diff --git a/jconfig.h b/jconfig.h
index aa1fb18..b391dce 100644
--- a/jconfig.h
+++ b/jconfig.h
@@ -7,6 +7,9 @@
 /* libjpeg-turbo version */
 #define LIBJPEG_TURBO_VERSION 1.3.0
 
+/* RWK Nintendo THP fork */
+#define NINTENDO_THP_STREAM 1
+
 /* Support arithmetic encoding */
 #define C_ARITH_CODING_SUPPORTED 1
 
diff --git a/jdarith.c b/jdarith.c
index 6126eea..d556733 100644
--- a/jdarith.c
+++ b/jdarith.c
@@ -102,8 +102,6 @@ get_byte (j_decompress_ptr cinfo)
  * derived from Markus Kuhn's JBIG implementation.
  */
 
-#define NINTENDO_THP_STREAM 1
-
 LOCAL(int)
 arith_decode (j_decompress_ptr cinfo, unsigned char *st)
 {
@@ -120,9 +118,6 @@ arith_decode (j_decompress_ptr cinfo, unsigned char *st)
 	data = 0;		/* stuff zero data */
       else {
 	data = get_byte(cinfo);	/* read next input byte */
-#if NINTENDO_THP_STREAM
-        if (!cinfo->NINTENDO_THP_FLAG) {
-#endif
 	if (data == 0xFF) {	/* zero stuff or marker code */
 	  do data = get_byte(cinfo);
 	  while (data == 0xFF);	/* swallow extra 0xFF bytes */
@@ -139,9 +134,6 @@ arith_decode (j_decompress_ptr cinfo, unsigned char *st)
 	    data = 0;
 	  }
 	}
-#if NINTENDO_THP_STREAM
-        }
-#endif
       }
       e->c = (e->c << 8) | data; /* insert data into C register */
       if ((e->ct += 8) < 0)	 /* update bit shift counter */
diff --git a/jdhuff.c b/jdhuff.c
index 744e964..ef8ff2d 100644
--- a/jdhuff.c
+++ b/jdhuff.c
@@ -291,8 +291,6 @@ jpeg_make_d_derived_tbl (j_decompress_ptr cinfo, boolean isDC, int tblno,
 #define MIN_GET_BITS  (BIT_BUF_SIZE-7)
 #endif
 
-#define NINTENDO_THP_STREAM 1
-
 GLOBAL(boolean)
 jpeg_fill_bit_buffer (bitread_working_state * state,
 		      register bit_buf_type get_buffer, register int bits_left,
@@ -323,9 +321,7 @@ jpeg_fill_bit_buffer (bitread_working_state * state,
       c = GETJOCTET(*next_input_byte++);
 
       /* If it's 0xFF, check and discard stuffed zero byte */
-#if NINTENDO_THP_STREAM
-      if (!cinfo->NINTENDO_THP_FLAG) {
-#endif
+#if !NINTENDO_THP_STREAM
       if (c == 0xFF) {
 	/* Loop here to discard any padding FF's on terminating marker,
 	 * so that we can save a valid unread_marker value.  NOTE: we will
@@ -360,8 +356,6 @@ jpeg_fill_bit_buffer (bitread_working_state * state,
 	  goto no_more_bytes;
 	}
       }
-#if NINTENDO_THP_STREAM
-      }
 #endif
 
       /* OK, load c into get_buffer */
@@ -404,6 +398,17 @@ jpeg_fill_bit_buffer (bitread_working_state * state,
    handle markers.  We have to hand off any blocks with markers to the
    slower routines. */
 
+#if NINTENDO_THP_STREAM
+#define GET_BYTE \
+{ \
+  register int c0, c1; \
+  c0 = GETJOCTET(*buffer++); \
+  c1 = GETJOCTET(*buffer); \
+  /* Pre-execute most common case */ \
+  get_buffer = (get_buffer << 8) | c0; \
+  bits_left += 8; \
+}
+#else
 #define GET_BYTE \
 { \
   register int c0, c1; \
@@ -424,6 +429,7 @@ jpeg_fill_bit_buffer (bitread_working_state * state,
     } \
   } \
 }
+#endif
 
 #if __WORDSIZE == 64 || defined(_WIN64)
 
diff --git a/jdinput.c b/jdinput.c
index 2ec9dd3..0d3ae10 100644
--- a/jdinput.c
+++ b/jdinput.c
@@ -314,9 +314,6 @@ consume_markers (j_decompress_ptr cinfo)
 
   switch (val) {
   case JPEG_REACHED_SOS:	/* Found SOS */
-#if NINTENDO_THP_STREAM
-    cinfo->NINTENDO_THP_FLAG = 1;
-#endif
     if (inputctl->inheaders) {	/* 1st SOS */
       initial_setup(cinfo);
       inputctl->inheaders = FALSE;
@@ -331,9 +328,6 @@ consume_markers (j_decompress_ptr cinfo)
     }
     break;
   case JPEG_REACHED_EOI:	/* Found EOI */
-#if NINTENDO_THP_STREAM
-    cinfo->NINTENDO_THP_FLAG = 0;
-#endif
     inputctl->pub.eoi_reached = TRUE;
     if (inputctl->inheaders) {	/* Tables-only datastream, apparently */
       if (cinfo->marker->saw_SOF)
diff --git a/jpeglib.h b/jpeglib.h
index e039f70..5aa1010 100644
--- a/jpeglib.h
+++ b/jpeglib.h
@@ -705,11 +705,6 @@ struct jpeg_decompress_struct {
   struct jpeg_color_deconverter * cconvert;
   struct jpeg_color_quantizer * cquantize;
     
-#define NINTENDO_THP_STREAM 1
-#if NINTENDO_THP_STREAM
-  int NINTENDO_THP_FLAG;
-#endif
-    
 };