ggml: riscv: add riscv spacemit backend

cmake/riscv64-spacemit-linux-gnu-gcc.cmake

@@ -0,0 +1,31 @@
+# Copyright (c) 2023 SpacemiT. All rights reserved.
+set(CMAKE_SYSTEM_NAME Linux)
+SET(CMAKE_SYSTEM_PROCESSOR riscv64)
+set(CMAKE_SYSTEM_VERSION 1)
+
+if(CMAKE_HOST_SYSTEM_PROCESSOR MATCHES "^(riscv)")
+message(STATUS "HOST SYSTEM ${CMAKE_HOST_SYSTEM_PROCESSOR}")
+else()
+set(GNU_MACHINE riscv64-unknown-linux-gnu CACHE STRING "GNU compiler triple")
+if(DEFINED ENV{RISCV_ROOT_PATH})
+    file(TO_CMAKE_PATH $ENV{RISCV_ROOT_PATH} RISCV_ROOT_PATH)
+else()
+    message(FATAL_ERROR "RISCV_ROOT_PATH env must be defined")
+endif()
+
+set(RISCV_ROOT_PATH ${RISCV_ROOT_PATH} CACHE STRING "root path to riscv toolchain")
+set(CMAKE_C_COMPILER ${RISCV_ROOT_PATH}/bin/riscv64-unknown-linux-gnu-gcc)
+set(CMAKE_CXX_COMPILER ${RISCV_ROOT_PATH}/bin/riscv64-unknown-linux-gnu-g++)
+set(CMAKE_STRIP ${RISCV_ROOT_PATH}/bin/riscv64-unknown-linux-gnu-strip)
+set(CMAKE_FIND_ROOT_PATH "${RISCV_ROOT_PATH}/riscv64-unknown-linux-gnu")
+set(CMAKE_SYSROOT "${RISCV_ROOT_PATH}/sysroot")
+endif()
+
+set(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
+set(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY ONLY)
+set(CMAKE_FIND_ROOT_PATH_MODE_INCLUDE ONLY)
+set(CMAKE_FIND_ROOT_PATH_MODE_PACKAGE ONLY)
+set(CMAKE_C_FLAGS "-march=rv64gcv_zfh_zba_zicbop -mabi=lp64d ${CMAKE_C_FLAGS}")
+set(CMAKE_CXX_FLAGS "-march=rv64gcv_zfh_zba_zicbop -mabi=lp64d ${CXX_FLAGS}")
+set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -latomic")
+add_definitions(-D__fp16=_Float16)

