bit architectures that lack
mfhc1/mthc1 so this patch also handles the affected moves from the FPU for
FPXX too. Moves to the FPU were supported by an earlier commit.
Differential Revision: http://reviews.llvm.org/D4484
llvm-svn: 212938
Summary:
This is similar to r210771 which did the same thing for MTHC1.
Also corrected MTHC1_D32 and MTHC1_D64 which used AFGR64 and FGR64 on the
wrong definitions.
Differential Revision: http://reviews.llvm.org/D4483
llvm-svn: 212936
enabled and mthc1 and dmtc1 are not available (e.g. on MIPS32r1)
This prevents the upper 32-bits of a double precision value from being moved to
the FPU with mtc1 to an odd-numbered FPU register. This is necessary to ensure
that the code generated executes correctly regardless of the current FPU mode.
MIPS32r2 and above continues to use mtc1/mthc1, while MIPS-IV and above continue
to use dmtc1.
Differential Revision: http://reviews.llvm.org/D4465
llvm-svn: 212930
Rename member variables and functions for the MCStreamer for DWARF-like
unwinding management. Rename the Windows ones as well and make the naming and
handling similar across the two. No functional change intended.
llvm-svn: 212912
This adds a llvm.aarch64.hint intrinsic to mirror the llvm.arm.hint in order to
support the various hint intrinsic functions in the ACLE.
Add an optional pattern field that permits the subclass to specify the pattern
that matches the selection. The intrinsic pattern is set as mayLoad, mayStore,
so overload the value for the definition of the hint instruction.
llvm-svn: 212883
This commit fixes a bug in PPCRegisterInfo::isFrameOffsetLegal that
could result in the LocalStackAlloc pass creating an MI instruction
out-of-range displacement:
%vreg17<def> = LD 33184, %vreg31; mem:LD8[%g](align=32)
%G8RC:%vreg17 G8RC_and_G8RC_NOX0:%vreg31
(In final assembler output the top bits are stripped off, resulting
in a negative offset loading from below the stack pointer.)
Common code expects the isFrameOffsetLegal routine to verify whether
adding a given offset to the offset already present in the instruction
results in a valid displacement. However, on PowerPC the routine
did not take the already present instruction offset into account.
This commit fixes isFrameOffsetLegal to add the instruction offset,
and updates a local caller (needsFrameBaseReg) to no longer add the
instruction offset itself before calling isFrameOffsetLegal.
Reviewed by Hal Finkel.
llvm-svn: 212832
We need the intrinsics with offsets, so why not just add them all.
The R128 parameter will also be useful for reducing SGPR usage.
GL_ARB_image_load_store also adds some image GLSL modifiers like "coherent",
so Mesa will probably translate those to slc, glc, etc.
When LLVM 3.5 is released, I'll switch Mesa to these new intrinsics.
llvm-svn: 212830
No functional change. As I was trying to understand this function, I found
that variables were reused with confusing names and the broadcast case was a
bit too implicit. Hopefully, this is an improvement.
llvm-svn: 212795
It was computing the VL/n case as:
MemObjSize = VectorByteSize / ElemByteSize / Divider * ElemByteSize
ElemByteSize not only falls out but VectorByteSize/Divider now actually
matches the definition of VL/n.
Also some formatting fixes.
llvm-svn: 212794
Use alg. from LegalizeDAG.cpp
Move Expand setting to SIISellowering
v2: Extend existing tests instead of creating new ones
v3: use separate LowerFPTOSINT function
v4: use TargetLowering::expandFP_TO_SINT
add comment about using FP_TO_SINT for uints
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Reviewed-by: Tom Stellard <tom@stellard.net>
llvm-svn: 212773
passes in the mips back end. This, unfortunately, required a
bit of churn in the various predicates to use a pointer rather
than a reference.
llvm-svn: 212744
This patch teaches the AsmParser to accept some logical+immediate
instructions and convert them as shown:
bic Rd, Rn, #imm -> and Rd, Rn, #~imm
bics Rd, Rn, #imm -> ands Rd, Rn, #~imm
orn Rd, Rn, #imm -> orr Rd, Rn, #~imm
eon Rd, Rn, #imm -> eor Rd, Rn, #~imm
Those instructions are an alternate syntax available to assembly coders,
and are needed in order to support code already compiling with some other
assemblers. For example, the bic construct is used by the linux kernel.
llvm-svn: 212722
Summary:
When -mno-odd-spreg is in effect, 32-bit floating point values are not
permitted in odd FPU registers. The option also prohibits 32-bit and 64-bit
floating point comparison results from being written to odd registers.
