| Message ID | 20230316094102.2628727-4-alex@linutronix.de |
|---|---|
| State | New |
| Headers | show |
| Series | [1/6] runqemu: direct mesa to use its own drivers, rather than ones provided by host distro | expand |
On 16 Mar 2023, at 09:41, Alexander Kanavin via lists.openembedded.org <alex.kanavin=gmail.com@lists.openembedded.org> wrote:
> +PACKAGECONFIG:class-native = "${@bb.utils.contains('DISTRO_FEATURES', 'opengl', 'libllvm', '', d)}"
Currently the llvm drivers in Mesa are the only known user of libllvm in oe-core, but I don’t like making that a hard assertion in the recipe and the subsequent native/target specific changes to the recipe.
What’s the build impact on llvm-native of enabling/disabling libllvm and why shouldn’t we just do it always?
Ross
On Mon, 20 Mar 2023 at 17:20, Ross Burton <Ross.Burton@arm.com> wrote: > Currently the llvm drivers in Mesa are the only known user of libllvm in oe-core, but I don’t like making that a hard assertion in the recipe and the subsequent native/target specific changes to the recipe. > > What’s the build impact on llvm-native of enabling/disabling libllvm and why shouldn’t we just do it always? I think that's answered elsewhere in the series. It's still 4 times quicker than rust-native. Maybe we should just embrace this brave new world of toolchains with heavy build times. Alex
diff --git a/meta/recipes-devtools/llvm/llvm_git.bb b/meta/recipes-devtools/llvm/llvm_git.bb index b4e983d2d0..f133653903 100644 --- a/meta/recipes-devtools/llvm/llvm_git.bb +++ b/meta/recipes-devtools/llvm/llvm_git.bb @@ -57,9 +57,11 @@ def get_llvm_arch(bb, d, arch_var): def get_llvm_host_arch(bb, d): return get_llvm_arch(bb, d, 'HOST_ARCH') -PACKAGECONFIG ??= "" +PACKAGECONFIG ??= "libllvm" +PACKAGECONFIG:class-native = "${@bb.utils.contains('DISTRO_FEATURES', 'opengl', 'libllvm', '', d)}" # if optviewer OFF, force the modules to be not found or the ones on the host would be found PACKAGECONFIG[optviewer] = ",-DPY_PYGMENTS_FOUND=OFF -DPY_PYGMENTS_LEXERS_C_CPP_FOUND=OFF -DPY_YAML_FOUND=OFF,python3-pygments python3-pyyaml,python3-pygments python3-pyyaml" +PACKAGECONFIG[libllvm] = "" # # Default to build all OE-Core supported target arches (user overridable). @@ -102,14 +104,15 @@ do_compile:prepend:class-target() { } do_compile() { + if ${@bb.utils.contains('PACKAGECONFIG', 'libllvm', 'true', 'false', d)}; then ninja -v ${PARALLEL_MAKE} -} - -do_compile:class-native() { + else ninja -v ${PARALLEL_MAKE} llvm-config llvm-tblgen + fi } do_install() { + if ${@bb.utils.contains('PACKAGECONFIG', 'libllvm', 'true', 'false', d)}; then DESTDIR=${D} ninja -v install # llvm harcodes usr/lib as install path, so this corrects it to actual libdir @@ -121,9 +124,10 @@ do_install() { # reproducibility sed -i -e 's,${WORKDIR},,g' ${D}/${libdir}/cmake/llvm/LLVMConfig.cmake + fi } -do_install:class-native() { +do_install:append:class-native() { install -D -m 0755 ${B}/bin/llvm-tblgen ${D}${bindir}/llvm-tblgen${PV} install -D -m 0755 ${B}/bin/llvm-config ${D}${bindir}/llvm-config${PV} ln -sf llvm-config${PV} ${D}${bindir}/llvm-config
Also, enable that, if opengl is in native DISTRO_FEATURES: this allows mesa-native to build drivers that rely on libllvm, particularly llvmpipe, which is a (sort of) accelerated software renderer that is the only option on build hosts without dedicated or supported GPUs. Signed-off-by: Alexander Kanavin <alex@linutronix.de> --- meta/recipes-devtools/llvm/llvm_git.bb | 14 +++++++++----- 1 file changed, 9 insertions(+), 5 deletions(-)