X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?a=blobdiff_plain;ds=sidebyside;f=frontend%2Fcspace.c;h=33a981df27eb9cde28463d40d6124e8f83786e35;hb=HEAD;hp=f0c491266ffcef63efc9d5a6b878d326b7ae8c9e;hpb=c82f907a99f7d0d3071471489d247f4b4394a8b7;p=pcsx_rearmed.git

diff --git a/frontend/cspace.c b/frontend/cspace.c
index f0c49126..a3e3301f 100644
--- a/frontend/cspace.c
+++ b/frontend/cspace.c
@@ -1,5 +1,5 @@
 /*
- * (C) Gražvydas "notaz" Ignotas, 2011,2012
+ * (C) Gražvydas "notaz" Ignotas, 2011,2012,2022
  *
  * This work is licensed under the terms of any of these licenses
  * (at your option):
@@ -8,6 +8,7 @@
  * See the COPYING file in the top-level directory.
  */
 
+#include <stdint.h>
 #include "cspace.h"
 
 /*
@@ -15,23 +16,103 @@
  * in favor of NEON version or platform-specific conversion
  */
 
-#ifndef __ARM_NEON__
+#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+#define SWAP16(x) __builtin_bswap16(x)
+#define LE16TOHx2(x) ((SWAP16((x) >> 16) << 16) | SWAP16(x))
+#else
+#define LE16TOHx2(x) (x)
+#endif
+
+#if defined(HAVE_bgr555_to_rgb565)
+
+/* have bgr555_to_rgb565 somewhere else */
+
+#elif ((defined(__clang_major__) && __clang_major__ >= 4) \
+        || (defined(__GNUC__) && __GNUC__ >= 5)) \
+       && __BYTE_ORDER__ != __ORDER_BIG_ENDIAN__
+
+#include <assert.h>
+
+#if defined(__ARM_NEON) || defined(__ARM_NEON__)
+#include <arm_neon.h>
+#define gsli(d_, s_, n_) d_ = vsliq_n_u16(d_, s_, n_)
+#define gsri(d_, s_, n_) d_ = vsriq_n_u16(d_, s_, n_)
+#else
+#define gsli(d_, s_, n_) d_ |= s_ << n_
+#define gsri(d_, s_, n_) d_ |= s_ >> n_
+#endif
+
+typedef uint16_t gvu16  __attribute__((vector_size(16),aligned(16)));
+typedef uint16_t gvu16u __attribute__((vector_size(16),aligned(2)));
+#define gdup(v_) {v_, v_, v_, v_, v_, v_, v_, v_}
+#define do_one(s) ({ \
+  uint16_t d_ = (s) << 1; d_ = (d_ & 0x07c0) | (d_ << 10) | (d_ >> 11); d_; \
+})
+#define do_one_simd(d_, s_, c0x07c0_) { \
+  gvu16 s1 = s_ << 1; \
+  d_ = s1 & c0x07c0_; \
+  gsli(d_, s_, 11); \
+  gsri(d_, s1, 11); \
+}
+
+void bgr555_to_rgb565(void * __restrict__ dst_, const void *  __restrict__ src_, int bytes)
+{
+	const uint16_t * __restrict__ src = src_;
+	uint16_t * __restrict__ dst = dst_;
+	gvu16 c0x07c0 = gdup(0x07c0);
+
+	assert(!(((uintptr_t)dst | (uintptr_t)src | bytes) & 1));
+
+	// align the destination
+	if ((uintptr_t)dst & 0x0e)
+	{
+		uintptr_t left = 0x10 - ((uintptr_t)dst & 0x0e);
+		gvu16 d, s = *(const gvu16u *)src;
+		do_one_simd(d, s, c0x07c0);
+		*(gvu16u *)dst = d;
+		dst += left / 2;
+		src += left / 2;
+		bytes -= left;
+	}
+	// go
+	for (; bytes >= 16; dst += 8, src += 8, bytes -= 16)
+	{
+		gvu16 d, s = *(const gvu16u *)src;
+		do_one_simd(d, s, c0x07c0);
+		*(gvu16 *)dst = d;
+		__builtin_prefetch(src + 128/2);
+	}
+	// finish it
+	for (; bytes > 0; dst++, src++, bytes -= 2)
+		*dst = do_one(*src);
+}
+#undef do_one
+#undef do_one_simd
+
+#else
 
 void bgr555_to_rgb565(void *dst_, const void *src_, int bytes)
 {
-	const unsigned int *src = src_;
+	// source can be misaligned, but it's very rare, so just force
+	const unsigned int *src = (const void *)((intptr_t)src_ & ~3);
 	unsigned int *dst = dst_;
-	unsigned int p;
-	int x;
+	unsigned int x, p, r, g, b;
 
 	for (x = 0; x < bytes / 4; x++) {
-		p = src[x];
-		p = ((p & 0x7c007c00) >> 10) | ((p & 0x03e003e0) << 1)
-			| ((p & 0x001f001f) << 11);
-		dst[x] = p;
+		p = LE16TOHx2(src[x]);
+
+		r = (p & 0x001f001f) << 11;
+		g = (p & 0x03e003e0) << 1;
+		b = (p & 0x7c007c00) >> 10;
+
+		dst[x] = r | g | b;
 	}
 }
 
+#endif
+
+#ifndef HAVE_bgr888_to_x
+
 void bgr888_to_rgb565(void *dst_, const void *src_, int bytes)
 {
 	const unsigned char *src = src_;
@@ -45,8 +126,13 @@ void bgr888_to_rgb565(void *dst_, const void *src_, int bytes)
 		r2 = src[3] & 0xf8;
 		g2 = src[4] & 0xfc;
 		b2 = src[5] & 0xf8;
+#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+		*dst = (r1 << 24) | (g1 << 19) | (b1 << 13) |
+			(r2 << 8) | (g2 << 3) | (b2 >> 3);
+#else
 		*dst = (r2 << 24) | (g2 << 19) | (b2 << 13) |
 			(r1 << 8) | (g1 << 3) | (b1 >> 3);
+#endif
 	}
 }
 
@@ -129,7 +215,7 @@ void bgr555_to_uyvy(void *d, const void *s, int pixels)
   int r0, g0, b0, r1, g1, b1;
   int y0, y1, u, v;
 
-  for (; pixels > 0; src += 2, dst++, pixels -= 2)
+  for (; pixels > 1; src += 2, dst++, pixels -= 2)
   {
     b0 = (src[0] >> 10) & 0x1f;
     g0 = (src[0] >> 5) & 0x1f;