This option has three purposes:
* It allows support for certain MIPS implementations such as loongson-3a that
do not allow the use of odd registers for single precision arithmetic.
* When using -mfpxx, -mno-odd-spreg is the default and this allows us to
statically check that code is compliant with the O32 FPXX ABI since mtc1/mfc1
instructions to/from odd registers are guaranteed not to appear for any
reason. Once this has been established, the user can then re-enable
-modd-spreg to regain the use of all 32 single-precision registers.
* When using -mfp64 and -mno-odd-spreg together, an O32 extension named
O32 FP64A is used as the ABI. This is intended to provide almost all
functionality of an FR=1 processor but can also be executed on a FR=0 core
with the assistance of a hardware compatibility mode which emulates FR=0
behaviour on an FR=1 processor.
* Added '.module oddspreg' and '.module nooddspreg' each of which update
the .MIPS.abiflags section appropriately
* Moved setFpABI() call inside emitDirectiveModuleFP() so that the caller
doesn't have to remember to do it.
* MipsABIFlags now calculates the flags1 and flags2 member on demand rather
than trying to maintain them in the same format they will be emitted in.
There is one portion of the -mfp64 and -mno-odd-spreg combination that is not
implemented yet. Moves to/from odd-numbered double-precision registers must not
use mtc1. I will fix this in a follow-up.
Differential Revision: http://reviews.llvm.org/D4383
llvm-svn: 212717
This is minimal change for backend required to have "hello world" compiled and working on x32 target (x86_64-linux-gnux32). More patches for x32 will follow.
Differential Revision: http://reviews.llvm.org/D4181
llvm-svn: 212716
There's no real need to have Shift as a separate format type from Binary.
The comments for other format types were too specific and in some cases
no longer accurate.
Just a clean-up, no behavioral change intended.
llvm-svn: 212707
shuffle lowering: match shuffle patterns equivalent to an unpcklwd or
unpckhwd instruction.
This allows us to use generic lowering code for v8i16 shuffles and match
the unpack pattern late.
llvm-svn: 212705
Immediate fields that have no natural MVT type tended to use i8 if the
field was small enough. This was a bit confusing since i8 isn't a legal
type for the target. Fields for short immediates in a 32-bit or 64-bit
operation use i32 or i64 instead, so it would be better to do the same
for all fields.
No behavioral change intended.
llvm-svn: 212702
The dwarf FPR numbers are supposed to have the order F0, F2, F4, F6,
F1, F3, F5, F7, F8, etc., which matches the pairing of registers for
long doubles. E.g. a long double stored in F0 is paired with F2.
llvm-svn: 212701
Summary:
On MIPS32r6/MIPS64r6, floating point comparisons return 0 or -1 but integer
comparisons return 0 or 1.
Updated the various uses of getBooleanContents. Two simplifications had to be
disabled when float and int boolean contents differ:
- ScalarizeVecRes_VSELECT except when the kind of boolean contents is trivially
discoverable (i.e. when the condition of the VSELECT is a SETCC node).
- visitVSELECT (select C, 0, 1) -> (xor C, 1).
Come to think of it, this one could test for the common case of 'C'
being a SETCC too.
Preserved existing behaviour for all other targets and updated the affected
MIPS32r6/MIPS64r6 tests. This also fixes the pi benchmark where the 'low'
variable was counting in the wrong direction because it thought it could simply
add the result of the comparison.
Reviewers: hfinkel
Reviewed By: hfinkel
Subscribers: hfinkel, jholewinski, mcrosier, llvm-commits
Differential Revision: http://reviews.llvm.org/D4389
llvm-svn: 212697
combine into half-shuffles through unpack instructions that expand the
half to a whole vector without messing with the dword lanes.
This fixes some redundant instructions in splat-like lowerings for
v16i8, which are now getting to be *really* nice.
llvm-svn: 212695
that splat i8s into i16s.
Previously, we would try much too hard to arrange a sequence of i8s in
one half of the input such that we could unpack them into i16s and
shuffle those into place. This isn't always going to be a cheaper i8
shuffle than our other strategies. The case where it is always going to
be cheaper is when we can arrange all the necessary inputs into one half
using just i16 shuffles. It happens that viewing the problem this way
also makes it much easier to produce an efficient set of shuffles to
move the inputs into one half and then unpack them.
