Skip to content

tcg: improve logical op support for ppc hosts

Nathan Froyd qemu-devel 1 version In review

Get series with b4
b4 am [email protected]

Versions newest first

  1. v1 Latest10 patches 3 repliesIn review

Patches in v1 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/10Add TCG ops for various logical operationstcg/tcg-opc.h+32 -00 replies
2/10Specialize tcg_gen_not_i64 for 32-bit targetstcg/tcg-op.h+3 -01 reply
3/10Implement specialized andc_i{32,64}tcg/tcg-op.h+11 -00 replies
4/10Implement specialized eqv_{i32,i64}tcg/tcg-op.h+11 -00 replies
5/10Implement specialized nand_i{32,64}tcg/tcg-op.h+11 -00 replies
6/10Implement specialized nor_i{32,64}tcg/tcg-op.h+11 -00 replies
7/10Implement specialized orc_i{32,64}tcg/tcg-op.h+11 -00 replies
8/10Implement new logical instructions for ppctcg/ppc/tcg-target.c tcg/ppc/tcg-target.h+31 -00 replies
9/10Implement new logical instructions for ppc64tcg/ppc64/tcg-target.c tcg/ppc64/tcg-target.h+46 -00 replies
10/10Remove r0 from the allocation pool on ppc/ppc64tcg/ppc/tcg-target.c tcg/ppc64/tcg-target.c+0 -21 reply

Reviews trailers per patch

No Reviewed-by, Acked-by or Tested-by trailers for v1 yet.

Read the v1 thread

Cover letter [Qemu-devel] [PATCH 0/10] tcg: improve logical op support for ppc hosts

This patch series implements the necessary support for native TCG implementations the logical operations andc, orc, eqv, nor, and nand (as opposed to the synthesized implementations currently used by TCG). It also implements backend support for said operations in the ppc and ppc64 backends.

I have tested it sparsely on a ppc64 machine I have access to; Linux user-mode emulation seems to be broken on ppc hosts. I could not run a given binary (something as simple as ls or complicated as gcc) with ppc{,64}-linux-user targeted QEMU. Prior to the guest application crashing, however, I could see in qemu.log that several andc/orc/eqv/nand/nor instructions were being used and being translated correctly.

The patch series also includes one bugfix at the very end, which removes r0 from the allocation pool. The logic of the patch seems straightforward enough to me, but I confess to not being a TCG expert, so if TCG somehow avoids generating instructions with r0 in the "wrong" places, I'd be happy to hear of an explanation why.

-Nathan