Hi,
On 2019/08/04 1:00, Michael wrote:
On Sat, 3 Aug 2019 14:46:32 +0900
Rin Okuyama <[email protected]> wrote:
Maybe it's time to remove all non-32bit access to fb.
I expect it is not a very hard work for now ;-).
I seriously doubt that's the problem, because:
- 32bit powerpc doesn't really do 64bit accesses ( unlike sparc for
example ) and altivec is disabled for kernel code ( since gcc started
using altivec for optimized, inlined memcpy )
- at least one of the putchar_aa() methods used memcpy() in order to
speed things up by rendering scanlines into cached memory and then
quickly copying them into slow & uncached video memory, which worked
just fine everywhere I tried ( that is, mips, powerpc, sparc, sparc64
and arm )
Thank you for your suggestive comments!
I probably found the cause of failure; new rasops allocates buffer and
stamp dynamically via kmem_alloc. This may not work in early stages
during boot.
I removed dynamical allocations. Could you please test the attached patch?
PS
I ordered Mac Mini G4, although serial console is hopeless...
They're nice little machines which usually don't cause much trouble.
Opening them is quite painful though.
Yeah, I look to forward to playing with it :-).
Thanks,
rin
Index: sys/dev/rasops/rasops.c
===================================================================
RCS file: /cvsroot/src/sys/dev/rasops/rasops.c,v
retrieving revision 1.108
diff -p -u -r1.108 rasops.c
--- sys/dev/rasops/rasops.c 2 Aug 2019 23:24:37 -0000 1.108
+++ sys/dev/rasops/rasops.c 4 Aug 2019 02:19:31 -0000
@@ -50,6 +50,7 @@ __KERNEL_RCSID(0, "$NetBSD: rasops.c,v 1
#define _RASOPS_PRIVATE
#include <dev/rasops/rasops.h>
+#include <dev/rasops/rasops_masks.h> /* XXX for MBE */
#ifndef _KERNEL
#include <errno.h>
@@ -69,6 +70,16 @@ struct rasops_matchdata {
int ident;
};
+static const uint32_t rasops_lmask32[4 + 1] = {
+ MBE(0x00000000), MBE(0x00ffffff), MBE(0x0000ffff), MBE(0x000000ff),
+ MBE(0x00000000),
+};
+
+static const uint32_t rasops_rmask32[4 + 1] = {
+ MBE(0x00000000), MBE(0xff000000), MBE(0xffff0000), MBE(0xffffff00),
+ MBE(0xffffffff),
+};
+
/* ANSI colormap (R,G,B). Upper 8 are high-intensity */
const uint8_t rasops_cmap[256 * 3] = {
0x00, 0x00, 0x00, /* black */
@@ -429,7 +440,7 @@ rasops_reconfig(struct rasops_info *ri,
/* Clear the entire display */
if ((ri->ri_flg & RI_CLEAR) != 0)
- memset(ri->ri_bits, 0, ri->ri_stride * ri->ri_height);
+ rasops_memset32(ri->ri_bits, 0, ri->ri_stride * ri->ri_height);
/* Now centre our window if needs be */
if ((ri->ri_flg & RI_CENTER) != 0) {
@@ -495,21 +506,6 @@ rasops_reconfig(struct rasops_info *ri,
WSSCREEN_WSCOLORS | WSSCREEN_REVERSE;
}
- if (ri->ri_buf != NULL) {
- kmem_free(ri->ri_buf, ri->ri_buflen);
- ri->ri_buf = NULL;
- }
- len = (ri->ri_flg & RI_FULLCLEAR) ? ri->ri_stride : ri->ri_emustride;
- ri->ri_buflen = len;
- ri->ri_buf = kmem_alloc(len, KM_SLEEP);
-
-#ifndef RASOPS_SMALL
