From fec006a4f9f6b0ce43272d72c017ef9f29c8a296 Mon Sep 17 00:00:00 2001
From: Cameron Cawley <[EMAIL REDACTED]>
Date: Sat, 12 Oct 2024 17:20:32 +0100
Subject: [PATCH] Allow for more fine tuning of Duff's device routines
---
src/video/SDL_blit.h | 46 +++++++++++++++++++++++++++++++++---------
src/video/SDL_blit_1.c | 16 +++++++--------
src/video/SDL_blit_A.c | 22 ++++++++++----------
src/video/SDL_blit_N.c | 14 ++++++-------
4 files changed, 62 insertions(+), 36 deletions(-)
diff --git a/src/video/SDL_blit.h b/src/video/SDL_blit.h
index 816194d07841e..375446f366817 100644
--- a/src/video/SDL_blit.h
+++ b/src/video/SDL_blit.h
@@ -610,6 +610,15 @@ extern SDL_BlitFunc SDL_CalculateBlitA(SDL_Surface *surface);
#else
#define USE_DUFFS_LOOP
#endif
+
+#define DUFFS_LOOP1(pixel_copy_increment, width) \
+ { \
+ int n; \
+ for (n = width; n > 0; --n) { \
+ pixel_copy_increment; \
+ } \
+ }
+
#ifdef USE_DUFFS_LOOP
// 8-times unrolled loop
@@ -666,8 +675,26 @@ extern SDL_BlitFunc SDL_CalculateBlitA(SDL_Surface *surface);
} \
}
-// Use the 8-times version of the loop by default
+// 2-times unrolled loop
+#define DUFFS_LOOP2(pixel_copy_increment, width) \
+ { \
+ int n = (width + 1) / 2; \
+ switch (width & 1) { \
+ case 0: \
+ do { \
+ pixel_copy_increment; \
+ SDL_FALLTHROUGH; \
+ case 1: \
+ pixel_copy_increment; \
+ } while (--n > 0); \
+ } \
+ }
+
+// Use the 4-times version of the loop by default
#define DUFFS_LOOP(pixel_copy_increment, width) \
+ DUFFS_LOOP4(pixel_copy_increment, width)
+// Use the 8-times version of the loop for simple routines
+#define DUFFS_LOOP_TRIVIAL(pixel_copy_increment, width) \
DUFFS_LOOP8(pixel_copy_increment, width)
// Special version of Duff's device for even more optimization
@@ -701,20 +728,19 @@ extern SDL_BlitFunc SDL_CalculateBlitA(SDL_Surface *surface);
// Don't use Duff's device to unroll loops
#define DUFFS_LOOP(pixel_copy_increment, width) \
- { \
- int n; \
- for (n = width; n > 0; --n) { \
- pixel_copy_increment; \
- } \
- }
+ DUFFS_LOOP1(pixel_copy_increment, width)
+#define DUFFS_LOOP_TRIVIAL(pixel_copy_increment, width) \
+ DUFFS_LOOP1(pixel_copy_increment, width)
#define DUFFS_LOOP8(pixel_copy_increment, width) \
- DUFFS_LOOP(pixel_copy_increment, width)
+ DUFFS_LOOP1(pixel_copy_increment, width)
#define DUFFS_LOOP4(pixel_copy_increment, width) \
- DUFFS_LOOP(pixel_copy_increment, width)
+ DUFFS_LOOP1(pixel_copy_increment, width)
+#define DUFFS_LOOP2(pixel_copy_increment, width) \
+ DUFFS_LOOP1(pixel_copy_increment, width)
#define DUFFS_LOOP_124(pixel_copy_increment1, \
pixel_copy_increment2, \
pixel_copy_increment4, width) \
- DUFFS_LOOP(pixel_copy_increment1, width)
+ DUFFS_LOOP1(pixel_copy_increment1, width)
#endif // USE_DUFFS_LOOP
diff --git a/src/video/SDL_blit_1.c b/src/video/SDL_blit_1.c
index 63e7c873b5670..1cc0a8c028316 100644
--- a/src/video/SDL_blit_1.c
+++ b/src/video/SDL_blit_1.c
@@ -48,7 +48,7 @@ static void Blit1to1(SDL_BlitInfo *info)
while (height--) {
#ifdef USE_DUFFS_LOOP
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP(
+ DUFFS_LOOP_TRIVIAL(
