sci-misc/stable-diffusion-cpp
Diffusion model(SD,Flux,Wan,Qwen Image,Z-Image,...) inference in pure C/C++
Screenshots
-
stable-diffusion-cpp-9999openblas blis rocm cuda opencl vulkan flexiblas webm webp wmma +amdgpu_targets_gfx908 +amdgpu_targets_gfx90a +amdgpu_targets_gfx942 +amdgpu_targets_gfx1030 +amdgpu_targets_gfx1100 +amdgpu_targets_gfx1101 +amdgpu_targets_gfx1200 +amdgpu_targets_gfx1201 amdgpu_targets_gfx803 amdgpu_targets_gfx900 amdgpu_targets_gfx906 amdgpu_targets_gfx940 amdgpu_targets_gfx941 amdgpu_targets_gfx1010 amdgpu_targets_gfx1011 amdgpu_targets_gfx1012 amdgpu_targets_gfx1031 amdgpu_targets_gfx1102 amdgpu_targets_gfx1103 amdgpu_targets_gfx1150 amdgpu_targets_gfx1151
View 
Download 
Browse License: MIT Overlay: guru -
stable-diffusion-cpp-9999openblas blis flexiblas rocm cuda opencl vulkan webm webp cpu_flags_x86_avx cpu_flags_x86_avx2 cpu_flags_x86_avx512f cpu_flags_x86_avx512vbmi cpu_flags_x86_bmi2 cpu_flags_x86_f16c cpu_flags_x86_fma3 cpu_flags_x86_sse4_2 +amdgpu_targets_gfx908 +amdgpu_targets_gfx90a +amdgpu_targets_gfx942 +amdgpu_targets_gfx950 +amdgpu_targets_gfx1030 +amdgpu_targets_gfx1100 +amdgpu_targets_gfx1101 +amdgpu_targets_gfx1200 +amdgpu_targets_gfx1201 amdgpu_targets_gfx803 amdgpu_targets_gfx900 amdgpu_targets_gfx906 amdgpu_targets_gfx940 amdgpu_targets_gfx941 amdgpu_targets_gfx1010 amdgpu_targets_gfx1011 amdgpu_targets_gfx1012 amdgpu_targets_gfx1031 amdgpu_targets_gfx1102 amdgpu_targets_gfx1103 amdgpu_targets_gfx1150 amdgpu_targets_gfx1151
View Download Browse License: MIT Overlay: stuff -
stable-diffusion-cpp-0_pre841~amd64openblas blis flexiblas rocm cuda opencl vulkan webm webp cpu_flags_x86_avx cpu_flags_x86_avx2 cpu_flags_x86_avx512f cpu_flags_x86_avx512vbmi cpu_flags_x86_bmi2 cpu_flags_x86_f16c cpu_flags_x86_fma3 cpu_flags_x86_sse4_2 +amdgpu_targets_gfx908 +amdgpu_targets_gfx90a +amdgpu_targets_gfx942 +amdgpu_targets_gfx950 +amdgpu_targets_gfx1030 +amdgpu_targets_gfx1100 +amdgpu_targets_gfx1101 +amdgpu_targets_gfx1200 +amdgpu_targets_gfx1201 amdgpu_targets_gfx803 amdgpu_targets_gfx900 amdgpu_targets_gfx906 amdgpu_targets_gfx940 amdgpu_targets_gfx941 amdgpu_targets_gfx1010 amdgpu_targets_gfx1011 amdgpu_targets_gfx1012 amdgpu_targets_gfx1031 amdgpu_targets_gfx1102 amdgpu_targets_gfx1103 amdgpu_targets_gfx1150 amdgpu_targets_gfx1151
View Download Browse License: MIT Overlay: stuff -
stable-diffusion-cpp-0_pre829~amd64openblas blis flexiblas rocm cuda opencl vulkan webm webp cpu_flags_x86_avx cpu_flags_x86_avx2 cpu_flags_x86_avx512f cpu_flags_x86_avx512vbmi cpu_flags_x86_bmi2 cpu_flags_x86_f16c cpu_flags_x86_fma3 cpu_flags_x86_sse4_2 +amdgpu_targets_gfx908 +amdgpu_targets_gfx90a +amdgpu_targets_gfx942 +amdgpu_targets_gfx950 +amdgpu_targets_gfx1030 +amdgpu_targets_gfx1100 +amdgpu_targets_gfx1101 +amdgpu_targets_gfx1200 +amdgpu_targets_gfx1201 amdgpu_targets_gfx803 amdgpu_targets_gfx900 amdgpu_targets_gfx906 amdgpu_targets_gfx940 amdgpu_targets_gfx941 amdgpu_targets_gfx1010 amdgpu_targets_gfx1011 amdgpu_targets_gfx1012 amdgpu_targets_gfx1031 amdgpu_targets_gfx1102 amdgpu_targets_gfx1103 amdgpu_targets_gfx1150 amdgpu_targets_gfx1151
