upstream-archive/stable-diffusion.cpp
1
0
Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2026-09-26 07:57:43 -05:00
Code Issues Packages Projects Releases Wiki Activity

feat: restore legacy fp8 handling when building with upstream ggml (#2001)

Browse Source
This commit is contained in:
Wagner Bruna
2026-09-21 00:48:30 +08:00
committed by GitHub
parent 137f7409bb
commit 008ca5b492
8 changed files with 106 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/model_loader.cpp
+70
View File
@@ -35,6 +35,66 @@
/*================================================= Preprocess ==================================================*/
#ifdef SD_USE_UPSTREAM_GGML
uint16_t f8_e4m3_to_f16(uint8_t f8) {
const uint32_t exponent_bias = 7;
if (f8 == 0xff) {
return ggml_fp32_to_fp16(-NAN);
} else if (f8 == 0x7f) {
return ggml_fp32_to_fp16(NAN);
}
uint32_t sign = f8 & 0x80;
uint32_t exponent = (f8 & 0x78) >> 3;
uint32_t mantissa = f8 & 0x07;
uint32_t result = sign << 24;
if (exponent == 0) {
if (mantissa > 0) {
exponent = 0x7f - exponent_bias;
// yes, 2 times
if ((mantissa & 0x04) == 0) {
mantissa &= 0x03;
mantissa <<= 1;
exponent -= 1;
}
if ((mantissa & 0x04) == 0) {
mantissa &= 0x03;
mantissa <<= 1;
exponent -= 1;
}
result |= (mantissa & 0x03) << 21;
result |= exponent << 23;
}
} else {
result |= mantissa << 20;
exponent += 0x7f - exponent_bias;
result |= exponent << 23;
}
return ggml_fp32_to_fp16(*reinterpret_cast<const float*>(&result));
}
uint16_t f8_e5m2_to_f16(uint8_t fp8) {
return static_cast<uint16_t>(fp8) << 8;
}
void f8_e4m3_to_f16_vec(uint8_t* src, uint16_t* dst, int64_t n) {
// support inplace op
for (int64_t i = n - 1; i >= 0; i--) {
dst[i] = f8_e4m3_to_f16(src[i]);
}
}
void f8_e5m2_to_f16_vec(uint8_t* src, uint16_t* dst, int64_t n) {
// support inplace op
for (int64_t i = n - 1; i >= 0; i--) {
dst[i] = f8_e5m2_to_f16(src[i]);
}
}
#endif
void f64_to_f32_vec(double* src, float* dst, int64_t n) {
// support inplace op
for (int64_t i = 0; i < n; i++) {
@@ -861,6 +921,10 @@ std::vector<MmapTensorStore> ModelLoader::mmap_tensors(std::map<std::string, ggm
if (tensor_storage.is_f64 ||
tensor_storage.is_i64 ||
#ifdef SD_USE_UPSTREAM_GGML
tensor_storage.is_f8_e4m3 ||
tensor_storage.is_f8_e5m2 ||
#endif
tensor_storage.type != dst_tensor->type) {
continue;
}
@@ -1150,6 +1214,12 @@ bool ModelLoader::load_tensors(on_new_tensor_cb_t on_new_tensor_cb,
f64_to_f32_vec((double*)read_buf, (float*)target_buf, tensor_storage.nelements());
} else if (tensor_storage.is_i64) {
i64_to_i32_vec((int64_t*)read_buf, (int32_t*)target_buf, tensor_storage.nelements());
#ifdef SD_USE_UPSTREAM_GGML
} else if (tensor_storage.is_f8_e4m3) {
f8_e4m3_to_f16_vec((uint8_t*)read_buf, (uint16_t*)target_buf, tensor_storage.nelements());
} else if (tensor_storage.is_f8_e5m2) {
f8_e5m2_to_f16_vec((uint8_t*)read_buf, (uint16_t*)target_buf, tensor_storage.nelements());
#endif
}
if (tensor_storage.type != dst_tensor->type) {
if (convert_buf == nullptr) {