{
*dst = map[*src];
}
@@ -100,7 +100,7 @@ static void Blit1to2(SDL_BlitInfo *info)
#ifdef USE_DUFFS_LOOP
while (height--) {
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP(
+ DUFFS_LOOP_TRIVIAL(
{
*(Uint16 *)dst = map[*src++];
dst += 2;
@@ -256,7 +256,7 @@ static void Blit1to4(SDL_BlitInfo *info)
while (height--) {
#ifdef USE_DUFFS_LOOP
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP(
+ DUFFS_LOOP_TRIVIAL(
*dst++ = map[*src++];
, width);
/* *INDENT-ON* */ // clang-format on
@@ -297,7 +297,7 @@ static void Blit1to1Key(SDL_BlitInfo *info)
if (palmap) {
while (height--) {
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP(
+ DUFFS_LOOP_TRIVIAL(
{
if ( *src != ckey ) {
*dst = palmap[*src];
@@ -313,7 +313,7 @@ static void Blit1to1Key(SDL_BlitInfo *info)
} else {
while (height--) {
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP(
+ DUFFS_LOOP_TRIVIAL(
{
if ( *src != ckey ) {
*dst = *src;
@@ -345,7 +345,7 @@ static void Blit1to2Key(SDL_BlitInfo *info)
while (height--) {
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP(
+ DUFFS_LOOP_TRIVIAL(
{
if ( *src != ckey ) {
*dstp=palmap[*src];
@@ -408,7 +408,7 @@ static void Blit1to4Key(SDL_BlitInfo *info)
while (height--) {
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP(
+ DUFFS_LOOP_TRIVIAL(
{
if ( *src != ckey ) {
*dstp = palmap[*src];
@@ -444,7 +444,7 @@ static void Blit1toNAlpha(SDL_BlitInfo *info)
while (height--) {
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP4(
+ DUFFS_LOOP(
{
sR = srcpal[*src].r;
sG = srcpal[*src].g;
diff --git a/src/video/SDL_blit_A.c b/src/video/SDL_blit_A.c
index f39ff80388bd5..403c482e77065 100644
--- a/src/video/SDL_blit_A.c
+++ b/src/video/SDL_blit_A.c
@@ -46,7 +46,7 @@ static void BlitNto1SurfaceAlpha(SDL_BlitInfo *info)
while (height--) {
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP4(
+ DUFFS_LOOP(
{
DISEMBLE_RGB(src, srcbpp, srcfmt, Pixel, sR, sG, sB);
dR = dstpal[*dst].r;
@@ -91,7 +91,7 @@ static void BlitNto1PixelAlpha(SDL_BlitInfo *info)
while (height--) {
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP4(
+ DUFFS_LOOP(
{
DISEMBLE_RGBA(src,srcbpp,srcfmt,Pixel,sR,sG,sB,sA);
dR = dstpal[*dst].r;
@@ -253,7 +253,7 @@ static void BlitRGBtoRGBSurfaceAlpha128(SDL_BlitInfo *info)
while (height--) {
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP4({
+ DUFFS_LOOP({
Uint32 s = *srcp++;
Uint32 d = *dstp;
*dstp++ = ((((s & 0x00fefefe) + (d & 0x00fefefe)) >> 1)
@@ -283,7 +283,7 @@ static void BlitRGBtoRGBSurfaceAlpha(SDL_BlitInfo *info)
while (height--) {
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP4({
+ DUFFS_LOOP({
s = *srcp;
d = *dstp;
@@ -705,7 +705,7 @@ static void Blit565to565SurfaceAlpha(SDL_BlitInfo *info)
while (height--) {
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP4({
+ DUFFS_LOOP({
Uint32 s = *srcp++;
Uint32 d = *dstp;
/*
@@ -743,7 +743,7 @@ static void Blit555to555SurfaceAlpha(SDL_BlitInfo *info)
while (height--) {
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP4({
+ DUFFS_LOOP({
Uint32 s = *srcp++;
Uint32 d = *dstp;
/*