- if (ri->ri_stamp != NULL) {
- kmem_free(ri->ri_stamp, ri->ri_stamp_len);
- ri->ri_stamp = NULL;
- }
-#endif
-
switch (ri->ri_depth) {
#if NRASOPS1 > 0
case 1:
@@ -980,9 +976,8 @@ void
rasops_eraserows(void *cookie, int row, int num, long attr)
{
struct rasops_info *ri = (struct rasops_info *)cookie;
- uint32_t *buf = (uint32_t *)ri->ri_buf;
uint32_t *rp, *hp, clr;
- int stride, cnt;
+ int stride;
hp = NULL; /* XXX GCC */
@@ -1021,13 +1016,10 @@ rasops_eraserows(void *cookie, int row,
hp = (uint32_t *)(ri->ri_hwbits + row * ri->ri_yscale);
}
- for (cnt = 0; cnt < stride >> 2; cnt++)
- buf[cnt] = clr;
-
while (num--) {
- memcpy(rp, buf, stride);
+ rasops_memset32(rp, clr, stride);
if (ri->ri_hwbits) {
- memcpy(hp, buf, stride);
+ memcpy(hp, rp, stride);
DELTA(hp, ri->ri_stride, uint32_t *);
}
DELTA(rp, ri->ri_stride, uint32_t *);
@@ -1042,9 +1034,8 @@ static void
rasops_do_cursor(struct rasops_info *ri)
{
int full, height, cnt, slop1, slop2, row, col;
- uint32_t tmp32, msk1, msk2;
- uint8_t tmp8;
- uint8_t *dp, *rp, *hp;
+ uint32_t mask1, mask2, *dp;
+ uint8_t tmp8, *rp, *hp;
hp = NULL; /* XXX GCC */
@@ -1108,28 +1099,24 @@ rasops_do_cursor(struct rasops_info *ri)
rp = (uint8_t *)((uintptr_t)rp & ~3);
hp = (uint8_t *)((uintptr_t)hp & ~3);
- msk1 = !slop1 ? 0 : be32toh(0xffffffffU >> (32 - (8 * slop1)));
- msk2 = !slop2 ? 0 : be32toh(0xffffffffU << (32 - (8 * slop2)));
+ mask1 = rasops_lmask32[4 - slop1];
+ mask2 = rasops_rmask32[slop2];
while (height--) {
- dp = rp;
+ dp = (uint32_t *)rp;
if (slop1) {
- tmp32 = *(uint32_t *)dp ^ msk1;
- *(uint32_t *)dp = tmp32;
- dp += 4;
+ *dp = *dp ^ mask1;
+ dp++;
}
for (cnt = full; cnt; cnt--) {
- tmp32 = ~*(uint32_t *)dp;
- *(uint32_t *)dp = tmp32;
- dp += 4;
+ *dp = ~*(uint32_t *)dp;
+ dp++;
}
- if (slop2) {
- tmp32 = *(uint32_t *)dp ^ msk2;
- *(uint32_t *)dp = tmp32;
- }
+ if (slop2)
+ *dp = *dp ^ mask2;
if (ri->ri_hwbits) {
memcpy(hp, rp, ((slop1 != 0) + full +
@@ -1147,9 +1134,8 @@ void
rasops_erasecols(void *cookie, int row, int col, int num, long attr)
{
struct rasops_info *ri = (struct rasops_info *)cookie;
- uint32_t *buf = ri->ri_buf;
- int height, cnt, clr;
- uint32_t *dp, *rp, *hp;
+ int height, clr;
+ uint32_t *rp, *hp;
hp = NULL; /* XXX GCC */
@@ -1177,25 +1163,13 @@ rasops_erasecols(void *cookie, int row,
height = ri->ri_font->fontheight;
clr = ri->ri_devcmap[((uint32_t)attr >> 16) & 0xf];
- dp = buf;
-
- /* Write 4 bytes per loop */
- for (cnt = num >> 2; cnt; cnt--)
- *dp++ = clr;
-
- /* Write unaligned trailing slop */
- for (cnt = num & 3; cnt; cnt--) {
- *(uint8_t *)dp = clr;
- DELTA(dp, 1, uint32_t *);
- }
-
while (height--) {
- memcpy(rp, buf, num);
- DELTA(rp, ri->ri_stride, uint32_t *);
+ rasops_memset32(rp, clr, num);
if (ri->ri_hwbits) {
- memcpy(hp, buf, num);
+ memcpy(hp, rp, num);
DELTA(hp, ri->ri_stride, uint32_t *);
}
+ DELTA(rp, ri->ri_stride, uint32_t *);
}
}