With this, our splat code gets one step closer to being not terrible
with the new experimental lowering strategy. It also exposes two
combines missing which I will add next.
llvm-svn: 212692
shuffles specifically for cases where a small subset of the elements in
the input vector are actually used.
This is specifically targetted at improving the shuffles generated for
trunc operations, but also helps out splat-like operations.
There is still some really low-hanging fruit here that I want to address
but this is a huge step in the right direction.
llvm-svn: 212680
don't need to set it manually.
This is based on feedback from Tom who pointed out that if every target
needs to handle this we need to reach out to those maintainers. In fact,
it doesn't make sense to duplicate everything when anything other than
expand seems unlikely at this stage.
llvm-svn: 212661
Storing will generally be immediately preceded by rounding from an f32
or f64, so make sure to match those patterns directly to convert into the
FPR16 register class directly rather than going through the integer GPRs.
This also eliminates an extra step in the convert-from-f64 path
which was first converting to f32 and then to f16 from there.
rdar://17594379
llvm-svn: 212638
This lets us experiment with 512-bit vectorization without passing
force-vector-width manually.
The code generated for a simple integer memset loop is properly vectorized.
Disassembly is still broken for it though :(.
llvm-svn: 212634
This is a follow up to r212492. There should be no functional difference, but
this patch makes it clear that SrcVT must be an i1/i8/16/i32 and DestVT must be
an i8/i16/i32/i64.
rdar://17516686
llvm-svn: 212633
Turns out my trick of using the same masks for SSE4.1 and AVX2 didn't work out
as we have to blend two vectors. While there remove unecessary cross-lane moves
from the shuffles so the backend can lower it to palignr instead of vperm.
Fixes PR20118, a miscompilation of vector sdiv by constant on AVX2.
llvm-svn: 212611
vector types to be legal and a ZERO_EXTEND node is encountered.
When we use widening to legalize vector types, extend nodes are a real
challenge. Either the input or output is likely to be legal, but in many
cases not both. As a consequence, we don't really have any way to
represent this situation and the prior code in the widening legalization
framework would just scalarize the extend operation completely.
This patch introduces a new DAG node to represent doing a zero extend of
a vector "in register". The core of the idea is to allow legal but
different vector types in the input and output. The output vector must
have fewer lanes but wider elements. The operation is defined to zero
extend the low elements of the input to the size of the output elements,
and drop all of the high elements which don't have a corresponding lane
in the output vector.
It also includes generic expansion of this node in terms of blending
a zero vector into the high elements of the vector and bitcasting
across. This in turn yields extremely nice code for x86 SSE2 when we use
the new widening legalization logic in conjunction with the new shuffle
lowering logic.
There is still more to do here. We need to support sign extension, any
extension, and potentially int-to-float conversions. My current plan is
to continue using similar synthetic nodes to model each of these
transitions with generic lowering code for each one.
However, with this patch LLVM already reaches performance parity with
GCC for the core C loops of the x264 code (assuming you disable the
hand-written assembly versions) when compiling for SSE2 and SSE3
architectures and enabling the new widening and lowering logic for
vectors.
Differential Revision: http://reviews.llvm.org/D4405
llvm-svn: 212610
Summary:
It seems we accidentally read the wrong column of the table MIPS64r6 spec
and used the names for c.cond.fmt instead of cmp.cond.fmt.
Differential Revision: http://reviews.llvm.org/D4387
llvm-svn: 212607
Summary:
This completes the change to use JALR instead of JR on MIPS32r6/MIPS64r6.
Reviewers: jkolek, vmedic, zoran.jovanovic, dsanders
Reviewed By: dsanders
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D4269
llvm-svn: 212605
Summary:
RET, and RET_MM have been replaced by a pseudo named PseudoReturn.
In addition a version with a 64-bit GPR named PseudoReturn64 has been
added.
Instruction selection for a return matches RetRA, which is expanded post
register allocation to PseudoReturn/PseudoReturn64. During MipsAsmPrinter,
this PseudoReturn/PseudoReturn64 are emitted as:
- (JALR64 $zero, $rs) on MIPS64r6
- (JALR $zero, $rs) on MIPS32r6
- (JR_MM $rs) on microMIPS
- (JR $rs) otherwise
On MIPS32r6/MIPS64r6, 'jr $rs' is an alias for 'jalr $zero, $rs'. To aid
development and review (specifically, to ensure all cases of jr are
updated), these aliases are temporarily named 'r6.jr' instead of 'jr'.