View Download Browse License: MIT Overlay: stuff -
stable-diffusion-cpp-0_pre827~amd64openblas blis flexiblas rocm cuda opencl vulkan webm webp cpu_flags_x86_avx cpu_flags_x86_avx2 cpu_flags_x86_avx512f cpu_flags_x86_avx512vbmi cpu_flags_x86_bmi2 cpu_flags_x86_f16c cpu_flags_x86_fma3 cpu_flags_x86_sse4_2 +amdgpu_targets_gfx908 +amdgpu_targets_gfx90a +amdgpu_targets_gfx942 +amdgpu_targets_gfx950 +amdgpu_targets_gfx1030 +amdgpu_targets_gfx1100 +amdgpu_targets_gfx1101 +amdgpu_targets_gfx1200 +amdgpu_targets_gfx1201 amdgpu_targets_gfx803 amdgpu_targets_gfx900 amdgpu_targets_gfx906 amdgpu_targets_gfx940 amdgpu_targets_gfx941 amdgpu_targets_gfx1010 amdgpu_targets_gfx1011 amdgpu_targets_gfx1012 amdgpu_targets_gfx1031 amdgpu_targets_gfx1102 amdgpu_targets_gfx1103 amdgpu_targets_gfx1150 amdgpu_targets_gfx1151
View Download Browse License: MIT Overlay: stuff
ChangeLog
commit afd611bea21c7c2305b72c1614e1b209b3864e77
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Aug 30 21:58:31 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre820
Superseded by 0_pre841. Keeps the established three-version window on this
snapshot track (827 + 829 + 841).
commit 4e30bd61d1879a99332700cab0d23ba741de4cf3
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Aug 30 21:58:27 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre841
Snapshot bump, tag master-841-6b3edaa. Both pins re-read per the header
comment's instruction: MY_COMMIT 0a565f2 -> 6b3edaa, and the ggml submodule
gitlink MOVED for the first time in several bumps, 8e800cef -> e20c3a14
(read from the tag's tree, not assumed).
That moved gitlink is why the rocWMMA rationale was rewritten rather than
carried over. 0_pre829's note was verified against 8e800cef, so it said
nothing about the tree this ebuild actually ships. Re-verified against the
unpacked e20c3a14: zero GGML_HIP_ROCWMMA_FATTN occurrences, and the only
rocWMMA mention anywhere is a "TODO add support for AMD cards via rocWMMA"
comment in ggml-cuda/lightning-indexer.cu whose actual include is
nvcuda::wmma via <mma.h> -- NVIDIA-only, no rocWMMA dependency. So the flag
drop remains correct on the new pin.
Build interface reassessed from the unpacked trees on both sides: the
stable-diffusion.cpp option() surface is 16 knobs at 829 and 841 with a
byte-identical diff, and the ggml option() surface is identical across the
two gitlinks. No ebuild change needed beyond the two pins.
Build-verified as root with USE=rocm against the migrated ROCm 10.0 stack:
"HIP and hipBLAS found", every compile through hipcc (217 -c invocations, 222
hipcc calls in all), and sd-cli links libamdhip64.so.7, libhipblas.so.3,
libhipblaslt.so.1 and libamd_comgr.so.3 with gfx1150 device code embedded.