@@ -1670,15 +1644,3 @@ rasops_get_cmap(struct rasops_info *ri,
memcpy(palette, rasops_cmap, uimin(bytes, sizeof(rasops_cmap)));
return 0;
}
-
-#ifndef RASOPS_SMALL
-void
-rasops_allocstamp(struct rasops_info *ri, size_t len)
-{
-
- KASSERT(ri->ri_stamp == NULL);
- ri->ri_stamp_len = len;
- ri->ri_stamp = kmem_zalloc(len, KM_SLEEP);
- ri->ri_stamp_attr = 0;
-}
-#endif
Index: sys/dev/rasops/rasops.h
===================================================================
RCS file: /cvsroot/src/sys/dev/rasops/rasops.h,v
retrieving revision 1.43
diff -p -u -r1.43 rasops.h
--- sys/dev/rasops/rasops.h 3 Aug 2019 06:29:52 -0000 1.43
+++ sys/dev/rasops/rasops.h 4 Aug 2019 02:19:31 -0000
@@ -131,15 +131,6 @@ struct rasops_info {
/* Callbacks so we can share some code */
void (*ri_do_cursor)(struct rasops_info *);
- /* buffer capable of single-row pixels */
- void *ri_buf;
- size_t ri_buflen;
-
- /* 4x1 stamp for optimized character blitting */
- void *ri_stamp;
- long ri_stamp_attr;
- size_t ri_stamp_len;
-
#if NRASOPS_ROTATION > 0
/* Used to intercept putchar to permit display rotation */
struct wsdisplay_emulops ri_real_ops;
@@ -199,6 +190,45 @@ void rasops_allocstamp(struct rasops_inf
((uint8_t *)(font)->data + ((uc) - ((font)->firstchar)) * \
(ri)->ri_fontscale)
+static __inline void
+rasops_memset32(void *p, uint32_t val, size_t bytes)
+{
+ int slop1, slop2, full;
+ uint8_t *dp = (uint8_t *)p;
+
+ if (bytes == 1) {
+ *dp = val;
+ return;
+ }
+
+ slop1 = (4 - ((uintptr_t)dp & 3)) & 3;
+ slop2 = (bytes - slop1) & 3;
+ full = (bytes - slop1 /* - slop2 */) >> 2;
+
+ if (slop1 & 1)
+ *dp++ = val;
+
+ if (slop1 & 2) {
+ *(uint16_t *)dp = val;
+ dp += 2;
+ }
+
+ for (; full; full--) {
+ *(uint32_t *)dp = val;
+ dp += 4;
+ }
+
+ if (slop2 & 2) {
+ *(uint16_t *)dp = val;
+ dp += 2;
+ }
+
+ if (slop1 & 1)
+ *dp = val;
+
+ return;
+}
+
static __inline uint32_t
be32uatoh(uint8_t *p)
{
Index: sys/dev/rasops/rasops15.c
===================================================================
RCS file: /cvsroot/src/sys/dev/rasops/rasops15.c,v
retrieving revision 1.34
diff -p -u -r1.34 rasops15.c
--- sys/dev/rasops/rasops15.c 2 Aug 2019 04:40:53 -0000 1.34
+++ sys/dev/rasops/rasops15.c 4 Aug 2019 02:19:31 -0000
@@ -55,6 +55,11 @@ static void rasops15_makestamp(struct ra
#endif
#ifndef RASOPS_SMALL
+/* 4x1 stamp for optimized character blitting */
+static uint32_t stamp[32];
+static long stamp_attr;
+static struct rasops_info *stamp_ri;
+
/*
* offset = STAMP_SHIFT(fontbits, nibble #) & STAMP_MASK
* destination uint32_t[0] = STAMP_READ(offset)
@@ -104,7 +109,8 @@ rasops15_init(struct rasops_info *ri)
}
#ifndef RASOPS_SMALL
- rasops_allocstamp(ri, sizeof(uint32_t) * 32);
+ stamp_attr = 0;
+ stamp_ri = NULL;
#endif
}
@@ -118,13 +124,14 @@ rasops15_init(struct rasops_info *ri)
static void
rasops15_makestamp(struct rasops_info *ri, long attr)
{
- uint32_t *stamp = (uint32_t *)ri->ri_stamp;
uint32_t fg, bg;
int i;
