summaryrefslogtreecommitdiffstats
path: root/src
diff options
context:
space:
mode:
Diffstat (limited to 'src')
-rw-r--r--src/gallium/auxiliary/tgsi/tgsi_exec.c44
-rw-r--r--src/gallium/auxiliary/tgsi/tgsi_exec.h13
-rw-r--r--src/gallium/drivers/softpipe/sp_clear.c4
-rw-r--r--src/mesa/drivers/common/meta.c9
-rw-r--r--src/mesa/drivers/dri/intel/intel_pixel_draw.c16
-rw-r--r--src/mesa/drivers/dri/radeon/radeon_common.c4
-rw-r--r--src/mesa/drivers/dri/radeon/radeon_debug.c10
-rw-r--r--src/mesa/drivers/dri/radeon/radeon_debug.h60
-rw-r--r--src/mesa/main/texenvprogram.c61
9 files changed, 154 insertions, 67 deletions
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index 60ffa188df6..c79c56debd6 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -2756,19 +2756,32 @@ exec_instruction(
if (mach->ExecMask) {
/* do the call */
- /* push the Cond, Loop, Cont stacks */
+ /* First, record the depths of the execution stacks.
+ * This is important for deeply nested/looped return statements.
+ * We have to unwind the stacks by the correct amount. For a
+ * real code generator, we could determine the number of entries
+ * to pop off each stack with simple static analysis and avoid
+ * implementing this data structure at run time.
+ */
+ mach->CallStack[mach->CallStackTop].CondStackTop = mach->CondStackTop;
+ mach->CallStack[mach->CallStackTop].LoopStackTop = mach->LoopStackTop;
+ mach->CallStack[mach->CallStackTop].ContStackTop = mach->ContStackTop;
+ /* note that PC was already incremented above */
+ mach->CallStack[mach->CallStackTop].ReturnAddr = *pc;
+
+ mach->CallStackTop++;
+
+ /* Second, push the Cond, Loop, Cont, Func stacks */
assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING);
mach->CondStack[mach->CondStackTop++] = mach->CondMask;
assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask;
assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
mach->ContStack[mach->ContStackTop++] = mach->ContMask;
-
assert(mach->FuncStackTop < TGSI_EXEC_MAX_CALL_NESTING);
mach->FuncStack[mach->FuncStackTop++] = mach->FuncMask;
- /* note that PC was already incremented above */
- mach->CallStack[mach->CallStackTop++] = *pc;
+ /* Finally, jump to the subroutine */
*pc = inst->InstructionExtLabel.Label;
}
break;
@@ -2785,18 +2798,24 @@ exec_instruction(
*pc = -1;
return;
}
- *pc = mach->CallStack[--mach->CallStackTop];
- /* pop the Cond, Loop, Cont stacks */
- assert(mach->CondStackTop > 0);
- mach->CondMask = mach->CondStack[--mach->CondStackTop];
- assert(mach->LoopStackTop > 0);
- mach->LoopMask = mach->LoopStack[--mach->LoopStackTop];
- assert(mach->ContStackTop > 0);
- mach->ContMask = mach->ContStack[--mach->ContStackTop];
+ assert(mach->CallStackTop > 0);
+ mach->CallStackTop--;
+
+ mach->CondStackTop = mach->CallStack[mach->CallStackTop].CondStackTop;
+ mach->CondMask = mach->CondStack[mach->CondStackTop];
+
+ mach->LoopStackTop = mach->CallStack[mach->CallStackTop].LoopStackTop;
+ mach->LoopMask = mach->LoopStack[mach->LoopStackTop];
+
+ mach->ContStackTop = mach->CallStack[mach->CallStackTop].ContStackTop;
+ mach->ContMask = mach->ContStack[mach->ContStackTop];
+
assert(mach->FuncStackTop > 0);
mach->FuncMask = mach->FuncStack[--mach->FuncStackTop];
+ *pc = mach->CallStack[mach->CallStackTop].ReturnAddr;
+
UPDATE_EXEC_MASK(mach);
}
break;
@@ -3245,7 +3264,6 @@ tgsi_exec_machine_run( struct tgsi_exec_machine *mach )
mach->FuncMask = 0xf;
mach->ExecMask = 0xf;
- mach->CondStackTop = 0; /* temporarily subvert this assertion */
assert(mach->CondStackTop == 0);
assert(mach->LoopStackTop == 0);
assert(mach->ContStackTop == 0);
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.h b/src/gallium/auxiliary/tgsi/tgsi_exec.h
index 3baa94dbdde..c72f76809d4 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.h
@@ -186,6 +186,17 @@ struct tgsi_exec_labels
*/
#define TGSI_EXEC_MAX_CONST_BUFFER 4096
+
+/** function call/activation record */
+struct tgsi_call_record
+{
+ uint CondStackTop;
+ uint LoopStackTop;
+ uint ContStackTop;
+ uint ReturnAddr;
+};
+
+
/**
* Run-time virtual machine state for executing TGSI shader.
