aboutsummaryrefslogtreecommitdiffstats
path: root/src/svm
diff options
context:
space:
mode:
authorFlorin Coras <fcoras@cisco.com>2017-06-19 12:26:09 -0400
committerDave Barach <openvpp@barachs.net>2017-06-22 16:55:03 +0000
commitf6359c8cace5b73a813e5f4e3d1bc28f7752fcdf (patch)
tree38aa2a4dca15fb49c8c6ed967f1fd50d522926cc /src/svm
parent5be2d073ceb695104fd84bce592cb588f8aaaa66 (diff)
Improve svm fifo and tcp tx path performance (VPP-846)
- multiarch on svm fifo - avoid ip lookup on tx Change-Id: Iab0d85204a710979417bca1d692cc47877131203 Signed-off-by: Florin Coras <fcoras@cisco.com> Signed-off-by: Dave Barach <dbarach@cisco.com>
Diffstat (limited to 'src/svm')
-rw-r--r--src/svm/svm_fifo.c95
1 files changed, 90 insertions, 5 deletions
diff --git a/src/svm/svm_fifo.c b/src/svm/svm_fifo.c
index 6ca437cfd7b..aed5d6a7420 100644
--- a/src/svm/svm_fifo.c
+++ b/src/svm/svm_fifo.c
@@ -14,6 +14,7 @@
*/
#include <svm/svm_fifo.h>
+#include <vppinfra/cpu.h>
static inline u8
position_lt (svm_fifo_t * f, u32 a, u32 b)
@@ -417,10 +418,38 @@ svm_fifo_enqueue_internal (svm_fifo_t * f, u32 max_bytes, u8 * copy_from_here)
return (total_copy_bytes);
}
+#define SVM_ENQUEUE_CLONE_TEMPLATE(arch, fn, tgt) \
+ uword \
+ __attribute__ ((flatten)) \
+ __attribute__ ((target (tgt))) \
+ CLIB_CPU_OPTIMIZED \
+ fn ## _ ## arch ( svm_fifo_t * f, u32 max_bytes, u8 * copy_from_here) \
+ { return fn (f, max_bytes, copy_from_here);}
+
+static int
+svm_fifo_enqueue_nowait_ma (svm_fifo_t * f, u32 max_bytes,
+ u8 * copy_from_here)
+{
+ return svm_fifo_enqueue_internal (f, max_bytes, copy_from_here);
+}
+
+foreach_march_variant (SVM_ENQUEUE_CLONE_TEMPLATE,
+ svm_fifo_enqueue_nowait_ma);
+CLIB_MULTIARCH_SELECT_FN (svm_fifo_enqueue_nowait_ma);
+
int
svm_fifo_enqueue_nowait (svm_fifo_t * f, u32 max_bytes, u8 * copy_from_here)
{
- return svm_fifo_enqueue_internal (f, max_bytes, copy_from_here);
+#if CLIB_DEBUG > 0
+ return svm_fifo_enqueue_nowait_ma (f, max_bytes, copy_from_here);
+#else
+ static int (*fp) (svm_fifo_t *, u32, u8 *);
+
+ if (PREDICT_FALSE (fp == 0))
+ fp = (void *) svm_fifo_enqueue_nowait_ma_multiarch_select ();
+
+ return (*fp) (f, max_bytes, copy_from_here);
+#endif
}
/**
@@ -541,15 +570,43 @@ svm_fifo_dequeue_internal (svm_fifo_t * f, u32 max_bytes, u8 * copy_here)
return (total_copy_bytes);
}
-int
-svm_fifo_dequeue_nowait (svm_fifo_t * f, u32 max_bytes, u8 * copy_here)
+static int
+svm_fifo_dequeue_nowait_ma (svm_fifo_t * f, u32 max_bytes, u8 * copy_here)
{
return svm_fifo_dequeue_internal (f, max_bytes, copy_here);
}
+#define SVM_FIFO_DEQUEUE_CLONE_TEMPLATE(arch, fn, tgt) \
+ uword \
+ __attribute__ ((flatten)) \
+ __attribute__ ((target (tgt))) \
+ CLIB_CPU_OPTIMIZED \
+ fn ## _ ## arch ( svm_fifo_t * f, u32 max_bytes, \
+ u8 * copy_here) \
+ { return fn (f, max_bytes, copy_here);}
+
+foreach_march_variant (SVM_FIFO_DEQUEUE_CLONE_TEMPLATE,
+ svm_fifo_dequeue_nowait_ma);
+CLIB_MULTIARCH_SELECT_FN (svm_fifo_dequeue_nowait_ma);
+
int
-svm_fifo_peek (svm_fifo_t * f, u32 relative_offset, u32 max_bytes,
- u8 * copy_here)
+svm_fifo_dequeue_nowait (svm_fifo_t * f, u32 max_bytes, u8 * copy_here)
+{
+#if CLIB_DEBUG > 0
+ return svm_fifo_dequeue_nowait_ma (f, max_bytes, copy_here);
+#else
+ static int (*fp) (svm_fifo_t *, u32, u8 *);
+
+ if (PREDICT_FALSE (fp == 0))
+ fp = (void *) svm_fifo_dequeue_nowait_ma_multiarch_select ();
+
+ return (*fp) (f, max_bytes, copy_here);
+#endif
+}
+
+static int
+svm_fifo_peek_ma (svm_fifo_t * f, u32 relative_offset, u32 max_bytes,
+ u8 * copy_here)
{
u32 total_copy_bytes, first_copy_bytes, second_copy_bytes;
u32 cursize, nitems, real_head;
@@ -586,6 +643,34 @@ svm_fifo_peek (svm_fifo_t * f, u32 relative_offset, u32 max_bytes,
return total_copy_bytes;
}
+#define SVM_FIFO_PEEK_CLONE_TEMPLATE(arch, fn, tgt) \
+ uword \
+ __attribute__ ((flatten)) \
+ __attribute__ ((target (tgt))) \
+ CLIB_CPU_OPTIMIZED \
+ fn ## _ ## arch ( svm_fifo_t * f, u32 relative_offset, u32 max_bytes, \
+ u8 * copy_here) \
+ { return fn (f, relative_offset, max_bytes, copy_here);}
+
+foreach_march_variant (SVM_FIFO_PEEK_CLONE_TEMPLATE, svm_fifo_peek_ma);
+CLIB_MULTIARCH_SELECT_FN (svm_fifo_peek_ma);
+
+int
+svm_fifo_peek (svm_fifo_t * f, u32 relative_offset, u32 max_bytes,
+ u8 * copy_here)
+{
+#if CLIB_DEBUG > 0
+ return svm_fifo_peek_ma (f, relative_offset, max_bytes, copy_here);
+#else
+ static int (*fp) (svm_fifo_t *, u32, u32, u8 *);
+
+ if (PREDICT_FALSE (fp == 0))
+ fp = (void *) svm_fifo_peek_ma_multiarch_select ();
+
+ return (*fp) (f, relative_offset, max_bytes, copy_here);
+#endif
+}
+
int
svm_fifo_dequeue_drop (svm_fifo_t * f, u32 max_bytes)
{