diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index 20e4d71..50ada52 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -563,7 +563,10 @@
   MACROBLOCK *const x = &cpi->mb;
   MACROBLOCKD *const xd = &x->e_mbd;
   int orig_rdmult = x->rdmult;
-  double rdmult_ratio = 1.0;
+  double rdmult_ratio;
+
+  vp9_clear_system_state();  // __asm emms;
+  rdmult_ratio = 1.0;  // avoid uninitialized warnings
 
   // Use the lower precision, but faster, 32x32 fdct for mode selection.
   x->use_lp32x32fdct = 1;
@@ -602,7 +605,10 @@
   if (cpi->oxcf.tuning == VP8_TUNE_SSIM)
     vp9_activity_masking(cpi, x);
 
-  x->rdmult = round(x->rdmult * rdmult_ratio);
+  if (cpi->sf.variance_adaptive_quantization) {
+    vp9_clear_system_state();  // __asm emms;
+    x->rdmult = round(x->rdmult * rdmult_ratio);
+  }
 
   // Find best coding mode & reconstruct the MB so it is available
   // as a predictor for MBs that follow in the SB
@@ -618,9 +624,13 @@
                                     totaldist, bsize, ctx, best_rd);
   }
 
-  x->rdmult = orig_rdmult;
-  if (*totalrate != INT_MAX)
-    *totalrate = round(*totalrate * rdmult_ratio);
+  if (cpi->sf.variance_adaptive_quantization) {
+    x->rdmult = orig_rdmult;
+    if (*totalrate != INT_MAX) {
+      vp9_clear_system_state();  // __asm emms;
+      *totalrate = round(*totalrate * rdmult_ratio);
+    }
+  }
 }
 
 static void update_stats(VP9_COMP *cpi) {
diff --git a/vp9/encoder/vp9_firstpass.c b/vp9/encoder/vp9_firstpass.c
index 7157451..caf4162 100644
--- a/vp9/encoder/vp9_firstpass.c
+++ b/vp9/encoder/vp9_firstpass.c
@@ -554,7 +554,10 @@
       int this_error;
       int gf_motion_error = INT_MAX;
       int use_dc_pred = (mb_col || mb_row) && (!mb_col || !mb_row);
-      double error_weight = 1.0;
+      double error_weight;
+
+      vp9_clear_system_state();  // __asm emms;
+      error_weight = 1.0;  // avoid uninitialized warnings
 
       xd->plane[0].dst.buf = new_yv12->y_buffer + recon_yoffset;
       xd->plane[1].dst.buf = new_yv12->u_buffer + recon_uvoffset;
@@ -587,7 +590,11 @@
       }
 
       // do intra 16x16 prediction
-      this_error = error_weight * vp9_encode_intra(x, use_dc_pred);
+      this_error = vp9_encode_intra(x, use_dc_pred);
+      if (cpi->sf.variance_adaptive_quantization) {
+        vp9_clear_system_state();  // __asm emms;
+        this_error *= error_weight;
+      }
 
       // intrapenalty below deals with situations where the intra and inter
       // error scores are very low (eg a plain black frame).
@@ -622,7 +629,10 @@
         first_pass_motion_search(cpi, x, &best_ref_mv,
                                  &mv.as_mv, lst_yv12,
                                  &motion_error, recon_yoffset);
-        motion_error *= error_weight;
+        if (cpi->sf.variance_adaptive_quantization) {
+          vp9_clear_system_state();  // __asm emms;
+          motion_error *= error_weight;
+        }
 
         // If the current best reference mv is not centered on 0,0 then do a 0,0
         // based search as well.
@@ -630,7 +640,10 @@
           tmp_err = INT_MAX;
           first_pass_motion_search(cpi, x, &zero_ref_mv, &tmp_mv.as_mv,
                                    lst_yv12, &tmp_err, recon_yoffset);
-          tmp_err *= error_weight;
+          if (cpi->sf.variance_adaptive_quantization) {
+            vp9_clear_system_state();  // __asm emms;
+            tmp_err *= error_weight;
+          }
 
           if (tmp_err < motion_error) {
             motion_error = tmp_err;
@@ -647,7 +660,10 @@
           first_pass_motion_search(cpi, x, &zero_ref_mv,
                                    &tmp_mv.as_mv, gld_yv12,
                                    &gf_motion_error, recon_yoffset);
-          gf_motion_error *= error_weight;
+          if (cpi->sf.variance_adaptive_quantization) {
+            vp9_clear_system_state();  // __asm emms;
+            gf_motion_error *= error_weight;
+          }
 
           if ((gf_motion_error < motion_error) &&
               (gf_motion_error < this_error)) {
diff --git a/vp9/encoder/vp9_vaq.c b/vp9/encoder/vp9_vaq.c
index 3d3b4b0..3179ae3 100644
--- a/vp9/encoder/vp9_vaq.c
+++ b/vp9/encoder/vp9_vaq.c
@@ -37,25 +37,36 @@
 
 unsigned int vp9_vaq_segment_id(int energy) {
   ENERGY_IN_BOUNDS(energy);
+
   return SEGMENT_ID(energy);
 }
 
 double vp9_vaq_rdmult_ratio(int energy) {
   ENERGY_IN_BOUNDS(energy);
+
+  vp9_clear_system_state();  // __asm emms;
+
   return RDMULT_RATIO(energy);
 }
 
 double vp9_vaq_inv_q_ratio(int energy) {
   ENERGY_IN_BOUNDS(energy);
+
+  vp9_clear_system_state();  // __asm emms;
+
   return Q_RATIO(-energy);
 }
 
 void vp9_vaq_init() {
   int i;
-  double base_ratio = 1.8;
+  double base_ratio;
 
   assert(ENERGY_SPAN <= MAX_SEGMENTS);
 
+  vp9_clear_system_state();  // __asm emms;
+
+  base_ratio = 1.8;
+
   for (i = ENERGY_MIN; i <= ENERGY_MAX; i++) {
     Q_RATIO(i) = pow(base_ratio, i/3.0);
   }
@@ -74,6 +85,8 @@
 
   seg->abs_delta = SEGMENT_DELTADATA;
 
+  vp9_clear_system_state();  // __asm emms;
+
   for (i = ENERGY_MIN; i <= ENERGY_MAX; i++) {
     int qindex_delta, segment_rdmult;
 
@@ -89,6 +102,7 @@
 
     segment_rdmult = vp9_compute_rd_mult(cpi, cm->base_qindex + qindex_delta +
                                          cm->y_dc_delta_q);
+
     RDMULT_RATIO(i) = (double) segment_rdmult / base_rdmult;
   }
 }
@@ -120,9 +134,14 @@
 }
 
 int vp9_block_energy(VP9_COMP *cpi, MACROBLOCK *x, BLOCK_SIZE bs) {
+  double energy;
+  unsigned int var = block_variance(cpi, x, bs);
+
+  vp9_clear_system_state();  // __asm emms;
+
   // if (var <= 1000)
   //   return 0;
-  unsigned int var = block_variance(cpi, x, bs);
-  double energy = 0.9*(logf(var + 1) - 10.0);
+
+  energy = 0.9*(logf(var + 1) - 10.0);
   return clamp(round(energy), ENERGY_MIN, ENERGY_MAX);
 }