struct ureg_src samplers[PIPE_MAX_SAMPLERS];
struct ureg_src systemValues[SYSTEM_VALUE_MAX];
- /* Extra info for handling point size clamping in vertex shader */
- struct ureg_dst pointSizeResult; /**< Actual point size output register */
- struct ureg_src pointSizeConst; /**< Point size range constant register */
- GLint pointSizeOutIndex; /**< Temp point size output register */
- GLboolean prevInstWrotePointSize;
-
const GLuint *inputMapping;
const GLuint *outputMapping;
/** Map Mesa's SYSTEM_VALUE_x to TGSI_SEMANTIC_x */
static unsigned mesa_sysval_to_semantic[SYSTEM_VALUE_MAX] = {
TGSI_SEMANTIC_FACE,
+ TGSI_SEMANTIC_VERTEXID,
TGSI_SEMANTIC_INSTANCEID
};
return t->temps[index];
case PROGRAM_OUTPUT:
- if (t->procType == TGSI_PROCESSOR_VERTEX && index == VERT_RESULT_PSIZ)
- t->prevInstWrotePointSize = GL_TRUE;
-
if (t->procType == TGSI_PROCESSOR_VERTEX)
assert(index < VERT_RESULT_MAX);
else if (t->procType == TGSI_PROCESSOR_FRAGMENT)
case PROGRAM_TEMPORARY:
assert(index >= 0);
+ assert(index < Elements(t->temps));
if (ureg_dst_is_undef(t->temps[index]))
t->temps[index] = ureg_DECL_temporary( t->ureg );
- assert(index < Elements(t->temps));
return ureg_src(t->temps[index]);
case PROGRAM_NAMED_PARAM:
/**
* Map mesa texture target to TGSI texture target.
*/
-static unsigned
-translate_texture_target( GLuint textarget,
+unsigned
+st_translate_texture_target( GLuint textarget,
GLboolean shadow )
{
if (shadow) {
case TEXTURE_1D_INDEX: return TGSI_TEXTURE_SHADOW1D;
case TEXTURE_2D_INDEX: return TGSI_TEXTURE_SHADOW2D;
case TEXTURE_RECT_INDEX: return TGSI_TEXTURE_SHADOWRECT;
+ case TEXTURE_1D_ARRAY_INDEX: return TGSI_TEXTURE_SHADOW1D_ARRAY;
+ case TEXTURE_2D_ARRAY_INDEX: return TGSI_TEXTURE_SHADOW2D_ARRAY;
+ case TEXTURE_CUBE_INDEX: return TGSI_TEXTURE_SHADOWCUBE;
default: break;
}
}
case TEXTURE_3D_INDEX: return TGSI_TEXTURE_3D;
case TEXTURE_CUBE_INDEX: return TGSI_TEXTURE_CUBE;
case TEXTURE_RECT_INDEX: return TGSI_TEXTURE_RECT;
+ case TEXTURE_1D_ARRAY_INDEX: return TGSI_TEXTURE_1D_ARRAY;
+ case TEXTURE_2D_ARRAY_INDEX: return TGSI_TEXTURE_2D_ARRAY;
+ case TEXTURE_EXTERNAL_INDEX: return TGSI_TEXTURE_2D;
default:
debug_assert( 0 );
return TGSI_TEXTURE_1D;
static struct ureg_dst
translate_dst( struct st_translate *t,
const struct prog_dst_register *DstReg,
- boolean saturate )
+ boolean saturate,
+ boolean clamp_color)
{
struct ureg_dst dst = dst_register( t,
DstReg->File,
if (saturate)
dst = ureg_saturate( dst );
+ else if (clamp_color && DstReg->File == PROGRAM_OUTPUT) {
+ /* Clamp colors for ARB_color_buffer_float. */
+ switch (t->procType) {
+ case TGSI_PROCESSOR_VERTEX:
+ /* XXX if the geometry shader is present, this must be done there
+ * instead of here. */
+ if (DstReg->Index == VERT_RESULT_COL0 ||
+ DstReg->Index == VERT_RESULT_COL1 ||
+ DstReg->Index == VERT_RESULT_BFC0 ||
+ DstReg->Index == VERT_RESULT_BFC1) {
+ dst = ureg_saturate(dst);
+ }
+ break;
+
+ case TGSI_PROCESSOR_FRAGMENT:
+ if (DstReg->Index >= FRAG_RESULT_COLOR) {
+ dst = ureg_saturate(dst);
+ }
+ break;
+ }
+ }
if (DstReg->RelAddr)
dst = ureg_dst_indirect( dst, ureg_src(t->address[0]) );
static void
compile_instruction(
struct st_translate *t,
- const struct prog_instruction *inst )
+ const struct prog_instruction *inst,
+ boolean clamp_dst_color_output)
{
struct ureg_program *ureg = t->ureg;
GLuint i;
- struct ureg_dst dst[1];
+ struct ureg_dst dst[1] = { { 0 } };
struct ureg_src src[4];
unsigned num_dst;
unsigned num_src;
if (num_dst)
dst[0] = translate_dst( t,
&inst->DstReg,
- inst->SaturateMode );
+ inst->SaturateMode,
+ clamp_dst_color_output);
for (i = 0; i < num_src; i++)
src[i] = translate_src( t, &inst->SrcReg[i] );
ureg_tex_insn( ureg,
translate_opcode( inst->Opcode ),
dst, num_dst,
- translate_texture_target( inst->TexSrcTarget,
+ st_translate_texture_target( inst->TexSrcTarget,
inst->TexShadow ),
+ NULL, 0,
src, num_src );
return;
/**
- * Emit the TGSI instructions to adjust the WPOS pixel center convention
- * Basically, add (adjX, adjY) to the fragment position.