@@ -776,7 +776,7 @@ static void BlitARGBto565PixelAlpha(SDL_BlitInfo *info)
while (height--) {
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP4({
+ DUFFS_LOOP({
Uint32 s = *srcp;
unsigned alpha = s >> 27; // downscale alpha to 5 bits
/* Here we special-case opaque alpha since the
@@ -819,7 +819,7 @@ static void BlitARGBto555PixelAlpha(SDL_BlitInfo *info)
while (height--) {
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP4({
+ DUFFS_LOOP({
unsigned alpha;
Uint32 s = *srcp;
alpha = s >> 27; // downscale alpha to 5 bits
@@ -872,7 +872,7 @@ static void BlitNtoNSurfaceAlpha(SDL_BlitInfo *info)
if (sA) {
while (height--) {
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP4(
+ DUFFS_LOOP(
{
DISEMBLE_RGB(src, srcbpp, srcfmt, Pixel, sR, sG, sB);
DISEMBLE_RGBA(dst, dstbpp, dstfmt, Pixel, dR, dG, dB, dA);
@@ -910,7 +910,7 @@ static void BlitNtoNSurfaceAlphaKey(SDL_BlitInfo *info)
while (height--) {
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP4(
+ DUFFS_LOOP(
{
RETRIEVE_RGB_PIXEL(src, srcbpp, Pixel);
if (sA && Pixel != ckey) {
@@ -1302,7 +1302,7 @@ static void BlitNtoNPixelAlpha(SDL_BlitInfo *info)
dstbpp = dstfmt->bytes_per_pixel;
while (height--) {
- DUFFS_LOOP4(
+ DUFFS_LOOP(
{
DISEMBLE_RGBA(src, srcbpp, srcfmt, Pixel, sR, sG, sB, sA);
if (sA) {
diff --git a/src/video/SDL_blit_N.c b/src/video/SDL_blit_N.c
index 9a120a1142a6c..652a906623330 100644
--- a/src/video/SDL_blit_N.c
+++ b/src/video/SDL_blit_N.c
@@ -1800,7 +1800,7 @@ static void Blit_RGB555_ARGB1555(SDL_BlitInfo *info)
while (height--) {
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP(
+ DUFFS_LOOP_TRIVIAL(
{
*dst = *src | mask;
++dst;
@@ -1831,7 +1831,7 @@ static void Blit4to4MaskAlpha(SDL_BlitInfo *info)
while (height--) {
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP(
+ DUFFS_LOOP_TRIVIAL(
{
*dst = *src | mask;
++dst;
@@ -1848,7 +1848,7 @@ static void Blit4to4MaskAlpha(SDL_BlitInfo *info)
while (height--) {
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP(
+ DUFFS_LOOP_TRIVIAL(
{
*dst = *src & mask;
++dst;
@@ -2142,7 +2142,7 @@ static void Blit2to2Key(SDL_BlitInfo *info)
while (height--) {
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP(
+ DUFFS_LOOP_TRIVIAL(
{
if ( (*srcp & rgbmask) != ckey ) {
*dstp = *srcp;
@@ -2188,7 +2188,7 @@ static void BlitNtoNKey(SDL_BlitInfo *info)
Uint32 mask = ((Uint32)info->a) << dstfmt->Ashift;
while (height--) {
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP(
+ DUFFS_LOOP_TRIVIAL(
{
if ((*src32 & rgbmask) != ckey) {
*dst32 = *src32 | mask;
@@ -2206,7 +2206,7 @@ static void BlitNtoNKey(SDL_BlitInfo *info)
Uint32 mask = srcfmt->Rmask | srcfmt->Gmask | srcfmt->Bmask;
while (height--) {
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP(
+ DUFFS_LOOP_TRIVIAL(
{
if ((*src32 & rgbmask) != ckey) {
*dst32 = *src32 & mask;
@@ -2463,7 +2463,7 @@ static void BlitNtoNKeyCopyAlpha(SDL_BlitInfo *info)
Uint32 *dst32 = (Uint32 *)dst;
while (height--) {
/* *INDENT-OFF* */ // clang-format off
- DUFFS_LOOP(
+ DUFFS_LOOP_TRIVIAL(
{
if ((*src32 & rgbmask) != ckey) {
*dst32 = *src32;