A follow up patch will change them back to the correct mnemonic.
Added (JALR $zero, $rs) to MipsNaClELFStreamer's definition of an indirect
jump, and removed it from its definition of a call.
Note: I haven't accounted for MIPS64 in MipsNaClELFStreamer since it's
doesn't appear to account for any MIPS64-specifics.
The return instruction created as part of eh_return expansion is now expanded
using expandRetRA() so we use the right return instruction on MIPS32r6/MIPS64r6
('jalr $zero, $rs').
Also, fixed a misuse of isABI_N64() to detect 64-bit wide registers in
expandEhReturn().
Reviewers: jkolek, vmedic, mseaborn, zoran.jovanovic, dsanders
Reviewed By: dsanders
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D4268
llvm-svn: 212604
has settled without incident, removing the x86-specific and overly
strict 'isVectorSplat' routine in favor of generic and more powerful
splat detection.
The primary motivation and result of this is that the x86 backend can
now see through splats which contain undef elements. This is essential
if we are using a widening form of legalization and I've updated a test
case to also run in that mode as before this change the generated code
for the test case was completely scalarized.
This version of the patch much more carefully handles the undef lanes.
- We aren't overly conservative about them in the shift lowering
(where we will never use the splat itself).
- One place where the splat would have been re-used by the existing code
now explicitly constructs a new constant splat that will be safe.
- The broadcast lowering is much more reasonable with undefs by doing
a correct check of whether the splat is the only user of a loaded
value, checking that the splat actually crosses multiple lanes before
using a broadcast, and handling broadcasts of non-constant splats.
As a consequence of the last bullet, the weird usage of vpshufd instead
of vbroadcast is gone, and we actually can lower an AVX splat with
vbroadcastss where before we emitted a really strange pattern of
a vector load and a manual splat across the vector.
llvm-svn: 212602
Loading will generally extend to an f32 or an 64, so make sure
to match those patterns directly to load into the FPR16 register
class directly rather than going through the integer GPRs.
This also eliminates an extra step in the convert-to-f64 path
which was first converting to f32 and then to f64 from there.
rdar://17594379
llvm-svn: 212573
This changes the implementation of atomic NAND operations
from "a & ~b" (compatible with GCC < 4.4) to actual "~(a & b)"
(compatible with GCC >= 4.4).
This is in line with the common-code and ARM back-end change
implemented in r212433.
llvm-svn: 212547
Summary:
Follow on to r212519 to improve the encapsulation and limit the scope of the enums.
Also merged two very similar parser functions, fixed a bug where ASE's
were not being reported, and marked CPR1's as being 128-bit when MSA is
enabled.
Differential Revision: http://reviews.llvm.org/D4384
llvm-svn: 212522
aggressively from the x86 shuffle lowering to the generic SDAG vector
shuffle formation code.
This code already tried to fold away shuffles of splats! It just had
lots of bugs and couldn't handle the case my new x86 shuffle lowering
needed.
First, it failed to correctly compute whether N2 was undef because it
pre-computed this, then did transformations which could *make* N2 undef,
then failed to ever re-consider the precomputed state.
Second, it didn't look through bitcasts at all, even in the safe cases
where they are just element-type bitcasts with no change to the number
of elements.
Third, it didn't handle all-zero bit casts nicely the way my code in the
x86 side of things did, which is essential to getting good zext-shuffle
lowerings.
But all of these are generic. I just ported the code down to this layer
and fixed the surrounding bugs. Tests exercising this in the x86 backend
still pass and some silly code in widen_cast-6.ll gets better. I updated
that test to be a bit more precise but it's still pretty unclear what
the value of the test is in this day and age.
llvm-svn: 212517
As destination k0 is allowed but not as predicate/writemask.
I also modified the test to allow checking of error messages by the assembler.
I applied a similar approach to the test ret.s in the same directory.
llvm-svn: 212504
When combining a sequence of two PSHUFD dag nodes into a single PSHUFD,
make sure that we assign the correct type to the resulting PSHUFD.
X86ISD::PSHUFD dag nodes can be either MVT::v4i32 or MVT::v4f32.