Installs exactly sd-cli and sd-server (ggml is static here --
SD_BUILD_SHARED_LIBS=OFF -- so there is no libggml-hip.so to look for).
commit e5296d3072d1394f6dc64084972d332fe8384690
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Aug 30 20:48:08 2026 +0200
sci-misc/stable-diffusion-cpp: drop the dead wmma USE flag
The flag did nothing except pull in a dependency that blocked the ROCm 10.0
stack.
src_configure passed -DGGML_HIP_ROCWMMA_FATTN=$(usex wmma), but the vendored
ggml has no such option. Verified against the unpacked tree at the pinned
GGML_COMMIT 8e800cef, which all three release ebuilds share: zero occurrences
of that string anywhere in the source, the only hits in the entire build being
the -D this ebuild passed and the environment file derived from it. CMake
accepts an unknown -D silently, so the flag never changed a single byte of the
build -- while USE=wmma still pulled >=sci-libs/rocWMMA-$.
That dependency had a real cost. ::gentoo ships rocWMMA only to 7.2.0, and its
subslot is 0/7.2, so anything holding it pinned the whole ROCm closure to the
7.2 stack. Together with dev-util/amdsmi it is why `emerge -uvDN @world` could
not resolve against the installed 10.0 cohort: emerge tried to pull
hipcc-7.2.4:0/7.2 and rocm-smi-7.2.4:0/7.2 back in beside their 0/10.0
counterparts, which share SLOT 0.
sci-misc/llama-cpp already went through this at 0.3.0: upstream removed the
option in b10121 (PR #26046, "HIP: remove rocWMMA FlashAttention") and HIP
FlashAttention now runs through the shared ggml-cuda fattn-mma template
instances. Same reasoning applies here, so the flag, its REQUIRED_USE
constraint, its dependency and its metadata.xml description all go.
No revision bump: the flag was a no-op, so the installed artifact is
bit-identical either way, and nothing on this host had it enabled.
Applied to 9999 as well. Its ggml pin was not re-verified against live master,
and the comment there says so -- but the option is gone upstream, so a live
checkout cannot have it.
commit e79cb142a4706ddf0308d7ceaeaa2a65260ed067
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed Aug 26 02:06:49 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre829
Upstream release tag is master-829-0a565f2, so MY_COMMIT moves to 0a565f2.
GGML_COMMIT deliberately stays at 8e800cef: the ggml submodule gitlink is
byte-identical at master-827 and master-829, so the unchanged pin is
current, not stale. The top-level CMakeLists.txt is also byte-identical
between the two tags, so every -D flag the ebuild sets still exists and no
build-interface adjustment is needed.
commit b93a466208bbf8e43efd9c70f51737f94e6db511
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Aug 20 09:58:18 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre817
commit 3255d0e85e1d976914c52941ebc195927db85c89
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Aug 20 09:58:07 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre827
commit 3e8fa0b533cc2fbb76b8334beb398329d2024408
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Aug 14 06:29:41 2026 +0200
sci-misc/stable-diffusion-cpp: pass compiler executable to CUDAHOSTCXX
cuda_gccdir returns the compiler bindir, while CMake expects CUDAHOSTCXX to name an
executable. Append g++ so CUDA configuration selects the supported compiler
deterministically.
commit e1bb5b792e386c632ce20027bf50b4c4614980d6
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Aug 14 14:43:10 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre813
commit b87aaa4c65a8b12dbb08565262c486636a495426
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Aug 14 14:43:06 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre820
commit ca6850d5c4e186aa1610c09e3ee16645bc6e1525
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Tue Aug 11 23:25:41 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre809
commit 54441d4ee27945b8ef8a08f23eda8211c8ac19af
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Tue Aug 11 23:25:37 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre817
commit b7a266798064a1a76bac54924f32f3b992f8e6fc
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Aug 6 19:09:19 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre805
Retention: keep 0_pre809/0_pre813 (last two) plus the 9999 live ebuild.