+ stamp_attr = attr;
+ stamp_ri = ri;
+
fg = ri->ri_devcmap[((uint32_t)attr >> 24) & 0xf] & 0xffff;
bg = ri->ri_devcmap[((uint32_t)attr >> 16) & 0xf] & 0xffff;
- ri->ri_stamp_attr = attr;
for (i = 0; i < 32; i += 2) {
#if BYTE_ORDER == LITTLE_ENDIAN
Index: sys/dev/rasops/rasops2.c
===================================================================
RCS file: /cvsroot/src/sys/dev/rasops/rasops2.c,v
retrieving revision 1.29
diff -p -u -r1.29 rasops2.c
--- sys/dev/rasops/rasops2.c 2 Aug 2019 04:39:09 -0000 1.29
+++ sys/dev/rasops/rasops2.c 4 Aug 2019 02:19:31 -0000
@@ -58,6 +58,12 @@ static void rasops2_putchar16(void *, in
static void rasops2_makestamp(struct rasops_info *, long);
#endif
+#ifndef RASOPS_SMALL
+/* 4x1 stamp for optimized character blitting */
+static uint8_t stamp[16];
+static long stamp_attr;
+static struct rasops_info *stamp_ri;
+
/*
* offset = STAMP_SHIFT(fontbits, nibble #) & STAMP_MASK
* destination = STAMP_READ(offset)
@@ -65,6 +71,7 @@ static void rasops2_makestamp(struct ras
#define STAMP_SHIFT(fb, n) ((n) ? (fb) >> 4 : (fb))
#define STAMP_MASK 0xf
#define STAMP_READ(o) stamp[o]
+#endif
/*
* Initialize rasops_info struct for this colordepth.
@@ -90,14 +97,15 @@ rasops2_init(struct rasops_info *ri)
case 16:
ri->ri_ops.putchar = rasops2_putchar16;
break;
-#endif /* !RASOPS_SMALL */
+#endif
default:
ri->ri_ops.putchar = rasops2_putchar;
return;
}
#ifndef RASOPS_SMALL
- rasops_allocstamp(ri, sizeof(uint8_t) * 16);
+ stamp_attr = 0;
+ stamp_ri = NULL;
#endif
}
@@ -108,12 +116,13 @@ rasops2_init(struct rasops_info *ri)
static void
rasops2_makestamp(struct rasops_info *ri, long attr)
{
- uint8_t *stamp = (uint8_t *)ri->ri_stamp;
int i, fg, bg;
+ stamp_attr = attr;
+ stamp_ri = ri;
+
fg = ri->ri_devcmap[((uint32_t)attr >> 24) & 0xf] & 3;
bg = ri->ri_devcmap[((uint32_t)attr >> 16) & 0xf] & 3;
- ri->ri_stamp_attr = attr;
for (i = 0; i < 16; i++) {
#if BYTE_ORDER == BIG_ENDIAN
Index: sys/dev/rasops/rasops24.c
===================================================================
RCS file: /cvsroot/src/sys/dev/rasops/rasops24.c,v
retrieving revision 1.46
diff -p -u -r1.46 rasops24.c
--- sys/dev/rasops/rasops24.c 2 Aug 2019 23:24:37 -0000 1.46
+++ sys/dev/rasops/rasops24.c 4 Aug 2019 02:19:32 -0000
@@ -62,6 +62,12 @@ static void rasops24_putchar16(void *,
static void rasops24_makestamp(struct rasops_info *, long);
#endif
+#ifndef RASOPS_SMALL
+/* 4x1 stamp for optimized character blitting */
+static uint32_t stamp[64];
+static long stamp_attr;
+static struct rasops_info *stamp_ri;
+
/*
* offset = STAMP_SHIFT(fontbits, nibble #) & STAMP_MASK
* destination uint32_t[0] = STAMP_READ(offset)
@@ -71,6 +77,7 @@ static void rasops24_makestamp(struct ra
#define STAMP_SHIFT(fb, n) ((n) ? (fb) : (fb) << 4)
#define STAMP_MASK (0xf << 4)
#define STAMP_READ(o) (*(uint32_t *)((uint8_t *)stamp + (o)))
+#endif
/*
* Initialize rasops_info struct for this colordepth.