Before this change, an assertion failure was triggered in method
'DAGCombinerInfo::CombineTo' when trying to combine the shuffles from the test
below into a single PSHUFD.
define <4 x float> @test1(<4 x float> %V) {
%1 = shufflevector <4 x float> %V, <4 x float> undef, <4 x i32> <i32 3, i32 0, i32 2, i32 1>
%2 = shufflevector <4 x float> %1, <4 x float> undef, <4 x i32> <i32 3, i32 0, i32 2, i32 1>
ret <4 x float> %2
}
llvm-svn: 212498
Add custom lowering code for signed multiply instruction selection, because the
default FastISel instruction selection for ISD::MUL will use unsigned multiply
for the i8 type and signed multiply for all other types. This would set the
incorrect flags for the overflow check.
This fixes <rdar://problem/17549300>
llvm-svn: 212493
Currently AArch64FastISel crashes if it tries to extend an integer into an
MVT::i128. This can happen by creating 128 bit integers like so:
typedef unsigned int uint128_t __attribute__((mode(TI)));
typedef int sint128_t __attribute__((mode(TI)));
This patch makes EmitIntExt check for their presence and then falls back to
SelectionDAG.
Tests included.
rdar://17516686
llvm-svn: 212492
Arguments passed as "byval align" should get the specified alignment
in the parameter save area. There was some code in PPCISelLowering.cpp
that attempted to implement this, but this didn't work correctly:
while code did update the ArgOffset value, it neglected to update
the PtrOff value (which was already computed from the old ArgOffset),
and it also neglected to update GPR_idx -- fields skipped due to
alignment in the save area must likewise be skipped in GPRs.
This patch fixes and simplifies this logic by:
- handling argument offset alignment right at the beginning
of argument processing, using a new helper routine
CalculateStackSlotAlignment (this avoids having to update
PtrOff and other derived values later on)
- not tracking GPR_idx separately, but always computing the
correct GPR_idx for each argument *from* its ArgOffset
- removing some redundant computation in LowerFormalArguments:
MinReservedArea must equal ArgOffset after argument processing,
so there's no use in computing it twice.
[This doesn't change the behavior of the current clang front-end,
since that never creates "byval align" arguments at the moment.
This will change with a follow-on patch, however.]
llvm-svn: 212476
lanes in vector splats.
The core problem here is that undef lanes can't *unilaterally* be
considered to contribute to splats. Their handling needs to be more
cautious. There is also a reported failure of the nightly testers
(thanks Tobias!) that may well stem from the same core issue. I'm going
to fix this theoretical issue, factor the APIs a bit better, and then
verify that I don't see anything bad with Tobias's reduction from the
test suite before recommitting.
Original commit message for r212324:
[x86] Generalize BuildVectorSDNode::getConstantSplatValue to work for
any constant, constant FP, or undef splat and to tolerate any undef
lanes in a splat, then replace all uses of isSplatVector in X86's
lowering with it.
This fixes issues where undef lanes in an otherwise splat vector would
prevent the splat logic from firing. It is a touch more awkward to use
this interface, but it is much more accurate. Suggestions for better
interface structuring welcome.
With this fix, the code generated with the widening legalization
strategy for widen_cast-4.ll is *dramatically* improved as the special
lowering strategies for a v16i8 SRA kick in even though the high lanes
are undef.
We also get a slightly different choice for broadcasting an aligned
memory location, and use vpshufd instead of vbroadcastss. This looks
like a minor win for pipelining and domain crossing, but a minor loss
for the number of micro-ops. I suspect its a wash, but folks can
easily tweak the lowering if they want.
llvm-svn: 212475
essentially a DAG combine that never gets a chance to run.
We might typically expect DAG combining to remove shuffles-of-splats and
other similar patterns, but we don't get a chance to run the DAG
combiner when we recursively form sub-shuffles during the lowering of
a shuffle. So instead hand-roll a really important combine directly into
the lowering code to detect shuffles-of-splats, especially shuffles of
an all-zero splat which needn't even have the same element width, etc.
This lets the new vector shuffle lowering handle shuffles which
implement things like zero-extension really nicely. This will become
even more important when I wire the legalization of zero-extension to
vector shuffles with the new widening legalization strategy.
llvm-svn: 212444
We've been performing the wrong operation on ARM for "atomicrmw nand" for
years, since "a NAND b" is "~(a & b)" rather than ARM's very tempting "a & ~b".