docs/build-riscv64-spacemit.md

@@ -0,0 +1,87 @@
+> [!IMPORTANT]
+> This build documentation is specific only to RISC-V SpacemiT SOCs.
+
+## Build llama.cpp locally (for riscv64)
+
+1. Prepare Toolchain For RISCV
+~~~
+wget https://archive.spacemit.com/toolchain/spacemit-toolchain-linux-glibc-x86_64-v1.1.2.tar.xz
+~~~
+
+2. Build
+Below is the build script: it requires utilizing RISC-V vector instructions for acceleration. Ensure the `GGML_CPU_RISCV64_SPACEMIT` compilation option is enabled. The currently supported optimization version is `RISCV64_SPACEMIT_IME1`, corresponding to the `RISCV64_SPACEMIT_IME_SPEC` compilation option. Compiler configurations are defined in the `riscv64-spacemit-linux-gnu-gcc.cmake` file. Please ensure you have installed the RISC-V compiler and set the environment variable via `export RISCV_ROOT_PATH={your_compiler_path}`.
+```bash
+
+cmake -B build-riscv64-spacemit \
+    -DCMAKE_BUILD_TYPE=Release \
+    -DGGML_CPU_RISCV64_SPACEMIT=ON \
+    -DLLAMA_CURL=OFF \
+    -DGGML_RV_ZFH=ON \
+    -DRISCV64_SPACEMIT_IME_SPEC=RISCV64_SPACEMIT_IME1 \
+    -DCMAKE_TOOLCHAIN_FILE=${PWD}/cmake/riscv64-spacemit-linux-gnu-gcc.cmake \
+    -DCMAKE_INSTALL_PREFIX=build-riscv64-spacemit/installed
+
+cmake --build build-riscv64-spacemit --parallel $(nproc) --config Release
+
+pushd build-riscv64-spacemit
+make install
+popd
+```
+
+## Simulation
+You can use QEMU to perform emulation on non-RISC-V architectures.
+
+1. Download QEMU
+~~~
+wget https://archive.spacemit.com/spacemit-ai/qemu/jdsk-qemu-v0.0.14.tar.gz
+~~~
+
+2. Run Simulation
+After build your llama.cpp, you can run the executable file via QEMU for simulation, for example:
+~~~
+export QEMU_ROOT_PATH={your QEMU file path}
+export RISCV_ROOT_PATH_IME1={your RISC-V compiler path}
+
+${QEMU_ROOT_PATH}/bin/qemu-riscv64 -L ${RISCV_ROOT_PATH_IME1}/sysroot -cpu max,vlen=256,elen=64,vext_spec=v1.0 ${PWD}/build-riscv64-spacemit/bin/llama-cli -m ${PWD}/models/Qwen2.5-0.5B-Instruct-Q4_0.gguf -t 1
+~~~
+## Performance
+#### Quantization Support For Matrix
+~~~
+model name      : Spacemit(R) X60
+isa             : rv64imafdcv_zicbom_zicboz_zicntr_zicond_zicsr_zifencei_zihintpause_zihpm_zfh_zfhmin_zca_zcd_zba_zbb_zbc_zbs_zkt_zve32f_zve32x_zve64d_zve64f_zve64x_zvfh_zvfhmin_zvkt_sscofpmf_sstc_svinval_svnapot_svpbmt
+mmu             : sv39
+uarch           : spacemit,x60
+mvendorid       : 0x710
+marchid         : 0x8000000058000001
+~~~
+
+Q4_0
+|   Model    |   Size   | Params | backend | threads | test | t/s |
+| -----------| -------- | ------ | ------- | ------- | ---- |------|
+Qwen2.5 0.5B |403.20 MiB|630.17 M|   cpu   |    4    | pp512|64.12 ± 0.26|
+Qwen2.5 0.5B |403.20 MiB|630.17 M|   cpu   |    4    | tg128|10.03 ± 0.01|
+Qwen2.5 1.5B |1011.16 MiB| 1.78 B |   cpu   |    4    | pp512|24.16 ± 0.02|
+Qwen2.5 1.5B |1011.16 MiB| 1.78 B |   cpu   |    4    | tg128|3.83 ± 0.06|
+Qwen2.5 3B   | 1.86 GiB  | 3.40 B |   cpu   |    4    | pp512|12.08 ± 0.02|
+Qwen2.5 3B   | 1.86 GiB  | 3.40 B |   cpu   |    4    | tg128|2.23 ± 0.02|
+
+Q4_1
+|   Model    |   Size   | Params | backend | threads | test | t/s |
+| -----------| -------- | ------ | ------- | ------- | ---- |------|
+Qwen2.5 0.5B |351.50 MiB|494.03 M|   cpu   |    4    | pp512|62.07 ± 0.12|
+Qwen2.5 0.5B |351.50 MiB|494.03 M|   cpu   |    4    | tg128|9.91 ± 0.01|
+Qwen2.5 1.5B |964.06 MiB| 1.54 B |   cpu   |    4    | pp512|22.95 ± 0.25|
+Qwen2.5 1.5B |964.06 MiB| 1.54 B |   cpu   |    4    | tg128|4.01 ± 0.15|
+Qwen2.5 3B   | 1.85 GiB | 3.09 B |   cpu   |    4    | pp512|11.55 ± 0.16|
+Qwen2.5 3B   | 1.85 GiB | 3.09 B |   cpu   |    4    | tg128|2.25 ± 0.04|
+
+
+Q4_K
+|   Model    |   Size   | Params | backend | threads | test | t/s |
+| -----------| -------- | ------ | ------- | ------- | ---- |------|
+Qwen2.5 0.5B |462.96 MiB|630.17 M|   cpu   |    4    | pp512|9.29 ± 0.05|
+Qwen2.5 0.5B |462.96 MiB|630.17 M|   cpu   |    4    | tg128|5.67 ± 0.04|
+Qwen2.5 1.5B | 1.04 GiB | 1.78 B |   cpu   |    4    | pp512|10.38 ± 0.10|
+Qwen2.5 1.5B | 1.04 GiB | 1.78 B |   cpu   |    4    | tg128|3.17 ± 0.08|
+Qwen2.5 3B   | 1.95 GiB | 3.40 B |   cpu   |    4    | pp512|4.23 ± 0.04|
+Qwen2.5 3B   | 1.95 GiB | 3.40 B |   cpu   |    4    | tg128|1.73 ± 0.00|

ggml/src/ggml-cpu/CMakeLists.txt

@@ -433,7 +433,16 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
             ggml-cpu/arch/riscv/quants.c
             ggml-cpu/arch/riscv/repack.cpp
             )
-        if (GGML_RVV)
+        if (GGML_CPU_RISCV64_SPACEMIT)
+            list(APPEND ARCH_FLAGS -march=rv64gcv_zfh_zba_zicbop -mabi=lp64d -DGGML_RV_ZFH -D${RISCV64_SPACEMIT_IME_SPEC})
+            target_compile_definitions(${GGML_CPU_NAME} PRIVATE GGML_USE_CPU_RISCV64_SPACEMIT)
+            list(APPEND GGML_CPU_SOURCES
+                ggml-cpu/spacemit/ggml_spacemit_ime.cpp
+                ggml-cpu/spacemit/ggml_spacemit_ime.h
+                ggml-cpu/spacemit/ggml_spacemit_ime_kernels.cpp
+                ggml-cpu/spacemit/ggml_spacemit_ime_kernels.h
+            )
+        elseif (GGML_RVV)
             if (GGML_XTHEADVECTOR)
                 list(APPEND ARCH_FLAGS -march=rv64gc_xtheadvector -mabi=lp64d)
             elseif (GGML_RV_ZFH)