commit bbacd130797c171dacfc0b6ad1d08b24c477f17c
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Aug 6 19:09:06 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre813
MY_COMMIT eb7f35c -> bfbef5b (tag master-813-bfbef5b); the ggml submodule
gitlink is unchanged (eced84c...).
commit ab64bcd1e858c8ce410e1019d494c7fb02774e7d
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon Aug 3 09:40:01 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre802
commit 15255d4bc2fc38bc8aa3e719b017d57106c0deaa
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon Aug 3 09:39:55 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre809
commit a2bfde341f7c15bd3d8397e661af3412e6567b18
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Jul 31 09:52:15 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre797
Superseded snapshot; retention keeps the last two master snapshots
(0_pre802, 0_pre805) plus the live ebuild.
commit 0e7d63aa24fdb0ab69a66923255b84c757befec2
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Jul 31 09:52:10 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre805
Upstream master snapshot master-805-e31a86c. The ggml submodule gitlink
is unchanged from 0_pre802 (eced84c), so only MY_COMMIT is refreshed.
commit 755b9983c27f3d758ff0700bfea613ba4ecef9e8
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Jul 30 10:53:59 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre796
Superseded snapshot; retention keeps the last two master snapshots
(0_pre797, 0_pre802) plus the live ebuild.
commit 720f63f317ca905df706bd6b2bf1717a64afa2ba
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Jul 30 10:53:31 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre802
Upstream master snapshot master-802-e92e86f. The ggml submodule gitlink
is unchanged from 0_pre797 (eced84c), so only MY_COMMIT is refreshed.
commit 12e0028e2fe7ae454cfedc1f0a3eaa6632c8f288
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Tue Jul 28 09:46:49 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre795
Retention: 0_pre796 and 0_pre797 are the last two snapshots. Nothing
pins the older one.
commit 200526faeeb5a3b670759b55e5ade542b377a2c2
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Tue Jul 28 09:46:33 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre797
Snapshot bump to master-797-5ef4a75. The ggml submodule gitlink is
unchanged (eced84c8, same distfile), and the 796..797 diff is three
files under examples/ (common.cpp, the server routes, and api.md docs)
with no CMakeLists or option-surface change.
commit 21d42c56fe1b2fd7d13001858e5546ca9838d5ce
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon Jul 27 11:13:33 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre796, drop 0_pre789
Snapshot bump to upstream tag master-796-2d0385b. The ebuild differs
from 0_pre795 only in MY_COMMIT, which together with PV composes the
tag name.
sci-ml/lemonade is the only consumer and depends on this package
unversioned under USE=system-sdcpp, so no pin follows the snapshot
counter.
0_pre789 drops, leaving the two most recent snapshots alongside the
live ebuild.
commit 6cefeccf45f0bbe7f6b4c733bcb2daa19307a020
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sat Jul 25 10:58:15 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre788
commit 630f7bbd9aec4db916a2b015259753bfb8d06b17
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sat Jul 25 10:58:14 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre795
Snapshot bump to master-795-87a0177. ggml submodule gitlink unchanged (eced84c8) and
no cmake option-surface changes over master-789, so a straight MY_COMMIT bump.
commit 28d229ed1fc76b8e1697dde54b24e27d274fc8aa
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Jul 24 11:42:03 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre775, 0_pre778, 0_pre782
commit c0a87149c6627f9fdee5eef58bf18a9fd384e4ce
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Jul 24 11:18:51 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre789
commit 55067c7022017192b3941a6aca2e451caea87d96
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed Jul 22 23:50:39 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre788
commit 8be7041e0a250ae97abb748a51b1c1fd83fdadd4
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Jul 17 15:06:52 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre782
commit 121e74ba44944ea118be84e499f13c08fe13f79c
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed Jul 15 11:09:37 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre769
commit 0f97a68cfe681e37097e1281f2094c7d37d9cf9b
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed Jul 15 11:09:25 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre778
commit 83d5826c0d978626e2da690f5956fa220ee7e6f0
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Jul 12 09:16:56 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre767
commit b421a9ebfb2e9949d37e12a201f04e9d90897d12
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Jul 12 09:16:40 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre775
commit aa99d9adab91b76225f0c4935848b400474f4ecf
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Jul 10 15:20:52 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre769
commit 049141e9c629782d3f0dc4cbe067ea05b8a24601
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed Jul 8 12:01:39 2026 +0200
sci-misc/stable-diffusion-cpp: new package
Diffusion-model (Stable Diffusion, Flux, Wan, Qwen-Image) inference in pure
C/C++ -- the image-generation counterpart to the packaged ggml llama family.