This bled over into the generic expansion pass.
So I assume no-one has ever actually tried to do an atomic nand in the real
world. Oh well.
llvm-svn: 212443
This completes the handling for DLL import storage symbols when lowering
instructions. A DLL import storage symbol must have an additional load
performed prior to use. This is applicable to variables and functions.
This is particularly important for non-function symbols as it is possible to
handle function references by emitting a thunk which performs the translation
from the unprefixed __imp_ symbol to the proper symbol (although, this is a
non-optimal lowering). For a variable symbol, no such thunk can be
accommodated.
llvm-svn: 212431
Add support for tracking DLLImport storage class information on a per symbol
basis in the ARM instruction selection. Use that information to correctly
mangle the symbol (dllimport symbols are referenced via *__imp_<name>).
llvm-svn: 212430
Ensure that all paths that retrieve the symbol name go through GetARMGVSymbol
rather than getSymbol. This is desirable so that any global symbol mangling can
be centralised to this function. The motivation for this is handling of symbols
that are marked as having dll import dll storage. Such a symbol requires an
extra load that is currently handled in the backend and a __imp_ prefix on the
symbol name.
llvm-svn: 212429
The linker relies on relocation type info (e.g. is it a branch?) to perform the
correct actions, so we should keep that even when we end up using a scattered
relocation for whatever reason.
rdar://problem/17553104
llvm-svn: 212333
We have detected a documentation bug in the encoding tables of the released
MIPS64r6 specification that has resulted in the wrong encodings being used for
these instructions in LLVM. This commit corrects them.
llvm-svn: 212330
any constant, constant FP, or undef splat and to tolerate any undef
lanes in a splat, then replace all uses of isSplatVector in X86's
lowering with it.
This fixes issues where undef lanes in an otherwise splat vector would
prevent the splat logic from firing. It is a touch more awkward to use
this interface, but it is much more accurate. Suggestions for better
interface structuring welcome.
With this fix, the code generated with the widening legalization
strategy for widen_cast-4.ll is *dramatically* improved as the special
lowering strategies for a v16i8 SRA kick in even though the high lanes
are undef.
We also get a slightly different choice for broadcasting an aligned
memory location, and use vpshufd instead of vbroadcastss. This looks
like a minor win for pipelining and domain crossing, but a minor loss
for the number of micro-ops. I suspect its a wash, but folks can easily
tweak the lowering if they want.
llvm-svn: 212324
Silvermont can only decode one instruction per cycle if the instruction exceeds 8 bytes.
Also in Silvermont instructions with more than 3 prefixes will cause 3 cycle penalty.
Maximum nop length is limited to 7 bytes when used for padding on Silvermont.
For other x86 processors max nop length remains unchanged 15 bytes.
Differential Revision: http://reviews.llvm.org/D4374
llvm-svn: 212321
subtarget. This involved having the movt predicate take the current
function - since we care about size in instruction selection for
whether or not to use movw/movt take the function so we can check
the attributes. This required adding the current MachineFunction to
FastISel and propagating through.
llvm-svn: 212309
This patch:
1) Improves the cost model for x86 alternate shuffles (originally
added at revision 211339);
2) Teaches the Cost Model Analysis pass how to analyze alternate shuffles.
Alternate shuffles are a special kind of blend; on x86, we can often
easily lowered alternate shuffled into single blend
instruction (depending on the subtarget features).
The existing cost model didn't take into account subtarget features.
Also, it had a couple of "dead" entries for vector types that are never
legal (example: on x86 types v2i32 and v2f32 are not legal; those are
always either promoted or widened to 128-bit vector types).
The new x86 cost model takes into account what target features we have
before returning the shuffle cost (i.e. the number of instructions
after the blend is lowered/expanded).
This patch also teaches the Cost Model Analysis how to identify and analyze
alternate shuffles (i.e. 'SK_Alternate' shufflevector instructions):
- added function 'isAlternateVectorMask';
- added some logic to check if an instruction is a alternate shuffle and, in
case, call the target specific TTI to get the corresponding shuffle cost;
- added a test to verify the cost model analysis on alternate shuffles.
llvm-svn: 212296
This patch adds tablegen patterns to select F16C float-to-half-float
conversion instructions from 'f32_to_f16' and 'f16_to_f32' dag nodes.