@@ -113,7 +120,8 @@ rasops24_init(struct rasops_info *ri)
}
#ifndef RASOPS_SMALL
- rasops_allocstamp(ri, sizeof(uint32_t) * 64);
+ stamp_attr = 0;
+ stamp_ri = NULL;
#endif
}
@@ -121,13 +129,13 @@ rasops24_init(struct rasops_info *ri)
#include "rasops_putchar_aa.h"
static __inline void
-rasops24_makestamp1(struct rasops_info *ri, uint32_t *stamp,
+rasops24_makestamp1(struct rasops_info *ri, uint32_t *xstamp,
uint32_t c1, uint32_t c2, uint32_t c3, uint32_t c4)
{
- stamp[0] = (c1 << 8) | (c2 >> 16);
- stamp[1] = (c2 << 16) | (c3 >> 8);
- stamp[2] = (c3 << 24) | c4;
+ xstamp[0] = (c1 << 8) | (c2 >> 16);
+ xstamp[1] = (c2 << 16) | (c3 >> 8);
+ xstamp[2] = (c3 << 24) | c4;
#if BYTE_ORDER == LITTLE_ENDIAN
if ((ri->ri_flg & RI_BSWAP) == 0)
@@ -135,9 +143,9 @@ rasops24_makestamp1(struct rasops_info *
if ((ri->ri_flg & RI_BSWAP) != 0)
#endif
{
- stamp[0] = bswap32(stamp[0]);
- stamp[1] = bswap32(stamp[1]);
- stamp[2] = bswap32(stamp[2]);
+ xstamp[0] = bswap32(xstamp[0]);
+ xstamp[1] = bswap32(xstamp[1]);
+ xstamp[2] = bswap32(xstamp[2]);
}
}
@@ -148,13 +156,14 @@ rasops24_makestamp1(struct rasops_info *
static void
rasops24_makestamp(struct rasops_info *ri, long attr)
{
- uint32_t *stamp = (uint32_t *)ri->ri_stamp;
uint32_t fg, bg, c1, c2, c3, c4;
int i;
+ stamp_attr = attr;
+ stamp_ri = ri;
+
fg = ri->ri_devcmap[((uint32_t)attr >> 24) & 0xf] & 0xffffff;
bg = ri->ri_devcmap[((uint32_t)attr >> 16) & 0xf] & 0xffffff;
- ri->ri_stamp_attr = attr;
for (i = 0; i < 64; i += 4) {
#if BYTE_ORDER == LITTLE_ENDIAN
@@ -193,9 +202,8 @@ static void
rasops24_eraserows(void *cookie, int row, int num, long attr)
{
struct rasops_info *ri = (struct rasops_info *)cookie;
- uint32_t *buf = (uint32_t *)ri->ri_buf;
int full, slop, cnt, stride;
- uint32_t *rp, *dp, *hp, clr, stamp[3];
+ uint32_t *rp, *dp, *hp, clr, xstamp[3];
hp = NULL; /* XXX GCC */
@@ -222,7 +230,7 @@ rasops24_eraserows(void *cookie, int row
#endif
clr = ri->ri_devcmap[((uint32_t)attr >> 16) & 0xf] & 0xffffff;
- rasops24_makestamp1(ri, stamp, clr, clr, clr, clr);
+ rasops24_makestamp1(ri, xstamp, clr, clr, clr, clr);
/*
* XXX the wsdisplay_emulops interface seems a little deficient in
@@ -247,25 +255,23 @@ rasops24_eraserows(void *cookie, int row
full = stride / (4 * 3);
slop = (stride - full * (4 * 3)) / 4;
- dp = buf;
-
- for (cnt = full; cnt; cnt--) {
- dp[0] = stamp[0];
- dp[1] = stamp[1];
- dp[2] = stamp[2];
- dp += 3;
- }
-
- for (cnt = 0; cnt < slop; cnt++)
- *dp++ = stamp[cnt];
-
while (num--) {
- memcpy(rp, buf, stride);
- DELTA(rp, ri->ri_stride, uint32_t *);
+ dp = rp;
+ for (cnt = full; cnt; cnt--) {
+ dp[0] = xstamp[0];
+ dp[1] = xstamp[1];
+ dp[2] = xstamp[2];
+ dp += 3;
+ }
+ for (cnt = 0; cnt < slop; cnt++)
+ *dp++ = xstamp[cnt];
+
if (ri->ri_hwbits) {
- memcpy(hp, buf, stride);
+ memcpy(hp, rp, stride);