A pinned release (0_pre767) plus a live ebuild, with CPU, BLAS, Vulkan,
OpenCL, CUDA and ROCm backends.
ggml is an upstream git submodule that GitHub's release tarball omits, so the
pinned ebuild fetches leejet's ggml fork as a second distfile and stages it
where add_subdirectory(ggml) expects. The pnpm-driven server web frontend is
disabled for a network-free build; a small patch lets the ebuild embed the
release version the tarball would otherwise report as "unknown".
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Aug 30 21:58:31 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre820
Superseded by 0_pre841. Keeps the established three-version window on this
snapshot track (827 + 829 + 841).
commit 4e30bd61d1879a99332700cab0d23ba741de4cf3
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Aug 30 21:58:27 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre841
Snapshot bump, tag master-841-6b3edaa. Both pins re-read per the header
comment's instruction: MY_COMMIT 0a565f2 -> 6b3edaa, and the ggml submodule
gitlink MOVED for the first time in several bumps, 8e800cef -> e20c3a14
(read from the tag's tree, not assumed).
That moved gitlink is why the rocWMMA rationale was rewritten rather than
carried over. 0_pre829's note was verified against 8e800cef, so it said
nothing about the tree this ebuild actually ships. Re-verified against the
unpacked e20c3a14: zero GGML_HIP_ROCWMMA_FATTN occurrences, and the only
rocWMMA mention anywhere is a "TODO add support for AMD cards via rocWMMA"
comment in ggml-cuda/lightning-indexer.cu whose actual include is
nvcuda::wmma via <mma.h> -- NVIDIA-only, no rocWMMA dependency. So the flag
drop remains correct on the new pin.
Build interface reassessed from the unpacked trees on both sides: the
stable-diffusion.cpp option() surface is 16 knobs at 829 and 841 with a
byte-identical diff, and the ggml option() surface is identical across the
two gitlinks. No ebuild change needed beyond the two pins.
Build-verified as root with USE=rocm against the migrated ROCm 10.0 stack:
"HIP and hipBLAS found", every compile through hipcc (217 -c invocations, 222
hipcc calls in all), and sd-cli links libamdhip64.so.7, libhipblas.so.3,
libhipblaslt.so.1 and libamd_comgr.so.3 with gfx1150 device code embedded.
Installs exactly sd-cli and sd-server (ggml is static here --
SD_BUILD_SHARED_LIBS=OFF -- so there is no libggml-hip.so to look for).
commit e5296d3072d1394f6dc64084972d332fe8384690
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Aug 30 20:48:08 2026 +0200
sci-misc/stable-diffusion-cpp: drop the dead wmma USE flag
The flag did nothing except pull in a dependency that blocked the ROCm 10.0
stack.
src_configure passed -DGGML_HIP_ROCWMMA_FATTN=$(usex wmma), but the vendored
ggml has no such option. Verified against the unpacked tree at the pinned
GGML_COMMIT 8e800cef, which all three release ebuilds share: zero occurrences
of that string anywhere in the source, the only hits in the entire build being
the -D this ebuild passed and the environment file derived from it. CMake
accepts an unknown -D silently, so the flag never changed a single byte of the
build -- while USE=wmma still pulled >=sci-libs/rocWMMA-$.
