DTO API

Makefile

@@ -7,15 +7,16 @@ all: libdto dto-test-wodto
 DML_LIB_CXX=-D_GNU_SOURCE
 
 libdto: dto.c
-	gcc -shared -fPIC -Wl,-soname,libdto.so dto.c $(DML_LIB_CXX) -DDTO_STATS_SUPPORT -o libdto.so.1.0 -laccel-config -ldl -lnuma -mwaitpkg
+	gcc -shared -fPIC -Wl,-soname,libdto.so dto.c $(DML_LIB_CXX) -DDTO_STATS_SUPPORT -msse4.2 -o libdto.so.1.0 -laccel-config -ldl -lnuma -mwaitpkg
 
 libdto_nostats: dto.c
-	gcc -shared -fPIC -Wl,-soname,libdto.so dto.c $(DML_LIB_CXX) -o libdto.so.1.0 -laccel-config -ldl -lnuma -mwaitpkg
+	gcc -shared -fPIC -Wl,-soname,libdto.so dto.c $(DML_LIB_CXX) -msse4.2 -o libdto.so.1.0 -laccel-config -ldl -lnuma -mwaitpkg
 
 install:
 	cp libdto.so.1.0 /usr/lib64/
 	ln -sf /usr/lib64/libdto.so.1.0 /usr/lib64/libdto.so.1
 	ln -sf /usr/lib64/libdto.so.1.0 /usr/lib64/libdto.so
+	cp dto.h /usr/include/
 
 install-local:
 	ln -sf ./libdto.so.1.0 ./libdto.so.1

README.md

@@ -41,6 +41,65 @@ can be enabled or disabled using an environment variable DTO_AUTO_ADJUST_KNOBS.
 
 DTO can also be used to learn certain application characterics by building histogram of various API types and sizes. The histogram can be built using an environment variable DTO_COLLECT_STATS.
 
+## DTO API
+
+DTO provides drop-in replacements for standard memory operations. Each function
+is available in three forms:
+
+- `*_default` uses library defaults.
+- `*_cfg` accepts a `struct dto_call_cfg` to override per-call behavior.
+- The unsuffixed name derives configuration from a bitwise OR of `DTO_API_*`
+  flags.
+
+We also support CRC32C offload via `dto_memcpy_crc` and `dto_crc` functions. `dto_memcpy_crc`
+offloads both memcpy and CRC32C computation to DSA, while `dto_crc` offloads only CRC32C computation to DSA.
+
+Available entry points include:
+
+- `dto_memcpy_default`, `dto_memcpy_cfg`, `dto_memcpy`
+- `dto_memmove_default`, `dto_memmove_cfg`, `dto_memmove`
+- `dto_memset_default`, `dto_memset_cfg`, `dto_memset`
+- `dto_memcmp_default`, `dto_memcmp_cfg`, `dto_memcmp`
+- `dto_memcpy_crc_default`, `dto_memcpy_crc_cfg`, `dto_memcpy_crc`
+- `dto_crc_default`, `dto_crc_cfg`, `dto_crc`
+
+### Sample usage
+
+```c
+#include <stdint.h>
+#include "dto.h"
+
+int main(void)
+{
+    char src[64] = "example";
+    char dst[64];
+
+    /* Use defaults */
+    dto_memcpy_default(dst, src, sizeof(src));
+
+    /* Per-call configuration */
+    struct dto_call_cfg cfg = {
+        .auto_adjust = 0,
+        .cache_control = 1,
+        .wait_method = WAIT_BUSYPOLL,
+        .numa_mode = NA_BUFFER_CENTRIC,
+        .overlapping_action = OVERLAPPING_CPU,
+    };
+    int diff = dto_memcmp_cfg(dst, src, sizeof(src), &cfg);
+
+    /* Flags-based configuration */
+    dto_memset(dst, 0, sizeof(dst), DTO_API_WAIT_YIELD);
+
+    /* Copy while computing CRC */
+    uint32_t c1 = dto_memcpy_crc_default(dst, src, sizeof(src));
+
+    /* Compute CRC with flags-based configuration */
+    uint32_t c2 = dto_crc(src, sizeof(src), DTO_API_WAIT_YIELD, NULL, NULL);
+
+    return diff;
+}
+```
+
 ```bash
 dto.c: DSA Transparent Offload shared library
 dto-test.c: Sample multi-threaded test application

accelConfig.sh

@@ -4,6 +4,7 @@ echo "OPTIONAL Arg-1: DSA device id. Default: 0"
 echo "OPTIONAL Arg-2: Enable/Disable DSA device. Default: yes"
 echo "OPTIONAL Arg-3: SHARED WQ id. Default: 1"
 echo "OPTIONAL Arg-4: ENGINE count. Default: 4"
+echo "OPTIONAL Arg-5: Block on fault. Default: no, but this is needed for CRC32!"
 
 if [ "$#" -ge 5 ]; then
     echo "ERROR: Incorrect argument count. Expected arg count <= 4"
@@ -14,6 +15,7 @@ DEVID=${1:-0}
 ENABLE=${2:-yes}
 SWQID=${3:-1}
 NENGS=${4:-4}
+BLOCKONFAULT=${5:-no}
 
 DEV=dsa${DEVID}
 SWQ=${DEV}/wq${DEVID}.${SWQID}
@@ -41,7 +43,11 @@ accel-config config-wq ${SWQ} --priority=1
 accel-config config-wq ${SWQ} --wq-size=128
 accel-config config-wq ${SWQ} --max-batch-size=1024
 accel-config config-wq ${SWQ} --max-transfer-size=2147483648
-accel-config config-wq ${SWQ} --block-on-fault=0
+if [ "${BLOCKONFAULT}" == "yes" ]; then
+    accel-config config-wq ${SWQ} --block-on-fault=1
+else
+    accel-config config-wq ${SWQ} --block-on-fault=0
+
 accel-config config-wq ${SWQ} --type=user
 accel-config config-wq ${SWQ} --name="dsa-test"
 accel-config config-wq ${SWQ} --mode=shared