ref: 7f3dd7ed3c97b1f0030ac24753ab248a8d8374b4
parent: 189c55a1eb5902b09579efb79ed9a5b0f960b6c0
author: Chris Moeller <kode54@gmail.com>
date: Sun Jan 11 20:25:45 EST 2015
Sanity fix for ARM NEON code: vmlaq always adds onto the first input parameter
--- a/dumb/src/helpers/resampler.c
+++ b/dumb/src/helpers/resampler.c
@@ -626,8 +626,8 @@
{
temp1 = vld1q_f32( (const float32_t *)( kernel + i ) );
temp2 = vld1q_f32( (const float32_t *) out + i * 4 );
- temp1 = vmlaq_f32( temp2, temp1, samplex );
- vst1q_f32( (float32_t *) out + i * 4, temp1 );
+ temp2 = vmlaq_f32( temp2, temp1, samplex );
+ vst1q_f32( (float32_t *) out + i * 4, temp2 );
}
}
@@ -935,8 +935,8 @@
{
temp1 = vld1q_f32( (const float32_t *)( kernel + i ) );
temp2 = vld1q_f32( (const float32_t *) out + i * 4 );
- temp1 = vmlaq_f32( temp2, temp1, samplex );
- vst1q_f32( (float32_t *) out + i * 4, temp1 );
+ temp2 = vmlaq_f32( temp2, temp1, samplex );
+ vst1q_f32( (float32_t *) out + i * 4, temp2 );
}
}