That dependency had a real cost. ::gentoo ships rocWMMA only to 7.2.0, and its
subslot is 0/7.2, so anything holding it pinned the whole ROCm closure to the
7.2 stack. Together with dev-util/amdsmi it is why `emerge -uvDN @world` could
not resolve against the installed 10.0 cohort: emerge tried to pull
hipcc-7.2.4:0/7.2 and rocm-smi-7.2.4:0/7.2 back in beside their 0/10.0
counterparts, which share SLOT 0.
sci-misc/llama-cpp already went through this at 0.3.0: upstream removed the
option in b10121 (PR #26046, "HIP: remove rocWMMA FlashAttention") and HIP
FlashAttention now runs through the shared ggml-cuda fattn-mma template
instances. Same reasoning applies here, so the flag, its REQUIRED_USE
constraint, its dependency and its metadata.xml description all go.
No revision bump: the flag was a no-op, so the installed artifact is
bit-identical either way, and nothing on this host had it enabled.
Applied to 9999 as well. Its ggml pin was not re-verified against live master,
and the comment there says so -- but the option is gone upstream, so a live
checkout cannot have it.
commit e79cb142a4706ddf0308d7ceaeaa2a65260ed067
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed Aug 26 02:06:49 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre829
Upstream release tag is master-829-0a565f2, so MY_COMMIT moves to 0a565f2.
GGML_COMMIT deliberately stays at 8e800cef: the ggml submodule gitlink is
byte-identical at master-827 and master-829, so the unchanged pin is
current, not stale. The top-level CMakeLists.txt is also byte-identical
between the two tags, so every -D flag the ebuild sets still exists and no
build-interface adjustment is needed.
commit b93a466208bbf8e43efd9c70f51737f94e6db511
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Aug 20 09:58:18 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre817
commit 3255d0e85e1d976914c52941ebc195927db85c89
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Aug 20 09:58:07 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre827
commit 3e8fa0b533cc2fbb76b8334beb398329d2024408
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Aug 14 06:29:41 2026 +0200
sci-misc/stable-diffusion-cpp: pass compiler executable to CUDAHOSTCXX
cuda_gccdir returns the compiler bindir, while CMake expects CUDAHOSTCXX to name an
executable. Append g++ so CUDA configuration selects the supported compiler
deterministically.
commit e1bb5b792e386c632ce20027bf50b4c4614980d6
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Aug 14 14:43:10 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre813
commit b87aaa4c65a8b12dbb08565262c486636a495426
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Aug 14 14:43:06 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre820
commit ca6850d5c4e186aa1610c09e3ee16645bc6e1525
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Tue Aug 11 23:25:41 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre809
commit 54441d4ee27945b8ef8a08f23eda8211c8ac19af
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Tue Aug 11 23:25:37 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre817
commit b7a266798064a1a76bac54924f32f3b992f8e6fc
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Aug 6 19:09:19 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre805
Retention: keep 0_pre809/0_pre813 (last two) plus the 9999 live ebuild.
commit bbacd130797c171dacfc0b6ad1d08b24c477f17c
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Aug 6 19:09:06 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre813
MY_COMMIT eb7f35c -> bfbef5b (tag master-813-bfbef5b); the ggml submodule
gitlink is unchanged (eced84c...).
commit ab64bcd1e858c8ce410e1019d494c7fb02774e7d
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon Aug 3 09:40:01 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre802
commit 15255d4bc2fc38bc8aa3e719b017d57106c0deaa
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon Aug 3 09:39:55 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre809
commit a2bfde341f7c15bd3d8397e661af3412e6567b18
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Jul 31 09:52:15 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre797
Superseded snapshot; retention keeps the last two master snapshots
(0_pre802, 0_pre805) plus the live ebuild.
commit 0e7d63aa24fdb0ab69a66923255b84c757befec2
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Jul 31 09:52:10 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre805
Upstream master snapshot master-805-e31a86c. The ggml submodule gitlink
is unchanged from 0_pre802 (eced84c), so only MY_COMMIT is refreshed.
commit 755b9983c27f3d758ff0700bfea613ba4ecef9e8
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Jul 30 10:53:59 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre796
Superseded snapshot; retention keeps the last two master snapshots
(0_pre797, 0_pre802) plus the live ebuild.