DELTA(hp, ri->ri_stride, uint32_t *);
}
+
+ DELTA(rp, ri->ri_stride, uint32_t *);
}
}
@@ -276,9 +282,8 @@ static void
rasops24_erasecols(void *cookie, int row, int col, int num, long attr)
{
struct rasops_info *ri = (struct rasops_info *)cookie;
- void *buf = ri->ri_buf;
- int height, cnt, clr, stamp[3];
- uint32_t *dp;
+ int height, cnt, slop1, slop2, full;
+ uint32_t clr, xstamp[3], *dp;
uint8_t *rp, *hp, *dbp;
hp = NULL; /* XXX GCC */
@@ -317,33 +322,46 @@ rasops24_erasecols(void *cookie, int row
height = ri->ri_font->fontheight;
clr = ri->ri_devcmap[((uint32_t)attr >> 16) & 0xf] & 0xffffff;
- rasops24_makestamp1(ri, stamp, clr, clr, clr, clr);
-
- /* 4 pels per loop */
- dp = (uint32_t *)buf;
- for (cnt = num >> 2; cnt; cnt--) {
- dp[0] = stamp[0];
- dp[1] = stamp[1];
- dp[2] = stamp[2];
- dp += 3;
- }
-
- /* Trailing slop */
- dbp = (uint8_t *)dp;
- for (cnt = num & 3; cnt; cnt--) {
- *dbp++ = (clr >> 16);
- *dbp++ = (clr >> 8);
- *dbp++ = clr;
- }
+ rasops24_makestamp1(ri, xstamp, clr, clr, clr, clr);
+ slop1 = (uintptr_t)rp & 3;
+ cnt = slop1;
+ full = (num /* - cnt */) >> 2;
+ cnt += full << 2;
+ slop2 = num - cnt;
num *= 3;
while (height--) {
- memcpy(rp, buf, num);
- rp += ri->ri_stride;
+ /* Align to word boundary */
+ dbp = rp;
+ for (cnt = slop1; cnt; cnt--) {
+ *dbp++ = (clr >> 16);
+ *dbp++ = (clr >> 8);
+ *dbp++ = clr;
+ }
+
+ /* 4 pels per loop */
+ dp = (uint32_t *)dbp;
+ for (cnt = full; cnt; cnt--) {
+ dp[0] = xstamp[0];
+ dp[1] = xstamp[1];
+ dp[2] = xstamp[2];
+ dp += 3;
+ }
+
+ /* Trailing slop */
+ dbp = (uint8_t *)dp;
+ for (cnt = slop2; cnt; cnt--) {
+ *dbp++ = (clr >> 16);
+ *dbp++ = (clr >> 8);
+ *dbp++ = clr;
+ }
+
if (ri->ri_hwbits) {
- memcpy(hp, buf, num);
+ memcpy(hp, rp, num);
hp += ri->ri_stride;
}
+
+ rp += ri->ri_stride;
}
}
Index: sys/dev/rasops/rasops32.c
===================================================================
RCS file: /cvsroot/src/sys/dev/rasops/rasops32.c,v
retrieving revision 1.42
diff -p -u -r1.42 rasops32.c
--- sys/dev/rasops/rasops32.c 2 Aug 2019 04:40:53 -0000 1.42
+++ sys/dev/rasops/rasops32.c 4 Aug 2019 02:19:32 -0000
@@ -54,6 +54,12 @@ static void rasops32_putchar16(void *, i
static void rasops32_makestamp(struct rasops_info *, long);
#endif
+#ifndef RASOPS_SMALL
+/* 4x1 stamp for optimized character blitting */
+static uint32_t stamp[64];
+static long stamp_attr;
+static struct rasops_info *stamp_ri;
+
/*
* offset = STAMP_SHIFT(fontbits, nibble #) & STAMP_MASK
* destination uint32_t[0] = STAMP_READ(offset)
@@ -64,6 +70,7 @@ static void rasops32_makestamp(struct ra
#define STAMP_SHIFT(fb, n) ((n) ? (fb) : (fb) << 4)
#define STAMP_MASK (0xf << 4)
#define STAMP_READ(o) (*(uint32_t *)((uint8_t *)stamp + (o)))
+#endif
/*
* Initialize a 'rasops_info' descriptor for this depth.