*/
@@ -249,7 +260,7 @@ struct tgsi_exec_machine
int FuncStackTop;
/** Function call stack for saving/restoring the program counter */
- uint CallStack[TGSI_EXEC_MAX_CALL_NESTING];
+ struct tgsi_call_record CallStack[TGSI_EXEC_MAX_CALL_NESTING];
int CallStackTop;
struct tgsi_full_instruction *Instructions;
diff --git a/src/gallium/drivers/softpipe/sp_clear.c b/src/gallium/drivers/softpipe/sp_clear.c
index fa59277438c..d3af18e162b 100644
--- a/src/gallium/drivers/softpipe/sp_clear.c
+++ b/src/gallium/drivers/softpipe/sp_clear.c
@@ -85,5 +85,7 @@ softpipe_clear(struct pipe_context *pipe, unsigned buffers, const float *rgba,
/* non-cached surface */
pipe->surface_fill(pipe, ps, 0, 0, ps->width, ps->height, cv);
#endif
- }
+ }
+
+ softpipe->dirty_render_cache = TRUE;
}
diff --git a/src/mesa/drivers/common/meta.c b/src/mesa/drivers/common/meta.c
index 276e981a94c..0b9781027e7 100644
--- a/src/mesa/drivers/common/meta.c
+++ b/src/mesa/drivers/common/meta.c
@@ -1522,6 +1522,8 @@ _mesa_meta_draw_pixels(GLcontext *ctx,
}
else if (_mesa_is_stencil_format(format)) {
if (ctx->Extensions.ARB_fragment_program &&
+ ctx->Pixel.IndexShift == 0 &&
+ ctx->Pixel.IndexOffset == 0 &&
type == GL_UNSIGNED_BYTE) {
/* We'll store stencil as alpha. This only works for GLubyte
* image data because of how incoming values are mapped to alpha
@@ -1656,6 +1658,13 @@ _mesa_meta_draw_pixels(GLcontext *ctx,
_mesa_ColorMask(GL_FALSE, GL_FALSE, GL_FALSE, GL_FALSE);
_mesa_set_enable(ctx, GL_STENCIL_TEST, GL_TRUE);
+
+ /* set all stencil bits to 0 */
+ _mesa_StencilOp(GL_REPLACE, GL_REPLACE, GL_REPLACE);
+ _mesa_StencilFunc(GL_ALWAYS, 0, 255);
+ _mesa_DrawArrays(GL_TRIANGLE_FAN, 0, 4);
+
+ /* set stencil bits to 1 where needed */
_mesa_StencilOp(GL_KEEP, GL_KEEP, GL_REPLACE);
_mesa_BindProgram(GL_FRAGMENT_PROGRAM_ARB, drawpix->StencilFP);
diff --git a/src/mesa/drivers/dri/intel/intel_pixel_draw.c b/src/mesa/drivers/dri/intel/intel_pixel_draw.c
index 8c113881d6b..7fbb89fd6aa 100644
--- a/src/mesa/drivers/dri/intel/intel_pixel_draw.c
+++ b/src/mesa/drivers/dri/intel/intel_pixel_draw.c
@@ -42,6 +42,7 @@
#include "main/depth.h"
#include "main/hash.h"
#include "main/blend.h"
+#include "swrast/swrast.h"
#include "drivers/common/meta.h"
#include "intel_context.h"
@@ -260,9 +261,24 @@ intelDrawPixels(GLcontext * ctx,
const struct gl_pixelstore_attrib *unpack,
const GLvoid * pixels)
{
+#if 0
+ /* XXX this function doesn't seem to work reliably even when all
+ * the pre-requisite conditions are met.