commit 720f63f317ca905df706bd6b2bf1717a64afa2ba
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Jul 30 10:53:31 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre802
Upstream master snapshot master-802-e92e86f. The ggml submodule gitlink
is unchanged from 0_pre797 (eced84c), so only MY_COMMIT is refreshed.
commit 12e0028e2fe7ae454cfedc1f0a3eaa6632c8f288
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Tue Jul 28 09:46:49 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre795
Retention: 0_pre796 and 0_pre797 are the last two snapshots. Nothing
pins the older one.
commit 200526faeeb5a3b670759b55e5ade542b377a2c2
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Tue Jul 28 09:46:33 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre797
Snapshot bump to master-797-5ef4a75. The ggml submodule gitlink is
unchanged (eced84c8, same distfile), and the 796..797 diff is three
files under examples/ (common.cpp, the server routes, and api.md docs)
with no CMakeLists or option-surface change.
commit 21d42c56fe1b2fd7d13001858e5546ca9838d5ce
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon Jul 27 11:13:33 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre796, drop 0_pre789
Snapshot bump to upstream tag master-796-2d0385b. The ebuild differs
from 0_pre795 only in MY_COMMIT, which together with PV composes the
tag name.
sci-ml/lemonade is the only consumer and depends on this package
unversioned under USE=system-sdcpp, so no pin follows the snapshot
counter.
0_pre789 drops, leaving the two most recent snapshots alongside the
live ebuild.
commit 6cefeccf45f0bbe7f6b4c733bcb2daa19307a020
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sat Jul 25 10:58:15 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre788
commit 630f7bbd9aec4db916a2b015259753bfb8d06b17
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sat Jul 25 10:58:14 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre795
Snapshot bump to master-795-87a0177. ggml submodule gitlink unchanged (eced84c8) and
no cmake option-surface changes over master-789, so a straight MY_COMMIT bump.
commit 28d229ed1fc76b8e1697dde54b24e27d274fc8aa
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Jul 24 11:42:03 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre775, 0_pre778, 0_pre782
commit c0a87149c6627f9fdee5eef58bf18a9fd384e4ce
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Jul 24 11:18:51 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre789
commit 55067c7022017192b3941a6aca2e451caea87d96
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed Jul 22 23:50:39 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre788
commit 8be7041e0a250ae97abb748a51b1c1fd83fdadd4
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Jul 17 15:06:52 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre782
commit 121e74ba44944ea118be84e499f13c08fe13f79c
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed Jul 15 11:09:37 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre769
commit 0f97a68cfe681e37097e1281f2094c7d37d9cf9b
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed Jul 15 11:09:25 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre778
commit 83d5826c0d978626e2da690f5956fa220ee7e6f0
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Jul 12 09:16:56 2026 +0200
sci-misc/stable-diffusion-cpp: drop 0_pre767
commit b421a9ebfb2e9949d37e12a201f04e9d90897d12
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Jul 12 09:16:40 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre775
commit aa99d9adab91b76225f0c4935848b400474f4ecf
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Jul 10 15:20:52 2026 +0200
sci-misc/stable-diffusion-cpp: add 0_pre769
commit 049141e9c629782d3f0dc4cbe067ea05b8a24601
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed Jul 8 12:01:39 2026 +0200
sci-misc/stable-diffusion-cpp: new package
Diffusion-model (Stable Diffusion, Flux, Wan, Qwen-Image) inference in pure
C/C++ -- the image-generation counterpart to the packaged ggml llama family.
A pinned release (0_pre767) plus a live ebuild, with CPU, BLAS, Vulkan,
OpenCL, CUDA and ROCm backends.
ggml is an upstream git submodule that GitHub's release tarball omits, so the
pinned ebuild fetches leejet's ggml fork as a second distfile and stages it
where add_subdirectory(ggml) expects. The pnpm-driven server web frontend is
disabled for a network-free build; a small patch lets the ebuild embed the
release version the tarball would otherwise report as "unknown".