summaryrefslogtreecommitdiff
path: root/llama.cpp/ggml/src/ggml-cpu/spacemit/ime_kernels.h
diff options
context:
space:
mode:
authorMitja Felicijan <mitja.felicijan@gmail.com>2026-02-12 20:57:17 +0100
committerMitja Felicijan <mitja.felicijan@gmail.com>2026-02-12 20:57:17 +0100
commitb333b06772c89d96aacb5490d6a219fba7c09cc6 (patch)
tree211df60083a5946baa2ed61d33d8121b7e251b06 /llama.cpp/ggml/src/ggml-cpu/spacemit/ime_kernels.h
downloadllmnpc-b333b06772c89d96aacb5490d6a219fba7c09cc6.tar.gz
Engage!
Diffstat (limited to 'llama.cpp/ggml/src/ggml-cpu/spacemit/ime_kernels.h')
-rw-r--r--llama.cpp/ggml/src/ggml-cpu/spacemit/ime_kernels.h26
1 files changed, 26 insertions, 0 deletions
diff --git a/llama.cpp/ggml/src/ggml-cpu/spacemit/ime_kernels.h b/llama.cpp/ggml/src/ggml-cpu/spacemit/ime_kernels.h
new file mode 100644
index 0000000..7570634
--- /dev/null
+++ b/llama.cpp/ggml/src/ggml-cpu/spacemit/ime_kernels.h
@@ -0,0 +1,26 @@
+#pragma once
+
+#include <cstddef>
+
+namespace sqnbitgemm_spacemit_ime {
+namespace ime1 {
+size_t gemm_kernel_i8i4(size_t blk_len,
+ const std::byte * quant_a_ptr,
+ const std::byte * quant_b_data,
+ const float * quant_b_scale,
+ const std::byte * quant_b_zp,
+ float * c_ptr,
+ size_t count_m,
+ size_t count_n,
+ size_t count_k,
+ size_t block_count_k,
+ size_t ldc,
+ const float * bias,
+ const size_t scale_stride);
+
+void quantize_a_row_i8(size_t blk_len, const float * a_ptr, size_t count_k, std::byte * quant_a_ptr);
+
+void quantize_a_4row_i8(size_t blk_len, const float * a_ptr, size_t count_k, std::byte * quant_a_ptr);
+
+} // namespace ime1
+} // namespace sqnbitgemm_spacemit_ime