@@ -103,7 +110,8 @@ rasops32_init(struct rasops_info *ri)
}
#ifndef RASOPS_SMALL
- rasops_allocstamp(ri, sizeof(uint32_t) * 64);
+ stamp_attr = 0;
+ stamp_ri = NULL;
#endif
}
@@ -117,13 +125,14 @@ rasops32_init(struct rasops_info *ri)
static void
rasops32_makestamp(struct rasops_info *ri, long attr)
{
- uint32_t *stamp = (uint32_t *)ri->ri_stamp;
uint32_t fg, bg;
int i;
+ stamp_attr = attr;
+ stamp_ri = ri;
+
fg = ri->ri_devcmap[((uint32_t)attr >> 24) & 0xf];
bg = ri->ri_devcmap[((uint32_t)attr >> 16) & 0xf];
- ri->ri_stamp_attr = attr;
for (i = 0; i < 64; i += 4) {
stamp[i + 0] = i & 32 ? fg : bg;
Index: sys/dev/rasops/rasops4.c
===================================================================
RCS file: /cvsroot/src/sys/dev/rasops/rasops4.c,v
retrieving revision 1.24
diff -p -u -r1.24 rasops4.c
--- sys/dev/rasops/rasops4.c 2 Aug 2019 04:39:09 -0000 1.24
+++ sys/dev/rasops/rasops4.c 4 Aug 2019 02:19:32 -0000
@@ -58,6 +58,12 @@ static void rasops4_putchar16(void *, in
static void rasops4_makestamp(struct rasops_info *, long);
#endif
+#ifndef RASOPS_SMALL
+/* 4x1 stamp for optimized character blitting */
+static uint16_t stamp[16];
+static long stamp_attr;
+static struct rasops_info *stamp_ri;
+
/*
* offset = STAMP_SHIFT(fontbits, nibble #) & STAMP_MASK
* destination = STAMP_READ(offset)
@@ -65,6 +71,7 @@ static void rasops4_makestamp(struct ras
#define STAMP_SHIFT(fb, n) ((n) ? (fb) >> 4 : (fb))
#define STAMP_MASK 0xf
#define STAMP_READ(o) stamp[o]
+#endif
/*
* Initialize rasops_info struct for this colordepth.
@@ -97,7 +104,8 @@ rasops4_init(struct rasops_info *ri)
}
#ifndef RASOPS_SMALL
- rasops_allocstamp(ri, sizeof(uint16_t) * 16);
+ stamp_attr = 0;
+ stamp_ri = NULL;
#endif
}
@@ -108,12 +116,13 @@ rasops4_init(struct rasops_info *ri)
static void
rasops4_makestamp(struct rasops_info *ri, long attr)
{
- uint16_t *stamp = (uint16_t *)ri->ri_stamp;
int i, fg, bg;
+ stamp_attr = attr;
+ stamp_ri = ri;
+
fg = ri->ri_devcmap[((uint32_t)attr >> 24) & 0xf] & 0xf;
bg = ri->ri_devcmap[((uint32_t)attr >> 16) & 0xf] & 0xf;
- ri->ri_stamp_attr = attr;
for (i = 0; i < 16; i++) {
#if BYTE_ORDER == BIG_ENDIAN
Index: sys/dev/rasops/rasops8.c
===================================================================
RCS file: /cvsroot/src/sys/dev/rasops/rasops8.c,v
retrieving revision 1.47
diff -p -u -r1.47 rasops8.c
--- sys/dev/rasops/rasops8.c 2 Aug 2019 04:40:53 -0000 1.47
+++ sys/dev/rasops/rasops8.c 4 Aug 2019 02:19:32 -0000
@@ -54,6 +54,12 @@ static void rasops8_putchar16(void *, i
static void rasops8_makestamp(struct rasops_info *ri, long);
#endif
+#ifndef RASOPS_SMALL
+/* 4x1 stamp for optimized character blitting */
+static uint32_t stamp[16];
+static long stamp_attr;
+static struct rasops_info *stamp_ri;
+
/*
* offset = STAMP_SHIFT(fontbits, nibble #) & STAMP_MASK
* destination = STAMP_READ(offset)
@@ -61,6 +67,7 @@ static void rasops8_makestamp(struct ras
#define STAMP_SHIFT(fb, n) ((n) ? (fb) >> 2 : (fb) << 2)
#define STAMP_MASK (0xf << 2)
#define STAMP_READ(o) (*(uint32_t *)((uint8_t *)stamp + (o)))
+#endif
/*
* Initialize a 'rasops_info' descriptor for this depth.