If the target doesn't have F16C, then 'f32_to_f16' and 'f16_to_f32'
are expanded into library calls.
llvm-svn: 212293
mode.
This also runs the test in that mode which would reproduce the crash.
What I love is that *every single FIXME* in the test is addressed by
switching to widening.
llvm-svn: 212254
Finkel, Eric Christopher, and a bunch of other people I'm probably
forgetting (sorry), add an option to the x86 backend to widen vectors
during type legalization rather than promote them.
This still would promote vNi1 vectors to get the masks right, but would
widen other vectors. A lot of experiments are piling up right now
showing that widening should probably be the default legalization
strategy outside of vNi1 cases, but it is very hard to test the
rammifications of that and fix bugs in widening-based legalization
without an option that enables it. I'll be checking in tests shortly
that use this option to exercise cases where widening doesn't work well
and hopefully we'll be able to switch fully to this soon.
llvm-svn: 212249
vector type legalization strategies in a more fine grained manner, and
change the legalization of several v1iN types and v1f32 to be widening
rather than scalarization on AArch64.
This fixes an assertion failure caused by scalarizing nodes like "v1i32
trunc v1i64". As v1i64 is legal it will fail to scalarize v1i32.
This also provides a foundation for other targets to have more granular
control over how vector types are legalized.
Patch by Hao Liu, reviewed by Tim Northover. I'm committing it to allow
some work to start taking place on top of this patch as it adds some
really important hooks to the backend that I'd like to immediately start
using. =]
http://reviews.llvm.org/D4322
llvm-svn: 212242
This new multiclass, avx512_perm_table_3src derives from the current one and
provides the Pat<>. The next patch will add another Pat<> that uses the
writemask.
Note that I dropped the type annotation from the intrinsic call, i.e.: (v16f32
VR512:$src1) -> R512:$src1. I think that this should be fine (at least many
intrinsic calls don't provide them) and it greatly reduces the number of
template arguments.
llvm-svn: 212222
This includes assembler and codegen support (see the new tests in
avx512-encodings.s and avx512-shuffle.ll).
<rdar://problem/17492620>
llvm-svn: 212221
SGPRs are written by instructions that sometimes will ignore control flow,
which means if you have code like:
if (VGPR0) {
SGPR0 = S_MOV_B32 0
} else {
SGPR0 = S_MOV_B32 1
}
The value of SGPR0 will 1 no matter what the condition is.
In order to deal with this situation correctly, we need to view the
program as if it were a single basic block when we calculate the
live ranges for the SGPRs. They way we actually update the live
range is by iterating over all of the segments in each LiveRange
object and setting the end of each segment equal to the start of
the next segment. So a live range like:
[3888r,9312r:0)[10032B,10384B:0) 0@3888r
will become:
[3888r,10032B:0)[10032B,10384B:0) 0@3888r
This change will allow us to use SALU instructions within branches.
llvm-svn: 212215
heuristic.
By default, no functionality change.
This is a follow-up of r212099.
This hook provides a finer grain to control the optimization.
<rdar://problem/17444599>
llvm-svn: 212204
This reverts commits r212189 and r212190.
While this pass was accidentally disabled (until r212073), r205437
slipped in a use of `auto` that should have been `auto&`.
This fixes PR20188.
llvm-svn: 212201
Based on the support for .req on ARM. The aarch64 variant has to keep track if
the alias register was a vector register (v0-31) or a general purpose or
VFP/Advanced SIMD ([bhsdq]0-31) register.
Patch by Janne Grunau!
llvm-svn: 212161
Otherwise they get freed and the implicit "isa<XYZ>" tests following
turn out badly (at least under sanitizers).
Also corrects the ordering of unordered atomic stores.
llvm-svn: 212136
The argument list vector is never used after it has been passed to the
CallLoweringInfo and moving it to the CallLoweringInfo is cleaner and
pretty much as cheap as keeping a pointer to it.
llvm-svn: 212135
On targets without cmpxchg16b or cmpxchg8b, the borderline atomic
operations were slipping through the gaps.
X86AtomicExpand.cpp was delegating to ISelLowering. Generic
ISelLowering was delegating to X86ISelLowering and X86ISelLowering was
asserting. The correct behaviour is to expand to a libcall, preferably
in generic ISelLowering.