+ * Note that this function is never hit with conform.
+ * Fall back to swrast because even the _mesa_meta_draw_pixels() approach
+ * isn't working because of an apparent stencil bug.
+ */
if (intel_stencil_drawpixels(ctx, x, y, width, height, format, type,
unpack, pixels))
return;
+#else
+ (void) intel_stencil_drawpixels; /* silence warning */
+ if (format == GL_STENCIL_INDEX) {
+ _swrast_DrawPixels(ctx, x, y, width, height, format, type,
+ unpack, pixels);
+ return;
+ }
+#endif
_mesa_meta_draw_pixels(ctx, x, y, width, height, format, type,
unpack, pixels);
diff --git a/src/mesa/drivers/dri/radeon/radeon_common.c b/src/mesa/drivers/dri/radeon/radeon_common.c
index e53eb0904d2..a4c7b40798a 100644
--- a/src/mesa/drivers/dri/radeon/radeon_common.c
+++ b/src/mesa/drivers/dri/radeon/radeon_common.c
@@ -1234,7 +1234,9 @@ int rcommonFlushCmdBuf(radeonContextPtr rmesa, const char *caller)
UNLOCK_HARDWARE(rmesa);
if (ret) {
- fprintf(stderr, "drmRadeonCmdBuffer: %d\n", ret);
+ fprintf(stderr, "drmRadeonCmdBuffer: %d. Kernel failed to "
+ "parse or rejected command stream. See dmesg "
+ "for more info.\n", ret);
_mesa_exit(ret);
}
diff --git a/src/mesa/drivers/dri/radeon/radeon_debug.c b/src/mesa/drivers/dri/radeon/radeon_debug.c
index a1ed39683f1..3b6f0038037 100644
--- a/src/mesa/drivers/dri/radeon/radeon_debug.c
+++ b/src/mesa/drivers/dri/radeon/radeon_debug.c
@@ -32,6 +32,9 @@
#include "radeon_debug.h"
#include "radeon_common_context.h"
+#include <stdarg.h>
+#include <stdio.h>
+
static const struct dri_debug_control debug_control[] = {
{"fall", RADEON_FALLBACKS},
{"tex", RADEON_TEXTURE},
@@ -85,10 +88,10 @@ void _radeon_debug_remove_indent(void)
}
}
-extern void _radeon_print(const radeon_debug_type_t type,
+void _radeon_print(const radeon_debug_type_t type,
const radeon_debug_level_t level,
const char* message,
- va_list values)
+ ...)
{
GET_CURRENT_CONTEXT(ctx);
if (ctx) {
@@ -97,5 +100,8 @@ extern void _radeon_print(const radeon_debug_type_t type,
if (radeon->debug.indent_depth)
fprintf(stderr, "%s", radeon->debug.indent);
}
+ va_list values;
+ va_start( values, message );
vfprintf(stderr, message, values);
+ va_end( values );
}
diff --git a/src/mesa/drivers/dri/radeon/radeon_debug.h b/src/mesa/drivers/dri/radeon/radeon_debug.h
index 132e27351da..2a8302293b2 100644
--- a/src/mesa/drivers/dri/radeon/radeon_debug.h
+++ b/src/mesa/drivers/dri/radeon/radeon_debug.h
@@ -30,8 +30,7 @@
#ifndef RADEON_DEBUG_H_INCLUDED
#define RADEON_DEBUG_H_INCLUDED
-#include <stdarg.h>
-#include <stdio.h>
+#include <stdlib.h>
typedef enum radeon_debug_levels {
RADEON_CRITICAL = 0, /* Only errors */
@@ -102,57 +101,36 @@ static inline int radeon_is_debug_enabled(const radeon_debug_type_t type,
extern void _radeon_print(const radeon_debug_type_t type,
const radeon_debug_level_t level,
const char* message,
- va_list values);
-/**
- * Format attribute requires declaration for setting it. Don't ask me why!