@@ -101,7 +108,8 @@ rasops8_init(struct rasops_info *ri)
}
#ifndef RASOPS_SMALL
- rasops_allocstamp(ri, sizeof(uint32_t) * 16);
+ stamp_attr = 0;
+ stamp_ri = NULL;
#endif
}
@@ -115,13 +123,14 @@ rasops8_init(struct rasops_info *ri)
static void
rasops8_makestamp(struct rasops_info *ri, long attr)
{
- uint32_t *stamp = (uint32_t *)ri->ri_stamp;
uint32_t fg, bg;
int i;
+ stamp_attr = attr;
+ stamp_ri = ri;
+
fg = ri->ri_devcmap[((uint32_t)attr >> 24) & 0xf] & 0xff;
bg = ri->ri_devcmap[((uint32_t)attr >> 16) & 0xf] & 0xff;
- ri->ri_stamp_attr = attr;
for (i = 0; i < 16; i++) {
#if BYTE_ORDER == BIG_ENDIAN
Index: sys/dev/rasops/rasops_putchar_aa.h
===================================================================
RCS file: /cvsroot/src/sys/dev/rasops/rasops_putchar_aa.h,v
retrieving revision 1.6
diff -p -u -r1.6 rasops_putchar_aa.h
--- sys/dev/rasops/rasops_putchar_aa.h 31 Jul 2019 04:45:44 -0000 1.6
+++ sys/dev/rasops/rasops_putchar_aa.h 4 Aug 2019 02:19:32 -0000
@@ -71,16 +71,18 @@
#define SET_WIDTH(p, c) memset(p, clr[c], width)
#endif
+#define MAX_WIDTH 32
+
static void
PUTCHAR_AA(RASOPS_DEPTH)(void *cookie, int row, int col, u_int uc, long attr)
{
struct rasops_info *ri = (struct rasops_info *)cookie;
struct wsdisplay_font *font = PICK_FONT(ri, uc);
- PIXEL_TYPE *buf = (PIXEL_TYPE *)ri->ri_buf;
int height, width, x, y, off[2];
uint16_t r[2], g[2], b[2];
uint8_t *fr, aval;
PIXEL_TYPE *rp, *hp, R, G, B;
+ PIXEL_TYPE buf[MAX_WIDTH * PIXEL_BYTES / sizeof(PIXEL_TYPE)];
COLOR_TYPE clr[2];
hp = NULL; /* XXX GCC */
@@ -205,3 +207,5 @@ PUTCHAR_AA(RASOPS_DEPTH)(void *cookie, i
#undef PIXEL_BYTES
#undef SET_PIXEL
#undef SET_WIDTH
+
+#undef MAX_WIDTH
Index: sys/dev/rasops/rasops_putchar_width.h
===================================================================
RCS file: /cvsroot/src/sys/dev/rasops/rasops_putchar_width.h,v
retrieving revision 1.10
diff -p -u -r1.10 rasops_putchar_width.h
--- sys/dev/rasops/rasops_putchar_width.h 31 Jul 2019 02:04:14 -0000
1.10
+++ sys/dev/rasops/rasops_putchar_width.h 4 Aug 2019 02:19:33 -0000
@@ -204,7 +204,6 @@ PUTCHAR_WIDTH(RASOPS_DEPTH, RASOPS_WIDTH
{
struct rasops_info *ri = (struct rasops_info *)cookie;
struct wsdisplay_font *font = PICK_FONT(ri, uc);
- STAMP_TYPE *stamp = (STAMP_TYPE *)ri->ri_stamp;
int height, fs;
uint8_t *fr;
STAMP_TYPE *rp, *hp;
@@ -225,7 +224,7 @@ PUTCHAR_WIDTH(RASOPS_DEPTH, RASOPS_WIDTH
return;
/* Recompute stamp? */
- if (attr != ri->ri_stamp_attr)
+ if (attr != stamp_attr || __predict_false(ri != stamp_ri))
MAKESTAMP(RASOPS_DEPTH)(ri, attr);
rp = (STAMP_TYPE *)(ri->ri_bits + row * ri->ri_yscale +