ggml/src/ggml-cpu/ggml-cpu.c

@@ -3209,6 +3209,26 @@ void ggml_cpu_fp32_to_fp16(const float * x, ggml_fp16_t * y, int64_t n) {
         uint16x8_t v_y = vec_convert_to_fp16(v_yd, 0);
         vec_xst(v_y, 0, (ggml_fp16_t *)(y + i));
     }
+#elif defined(__riscv) && defined(__riscv_v) && defined(__riscv_zfh)
+    int64_t n_loop = n;
+    __asm__ volatile(
+            "LOOP%=:                                    \n\t"
+            "vsetvli        t0, %[n], e32, m4,tu,mu     \n\t"
+            "slli           t1, t0, 1                   \n\t"
+            "slli           t2, t0, 2                   \n\t"
+            "vle32.v        v0, (%[IN])                 \n\t"
+            "add            %[IN], %[IN], t2            \n\t"
+            "vsetvli        t0, %[n], e16, m2,tu,mu     \n\t"
+            "vfncvt.f.f.w   v4, v0                      \n\t"
+            "vse16.v        v4, (%[DST])                \n\t"
+            "add            %[DST], %[DST], t1          \n\t"
+            "sub            %[n],  %[n], t0             \n\t"
+            "bnez           %[n], LOOP%=                \n\t"
+
+            : [ IN ] "+r"(x), [ DST ] "+r"(y), [ n ] "+r"(n_loop)
+            :
+            : "cc", "t0", "t1", "t2");
+    i += n;
 #endif
     for (; i < n; ++i) {
         y[i] = GGML_CPU_FP32_TO_FP16(x[i]);
@@ -3250,6 +3270,26 @@ void ggml_cpu_fp16_to_fp32(const ggml_fp16_t * x, float * y, int64_t n) {
         float32x4_t v_yh = vec_extend_to_fp32_hi(v_yd, 0);
         vec_xst(v_yh, 0, (float *)(y + i));
     }
+#elif defined(__riscv) && defined(__riscv_v) && defined(__riscv_zfh)
+    int64_t n_loop = n;
+    __asm__ volatile(
+        "LOOP%=:                                    \n\t"
+        "vsetvli        t0, %[n], e16, m2,tu,mu     \n\t"
+        "slli           t1, t0, 2                   \n\t"
+        "slli           t2, t0, 1                   \n\t"
+        "vle16.v        v0, (%[IN])                 \n\t"
+        "add            %[IN], %[IN], t2            \n\t"
+        "vfwcvt.f.f.v   v4, v0                      \n\t"
+        "vsetvli        t0, %[n], e32, m4,tu,mu     \n\t"
+        "vse32.v        v4, (%[DST])                \n\t"
+        "add            %[DST], %[DST], t1          \n\t"
+        "sub            %[n],  %[n], t0             \n\t"
+        "bnez           %[n], LOOP%=                \n\t"
+
+        : [ IN ] "+r"(x), [ DST ] "+r"(y), [ n ] "+r"(n_loop)
+        :
+        : "cc", "t0", "t1", "t2");
+    i += n;
 #endif
 
     for (; i < n; ++i) {

ggml/src/ggml-cpu/ggml-cpu.cpp

@@ -18,6 +18,10 @@
 #    include "kleidiai/kleidiai.h"
 #endif
 
+#ifdef GGML_USE_CPU_RISCV64_SPACEMIT
+#    include "spacemit/ggml_spacemit_ime.h"
+#endif
+
 #if defined(_WIN32)
 #    define WIN32_LEAN_AND_MEAN
 #    ifndef NOMINMAX
@@ -45,6 +49,12 @@ std::vector<ggml_backend_buffer_type_t> & ggml_backend_cpu_get_extra_buffer_type
         }
 #endif
 
+#ifdef GGML_USE_CPU_RISCV64_SPACEMIT
+        if (ggml_backend_cpu_riscv64_spacemit_buffer_type()) {
+            bufts.push_back(ggml_backend_cpu_riscv64_spacemit_buffer_type());
+        }
+#endif
+
 #ifdef GGML_USE_CPU_KLEIDIAI
         if (ggml_backend_cpu_kleidiai_buffer_type()) {
             bufts.push_back(ggml_backend_cpu_kleidiai_buffer_type());