- */
-static inline void radeon_print(const radeon_debug_type_t type,
- const radeon_debug_level_t level,
- const char* message,
- ...) __attribute__((format(printf,3,4)));
-
+ ...) __attribute__((format(printf,3,4)));
/**
* Print out debug message if channel specified by type is enabled
* and compile time debugging level is at least as high as level parameter
*/
-static inline void radeon_print(const radeon_debug_type_t type,
- const radeon_debug_level_t level,
- const char* message,
- ...)
-{
- /* Compile out if level of message is too high */
- if (radeon_is_debug_enabled(type, level)) {
-
- va_list values;
- va_start( values, message );
- _radeon_print(type, level, message, values);
- va_end( values );
- }
-}
+#define radeon_print(type, level, message, ...) do { \
+ const radeon_debug_level_t _debug_level = (level); \
+ const radeon_debug_type_t _debug_type = (type); \
+ /* Compile out if level of message is too high */ \
+ if (radeon_is_debug_enabled(type, level)) { \
+ _radeon_print(_debug_type, _debug_level, \
+ (message), ## __VA_ARGS__); \
+ } \
+} while(0)
-static inline void radeon_error(const char* message, ...) __attribute__((format(printf,1,2)));
/**
* printf style function for writing error messages.
*/
-static inline void radeon_error(const char* message, ...)
-{
- va_list values;
- va_start( values, message );
- radeon_print(RADEON_GENERAL, RADEON_CRITICAL, message, values);
- va_end( values );
-}
+#define radeon_error(message, ...) do { \
+ radeon_print(RADEON_GENERAL, RADEON_CRITICAL, \
+ (message), ## __VA_ARGS__); \
+} while(0)
-static inline void radeon_warning(const char* message, ...) __attribute__((format(printf,1,2)));
/**
* printf style function for writing warnings.
*/
-static inline void radeon_warning(const char* message, ...)
-{
- va_list values;
- va_start( values, message );
- radeon_print(RADEON_GENERAL, RADEON_IMPORTANT, message, values);
- va_end( values );
-}
+#define radeon_warning(message, ...) do { \
+ radeon_print(RADEON_GENERAL, RADEON_IMPORTANT, \
+ (message), ## __VA_ARGS__); \
+} while(0)
extern void radeon_init_debug(void);
extern void _radeon_debug_add_indent(void);
diff --git a/src/mesa/main/texenvprogram.c b/src/mesa/main/texenvprogram.c
index 870970a3554..2f3e47e69ec 100644
--- a/src/mesa/main/texenvprogram.c
+++ b/src/mesa/main/texenvprogram.c
@@ -248,6 +248,40 @@ static GLuint translate_mode( GLenum envMode, GLenum mode )
/**
+ * Do we need to clamp the results of the given texture env/combine mode?
+ * If the inputs to the mode are in [0,1] we don't always have to clamp
+ * the results.
+ */
+static GLboolean
+need_saturate( GLuint mode )
+{
+ switch (mode) {
+ case MODE_REPLACE:
+ case MODE_MODULATE:
+ case MODE_INTERPOLATE:
+ return GL_FALSE;
+ case MODE_ADD:
+ case MODE_ADD_SIGNED:
+ case MODE_SUBTRACT:
+ case MODE_DOT3_RGB:
+ case MODE_DOT3_RGB_EXT:
+ case MODE_DOT3_RGBA:
+ case MODE_DOT3_RGBA_EXT:
+ case MODE_MODULATE_ADD_ATI:
+ case MODE_MODULATE_SIGNED_ADD_ATI:
+ case MODE_MODULATE_SUBTRACT_ATI:
+ case MODE_ADD_PRODUCTS:
+ case MODE_ADD_PRODUCTS_SIGNED:
+ case MODE_BUMP_ENVMAP_ATI:
+ return GL_TRUE;
+ default:
+ assert(0);
+ }
+}
+
+
+
+/**
* Translate TEXTURE_x_BIT to TEXTURE_x_INDEX.
