The baseline `allowsMemoryAccess()` is wrong for X86.
It assumes that aligned memory operations are always allowed,
but that is not true.
For example, We can not perform a 32-byte aligned non-temporal load
of a 32-byte vector, without AVX2 that is, yet `allowsMemoryAccess()`
will say it is allowed, so we may end up merging non-temporal loads,
only to split them up to legalize them, and here we go again.
NOTE: the test changes here are superfluous. The main effect is that without this change,
in D141777, we'd get stuck endlessly merging and splitting non-temporal stores.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D141776
This change adjusts the cost modeling used when the target does not have a schedule model with individual instruction latencies. After this change, we use the default latency information available from TargetSchedule. The default latency information essentially ends up treating most instructions as latency 1, with a few "expensive" ones getting a higher cost.
Previously, we unconditionally applied the first legal pattern - without any consideration of profitability. As a result, this change both prevents some patterns being applied, and changes which patterns are exercised. (i.e. previously the first pattern was applied, afterwards, maybe the second one is because the first wasn't profitable.)
The motivation here is two fold.
First, this brings the default behavior in line with the behavior when -mcpu or -mtune is specified. This improves test coverage, and generally makes it less likely we will have bad surprises when providing more information to the compiler.
Second, this enables some reassociation for ILP by default. Despite being unconditionally enabled, the prior code tended to "reassociate" repeatedly through an entire chain and simply moving the first operand to the end. The result was still a serial chain, just a different one. With this change, one of the intermediate transforms is unprofitable and we end up with a partially flattened tree.
Note that the resulting code diffs show significant room for improvement in the basic algorithm. I am intentionally excluding those from this patch.
For the test diffs, I don't seen any concerning regressions. I took a fairly close look at the RISCV ones, but only skimmed the x86 (particularly vector x86) changes.
Differential Revision: https://reviews.llvm.org/D141017
The first attempt missed changing test files for tools
(update_llc_test_checks.py).
Original commit message:
This implements the main suggested change from issue #56498.
Using the shorter (non-extending) instruction with only
-Oz ("minsize") rather than -Os ("optsize") is left as a
possible follow-up.
As noted in the bug report, the zero-extending load may have
shorter latency/better throughput across a wide range of x86
micro-arches, and it avoids a potential false dependency.
The cost is an extra instruction byte.
This could cause perf ups and downs from secondary effects,
but I don't think it is possible to account for those in
advance, and that will likely also depend on exact micro-arch.
This does bring LLVM x86 codegen more in line with existing
gcc codegen, so if problems are exposed they are more likely
to occur for both compilers.
Differential Revision: https://reviews.llvm.org/D129775
This implements the main suggested change from issue #56498.
Using the shorter (non-extending) instruction with only
-Oz ("minsize") rather than -Os ("optsize") is left as a
possible follow-up.
As noted in the bug report, the zero-extending load may have
shorter latency/better throughput across a wide range of x86
micro-arches, and it avoids a potential false dependency.
The cost is an extra instruction byte.
This could cause perf ups and downs from secondary effects,
but I don't think it is possible to account for those in
advance, and that will likely also depend on exact micro-arch.
This does bring LLVM x86 codegen more in line with existing
gcc codegen, so if problems are exposed they are more likely
to occur for both compilers.
Differential Revision: https://reviews.llvm.org/D129775
We already use combineAddOrSubToADCOrSBB to fold extended EFLAGS results into ISD::ADD/SUB ops as X86ISD::ADC/SBB carry ops.
This patch extends this to also try to fold EFLAGS results with X86ISD::ADD/SUB ops
Differential Revision: https://reviews.llvm.org/D125642
As suggested on PR35908, if we are adding/subtracting an extracted bit, attempt to use BT instead to fold the op and use a ADC/SBB op.
Reapply with extra type legality checks - LowerAndToBT was originally only used during lowering, now that it can occur earlier we might encounter illegal types that we can either promote to i32 or just bail.
Differential Revision: https://reviews.llvm.org/D122084
Ensure we don't attempt to fold to illegal types to ADC/SBB nodes.
After D122084 its possible for ADD(X,AND(SRL(Y,Z),1) patterns to be matched before type legalization.
As suggested on PR35908, if we are adding/subtracting an extracted bit, attempt to use BT instead to fold the op and use a ADC/SBB op.
Differential Revision: https://reviews.llvm.org/D122084
