conv_acc: Our code requires SSSE3, not just SSE3
The accelerated convolutional decoder uses SSSE3 instructions such
as PSIGNW (via _mm_sign_epi16) which go beyond what SSE3 offers. So
let's make sure we use the right compiler flag (-mssse3) and also the
right runtime check.
Without this patch, we would use illegal instructions e.g. on Opteron
Gen3 such as Opteron 2427, which are also used as build.opensuse.org
build hosts (build31 through build36) where we wouldn't pass "make
check" as a result.
Change-Id: I2754164384109f2821fd98ffb48f625893f2923d
Fixes: OS#2386
diff --git a/src/conv_acc.c b/src/conv_acc.c
index 33fe264..c16e436 100644
--- a/src/conv_acc.c
+++ b/src/conv_acc.c
@@ -48,7 +48,7 @@
static int init_complete = 0;
__attribute__ ((visibility("hidden"))) int avx2_supported = 0;
-__attribute__ ((visibility("hidden"))) int sse3_supported = 0;
+__attribute__ ((visibility("hidden"))) int ssse3_supported = 0;
__attribute__ ((visibility("hidden"))) int sse41_supported = 0;
/**
@@ -75,12 +75,12 @@
int16_t *osmo_conv_gen_vdec_malloc(size_t n);
void osmo_conv_gen_vdec_free(int16_t *ptr);
-#if defined(HAVE_SSE3)
+#if defined(HAVE_SSSE3)
int16_t *osmo_conv_sse_vdec_malloc(size_t n);
void osmo_conv_sse_vdec_free(int16_t *ptr);
#endif
-#if defined(HAVE_SSE3) && defined(HAVE_AVX2)
+#if defined(HAVE_SSSE3) && defined(HAVE_AVX2)
int16_t *osmo_conv_sse_avx_vdec_malloc(size_t n);
void osmo_conv_sse_avx_vdec_free(int16_t *ptr);
#endif
@@ -99,7 +99,7 @@
void osmo_conv_gen_metrics_k7_n4(const int8_t *seq, const int16_t *out,
int16_t *sums, int16_t *paths, int norm);
-#if defined(HAVE_SSE3)
+#if defined(HAVE_SSSE3)
void osmo_conv_sse_metrics_k5_n2(const int8_t *seq, const int16_t *out,
int16_t *sums, int16_t *paths, int norm);
void osmo_conv_sse_metrics_k5_n3(const int8_t *seq, const int16_t *out,
@@ -114,7 +114,7 @@
int16_t *sums, int16_t *paths, int norm);
#endif
-#if defined(HAVE_SSE3) && defined(HAVE_AVX2)
+#if defined(HAVE_SSSE3) && defined(HAVE_AVX2)
void osmo_conv_sse_avx_metrics_k5_n2(const int8_t *seq, const int16_t *out,
int16_t *sums, int16_t *paths, int norm);
void osmo_conv_sse_avx_metrics_k5_n3(const int8_t *seq, const int16_t *out,
@@ -654,8 +654,8 @@
avx2_supported = __builtin_cpu_supports("avx2");
#endif
- #ifdef HAVE_SSE3
- sse3_supported = __builtin_cpu_supports("sse3");
+ #ifdef HAVE_SSSE3
+ ssse3_supported = __builtin_cpu_supports("ssse3");
#endif
#ifdef HAVE_SSE4_1
@@ -667,16 +667,16 @@
* Usage of curly braces is mandatory,
* because we use multi-line define.
*/
-#if defined(HAVE_SSE3) && defined(HAVE_AVX2)
- if (sse3_supported && avx2_supported) {
+#if defined(HAVE_SSSE3) && defined(HAVE_AVX2)
+ if (ssse3_supported && avx2_supported) {
INIT_POINTERS(sse_avx);
- } else if (sse3_supported) {
+ } else if (ssse3_supported) {
INIT_POINTERS(sse);
} else {
INIT_POINTERS(gen);
}
-#elif defined(HAVE_SSE3)
- if (sse3_supported) {
+#elif defined(HAVE_SSSE3)
+ if (ssse3_supported) {
INIT_POINTERS(sse);
} else {
INIT_POINTERS(gen);