This can be achieved by X86ISelLowering deciding it doesn't want the
faff after all.
llvm-svn: 212134
The logic for expanding atomics that aren't natively supported in
terms of cmpxchg loops is much simpler to express at the IR level. It
also allows the normal optimisations and CodeGen improvements to help
out with atomics, instead of using a limited set of possible
instructions..
rdar://problem/13496295
llvm-svn: 212119
For now I only updated the _alt variants. The main variants are used by
codegen and that will need a bit more work to trigger.
<rdar://problem/17492620>
llvm-svn: 212114
Adding a writemask variant would require a third asm string to be passed to
the template. Generate the AsmString in the template instead.
No change in X86.td.expanded.
llvm-svn: 212113
seh_stackalloc 0 is not representable in Win64 SEH info, so emitting it
is a bug.
Reviewers: rnk
Differential Revision: http://reviews.llvm.org/D4334
Patch by Vadim Chugunov!
llvm-svn: 212081
In r212073 I missed a call of `use_begin()` that assumed the wrong
semantics. It's not clear to me at all what this code does without the
fix, so I'm not sure how to write a testcase.
llvm-svn: 212075
AArch64AddressTypePromotion was doing nothing because it was using the
old semantics of `Use` and `uses()`, when it really wanted to get at the
`users()`.
llvm-svn: 212073
This patch adds support for a new builtin instruction called
__builtin_ia32_rdpmc.
Builtin '__builtin_ia32_rdpmc' is defined as a 'GCC builtin'; on X86, it can
be used to read performance monitoring counters. It takes as input the index
of the performance counter to read, and returns the value of the specified
performance counter as a 64-bit number.
Calls to this new builtin will map to instruction RDPMC.
The index in input to the builtin call is moved to register %ECX. The result
of the builtin call is the value of the specified performance counter (RDPMC
would return that quantity in registers RDX:RAX).
This patch:
- Adds builtin int_x86_rdpmc as a GCCBuiltin;
- Adds a new x86 DAG node called 'RDPMC_DAG';
- Teaches how to lower this new builtin;
- Adds an ISel pattern to select instruction RDPMC;
- Fixes the definition of instruction RDPMC adding %RAX and %RDX as
implicit definitions, and adding %ECX as implicit use;
- Adds a LLVM test to verify that the new builtin is correctly selected.
llvm-svn: 212049
This exception format is not specific to Windows x64. A similar approach is
taken on nearly all architectures. Generalise the name to reflect reality.
This will eventually be used for Windows on ARM data emission as well.
Switch the enum and namespace into an enum class.
llvm-svn: 212000
Rename the routines to reflect the reality that they are more related to call
frame information than to Win64 EH. Although EH is implemented in an intertwined
manner by augmenting with an exception handler and an associated parameter, the
majority of these routines emit information required to unwind the frames. This
also helps identify that these routines are generic for most windows platforms
(they apply equally to nearly all architectures except x86) although the
encoding of the information is architecture dependent.
Unwinding data is emitted via EmitWinCFI* and exception handling information via
EmitWinEH*.
llvm-svn: 211994
lowering for v16i8.
ASan and some bots caught this bug with existing test cases. Fixing it
even fixed a miscompile with one of the test cases. I'm still a bit
suspicious of this test case as I've not taken a proper amount of time
to think about it, but the fix here is strict goodness.
llvm-svn: 211976
These show up really frequently, not the least with actual splats. =] We
lowered these quite badly before. The new code path tries to widen i8
shuffles to i16 shuffles in a splat-like way. There are still some
inefficiencies in our i16 splat logic though, so we aren't really done
here.
Also, for certain patterns (bit of a gather-and-splat) we still
generate pretty silly code, and I've left a fixme for addressing it.
However, I'm not actually worried about this code pattern as much. The
old shuffle lowering generates a 29 instruction monstrosity for it that
should execute much more slowly.
llvm-svn: 211974
lowering.
For maximum irony, I had already discovered this bug, diagnosed it, and
left FIXMEs about it in the test cases. =[ I just failed to go back over
those until after i had reduced a bootstrap miscompile down to a single
TU, stared at the assembly for an hour, and figured out the bug. Again.
Oh well.
llvm-svn: 211955
The address space of the pointer must be global (1) for these intrinsics. There must also be alignment metadata attached to the intrinsic calls, e.g.
%val = tail call i32 @llvm.nvvm.ldu.i.global.i32.p1i32(i32 addrspace(1)* %ptr), !align !0!0 = metadata !{i32 4}
llvm-svn: 211939