Skip to content

ggml-quants: Provide ggml_vqtbl1q_u8 for 64bit compatibility #5711

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Feb 25, 2024
Merged

ggml-quants: Provide ggml_vqtbl1q_u8 for 64bit compatibility #5711

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
a3ee7a1
[ggml-quants] Provide ggml_vqtbl1q_u8 for 64bit compatibility
rgryta Feb 25, 2024
f5006c1
[android-example] Remove abi filter after arm v7a fix
rgryta Feb 25, 2024
cc9288b
[github-workflows] Do not skip Android armeabi-v7a build
rgryta Feb 25, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 1 addition & 2 deletions .github/workflows/build.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -669,8 +669,7 @@ jobs:
run: |
cd examples/llama.android

# Skip armeabi-v7a for now (https://github.com/llvm/llvm-project/issues/65820).
./gradlew build --no-daemon -Pskip-armeabi-v7a
./gradlew build --no-daemon

# freeBSD-latest:
# runs-on: macos-12
Expand Down
8 changes: 2 additions & 6 deletions examples/llama.android/app/build.gradle.kts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,12 +21,8 @@ android {
useSupportLibrary = true
}
ndk {
// Workaround for https://github.com/llvm/llvm-project/issues/65820
// affecting armeabi-v7a. Skip armeabi-v7a when invoked with
// -Pskip-armeabi-v7a (e.g., ./gradlew build -Pskip-armeabi-v7a).
if (project.hasProperty("skip-armeabi-v7a")) {
abiFilters += listOf("arm64-v8a", "x86_64", "x86")
}
// Add NDK properties if wanted, e.g.
// abiFilters += listOf("arm64-v8a")
}
externalNativeBuild {
cmake {
Expand Down
33 changes: 29 additions & 4 deletions ggml-quants.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -462,6 +462,30 @@ inline static int8x16_t ggml_vqtbl1q_s8(int8x16_t a, uint8x16_t b) {
return res;
}

// NOTE: not tested
inline static int8x16_t ggml_vqtbl1q_u8(uint8x16_t a, uint8x16_t b) {
int8x16_t res;

res[ 0] = a[b[ 0]];
res[ 1] = a[b[ 1]];
res[ 2] = a[b[ 2]];
res[ 3] = a[b[ 3]];
res[ 4] = a[b[ 4]];
res[ 5] = a[b[ 5]];
res[ 6] = a[b[ 6]];
res[ 7] = a[b[ 7]];
res[ 8] = a[b[ 8]];
res[ 9] = a[b[ 9]];
res[10] = a[b[10]];
res[11] = a[b[11]];
res[12] = a[b[12]];
res[13] = a[b[13]];
res[14] = a[b[14]];
res[15] = a[b[15]];

return res;
}

#else

#define ggml_int16x8x2_t int16x8x2_t
Expand All @@ -476,6 +500,7 @@ inline static int8x16_t ggml_vqtbl1q_s8(int8x16_t a, uint8x16_t b) {
#define ggml_vld1q_s8_x2 vld1q_s8_x2
#define ggml_vld1q_s8_x4 vld1q_s8_x4
#define ggml_vqtbl1q_s8 vqtbl1q_s8
#define ggml_vqtbl1q_u8 vqtbl1q_u8

#endif

Expand Down Expand Up @@ -9488,17 +9513,17 @@ void ggml_vec_dot_iq3_s_q8_K (int n, float * GGML_RESTRICT s, size_t bs, const v
qs += 16;

vs.val[0] = vreinterpretq_u8_u32(vdupq_n_u32(signs[0] | (signs[1] << 16)));
vs.val[1] = vandq_u8(vqtbl1q_u8(vs.val[0], mask1.val[1]), mask2);
vs.val[0] = vandq_u8(vqtbl1q_u8(vs.val[0], mask1.val[0]), mask2);
vs.val[1] = vandq_u8(ggml_vqtbl1q_u8(vs.val[0], mask1.val[1]), mask2);
vs.val[0] = vandq_u8(ggml_vqtbl1q_u8(vs.val[0], mask1.val[0]), mask2);
vs.val[0] = vceqq_u8(vs.val[0], mask2);
vs.val[1] = vceqq_u8(vs.val[1], mask2);

q3s.val[0] = vsubq_s8(vreinterpretq_s8_u8(veorq_u8(vs.val[0], vreinterpretq_u8_u32(aux32x4_0))), vreinterpretq_s8_u8(vs.val[0]));
q3s.val[1] = vsubq_s8(vreinterpretq_s8_u8(veorq_u8(vs.val[1], vreinterpretq_u8_u32(aux32x4_1))), vreinterpretq_s8_u8(vs.val[1]));

vs.val[0] = vreinterpretq_u8_u32(vdupq_n_u32(signs[2] | (signs[3] << 16)));
vs.val[1] = vandq_u8(vqtbl1q_u8(vs.val[0], mask1.val[1]), mask2);
vs.val[0] = vandq_u8(vqtbl1q_u8(vs.val[0], mask1.val[0]), mask2);
vs.val[1] = vandq_u8(ggml_vqtbl1q_u8(vs.val[0], mask1.val[1]), mask2);
vs.val[0] = vandq_u8(ggml_vqtbl1q_u8(vs.val[0], mask1.val[0]), mask2);
vs.val[0] = vceqq_u8(vs.val[0], mask2);
vs.val[1] = vceqq_u8(vs.val[1], mask2);

Expand Down