*/
static GLuint translate_tex_src_bit( GLbitfield bit )
@@ -1116,7 +1150,7 @@ static struct ureg
emit_texenv(struct texenv_fragment_program *p, GLuint unit)
{
const struct state_key *key = p->state;
- GLboolean saturate;
+ GLboolean rgb_saturate, alpha_saturate;
GLuint rgb_shift, alpha_shift;
struct ureg out, dest;
@@ -1146,7 +1180,19 @@ emit_texenv(struct texenv_fragment_program *p, GLuint unit)
/* If we'll do rgb/alpha shifting don't saturate in emit_combine().
* We don't want to clamp twice.
*/
- saturate = !(rgb_shift || alpha_shift);
+ if (rgb_shift)
+ rgb_saturate = GL_FALSE; /* saturate after rgb shift */
+ else if (need_saturate(key->unit[unit].ModeRGB))
+ rgb_saturate = GL_TRUE;
+ else
+ rgb_saturate = GL_FALSE;
+
+ if (alpha_shift)
+ alpha_saturate = GL_FALSE; /* saturate after alpha shift */
+ else if (need_saturate(key->unit[unit].ModeA))
+ alpha_saturate = GL_TRUE;
+ else
+ alpha_saturate = GL_FALSE;
/* If this is the very last calculation, emit direct to output reg:
*/
@@ -1162,7 +1208,7 @@ emit_texenv(struct texenv_fragment_program *p, GLuint unit)
*/
if (key->unit[unit].ModeRGB == key->unit[unit].ModeA &&
args_match(key, unit)) {
- out = emit_combine( p, dest, WRITEMASK_XYZW, saturate,
+ out = emit_combine( p, dest, WRITEMASK_XYZW, rgb_saturate,
unit,
key->unit[unit].NumArgsRGB,
key->unit[unit].ModeRGB,
@@ -1170,7 +1216,7 @@ emit_texenv(struct texenv_fragment_program *p, GLuint unit)
}
else if (key->unit[unit].ModeRGB == MODE_DOT3_RGBA_EXT ||
key->unit[unit].ModeRGB == MODE_DOT3_RGBA) {
- out = emit_combine( p, dest, WRITEMASK_XYZW, saturate,
+ out = emit_combine( p, dest, WRITEMASK_XYZW, rgb_saturate,
unit,
key->unit[unit].NumArgsRGB,
key->unit[unit].ModeRGB,
@@ -1180,12 +1226,12 @@ emit_texenv(struct texenv_fragment_program *p, GLuint unit)
/* Need to do something to stop from re-emitting identical
* argument calculations here:
*/
- out = emit_combine( p, dest, WRITEMASK_XYZ, saturate,
+ out = emit_combine( p, dest, WRITEMASK_XYZ, rgb_saturate,
unit,
key->unit[unit].NumArgsRGB,
key->unit[unit].ModeRGB,
key->unit[unit].OptRGB);
- out = emit_combine( p, dest, WRITEMASK_W, saturate,
+ out = emit_combine( p, dest, WRITEMASK_W, alpha_saturate,
unit,
key->unit[unit].NumArgsA,
key->unit[unit].ModeA,
@@ -1196,8 +1242,7 @@ emit_texenv(struct texenv_fragment_program *p, GLuint unit)
*/
if (alpha_shift || rgb_shift) {
struct ureg shift;
-
- saturate = GL_TRUE; /* always saturate at this point */
+ GLboolean saturate = GL_TRUE; /* always saturate at this point */
if (rgb_shift == alpha_shift) {
shift = register_scalar_const(p, (GLfloat)(1<<rgb_shift));
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-16 05:40:47 +0000