[qt/clang/llvm-project]: Summary of bulk changes made

Git repository change summary for qt/clang/llvm-project
Pushed by mirror-service into branch 'upstream/users/lukel97/loop-vectorize/mask.beforefirst'.
Changed from 0000000000000000000000000000000000000000 to 323d58384a6f87739bc2567ee66f083cc5bdcc94
Acknowledgement was received that this change introduces only existing code that has been pushed to another public open source repository.

This change contains the following new commits:

Git commit 323d58384a6f87739bc2567ee66f083cc5bdcc94 by Luke Lau on 27/07/2026 at 15:25..
[VPlan] Use llvm.mask.beforefirst for computing early exit mask

This makes the cost more accurate and improves codegen for RISC-V. The codegen for an early exit loop with side effects then goes from:

    .LBB0_3:                                # %vector.body
                                            # =>This Inner Loop Header: Depth=1
    	vl1r.v	v16, (a3)
    	vsetvli	zero, zero, e8, m1, ta, ma
    	vmseq.vi	v16, v16, 0
    	vfirst.m	a7, v16
    	srli	t0, a7, 63
    	czero.nez	a7, a7, t0
    	czero.eqz	t0, a1, t0
    	or	a7, t0, a7
    	vsetvli	zero, zero, e64, m8, ta, ma
    	vmsltu.vx	v0, v8, a7
    	vle8.v	v17, (a5), v0.t
    	vcpop.m	t1, v16
    	vsetvli	zero, zero, e8, m1, ta, ma
    	vadd.vi	v16, v17, 1
    	vse8.v	v16, (a5), v0.t
    	sub	a4, a4, a2
    	bnez	t1, .LBB0_5

To

    .LBB0_3:                                # %vector.body
                                            # =>This Inner Loop Header: Depth=1
    	vl1r.v	v8, (a4)
    	vmseq.vi	v8, v8, 0
    	vmsbf.m	v0, v8
    	vle8.v	v9, (a5), v0.t
    	vcpop.m	t0, v8
    	vadd.vi	v9, v9, 1
    	vse8.v	v9, (a5), v0.t
    	sub	a2, a2, a3
    	bnez	t0, .LBB0_5

Some AArch64 tests have had their trip count increased since the reduced mask costs made a smaller fits-exactly-into-TC VF be selected. Increasing it keeps the original VF.
https://invent.kde.org/qt/clang/llvm-project/-/commit/323d58384a6f87739bc2567ee66f083cc5bdcc94