- */
-static void
-emit_adjusted_wpos( struct st_translate *t,
- const struct gl_program *program,
- GLfloat adjX, GLfloat adjY)
-{
- struct ureg_program *ureg = t->ureg;
- struct ureg_dst wpos_temp = ureg_DECL_temporary(ureg);
- struct ureg_src wpos_input = t->inputs[t->inputMapping[FRAG_ATTRIB_WPOS]];
-
- /* Note that we bias X and Y and pass Z and W through unchanged.
- * The shader might also use gl_FragCoord.w and .z.
- */
- ureg_ADD(ureg, wpos_temp, wpos_input,
- ureg_imm4f(ureg, adjX, adjY, 0.0f, 0.0f));
-
- t->inputs[t->inputMapping[FRAG_ATTRIB_WPOS]] = ureg_src(wpos_temp);
-}
-
-
-/**
- * Emit the TGSI instructions for inverting the WPOS y coordinate.
+ * Emit the TGSI instructions for inverting and adjusting WPOS.
* This code is unavoidable because it also depends on whether
* a FBO is bound (STATE_FB_WPOS_Y_TRANSFORM).
*/
static void
-emit_wpos_inversion( struct st_translate *t,
- const struct gl_program *program,
- boolean invert)
+emit_wpos_adjustment( struct st_translate *t,
+ const struct gl_program *program,
+ boolean invert,
+ GLfloat adjX, GLfloat adjY[2])
{
struct ureg_program *ureg = t->ureg;
wposTransformState);
struct ureg_src wpostrans = ureg_DECL_constant( ureg, wposTransConst );
- struct ureg_dst wpos_temp;
+ struct ureg_dst wpos_temp = ureg_DECL_temporary( ureg );
struct ureg_src wpos_input = t->inputs[t->inputMapping[FRAG_ATTRIB_WPOS]];
- /* MOV wpos_temp, input[wpos]
- */
- if (wpos_input.File == TGSI_FILE_TEMPORARY)
- wpos_temp = ureg_dst(wpos_input);
- else {
- wpos_temp = ureg_DECL_temporary( ureg );
+ /* First, apply the coordinate shift: */
+ if (adjX || adjY[0] || adjY[1]) {
+ if (adjY[0] != adjY[1]) {
+ /* Adjust the y coordinate by adjY[1] or adjY[0] respectively
+ * depending on whether inversion is actually going to be applied
+ * or not, which is determined by testing against the inversion
+ * state variable used below, which will be either +1 or -1.
+ */
+ struct ureg_dst adj_temp = ureg_DECL_temporary(ureg);
+
+ ureg_CMP(ureg, adj_temp,
+ ureg_scalar(wpostrans, invert ? 2 : 0),
+ ureg_imm4f(ureg, adjX, adjY[0], 0.0f, 0.0f),
+ ureg_imm4f(ureg, adjX, adjY[1], 0.0f, 0.0f));
+ ureg_ADD(ureg, wpos_temp, wpos_input, ureg_src(adj_temp));
+ } else {
+ ureg_ADD(ureg, wpos_temp, wpos_input,
+ ureg_imm4f(ureg, adjX, adjY[0], 0.0f, 0.0f));
+ }
+ wpos_input = ureg_src(wpos_temp);
+ } else {
+ /* MOV wpos_temp, input[wpos]
+ */
ureg_MOV( ureg, wpos_temp, wpos_input );
}
+ /* Now the conditional y flip: STATE_FB_WPOS_Y_TRANSFORM.xy/zw will be
+ * inversion/identity, or the other way around if we're drawing to an FBO.
+ */
if (invert) {
/* MAD wpos_temp.y, wpos_input, wpostrans.xxxx, wpostrans.yyyy
*/
const struct gl_fragment_program *fp =
(const struct gl_fragment_program *) program;
struct pipe_screen *pscreen = st->pipe->screen;
+ GLfloat adjX = 0.0f;
+ GLfloat adjY[2] = { 0.0f, 0.0f };
boolean invert = FALSE;
+ /* Query the pixel center conventions supported by the pipe driver and set
+ * adjX, adjY to help out if it cannot handle the requested one internally.
+ *
+ * The bias of the y-coordinate depends on whether y-inversion takes place
+ * (adjY[1]) or not (adjY[0]), which is in turn dependent on whether we are
+ * drawing to an FBO (causes additional inversion), and whether the the pipe
+ * driver origin and the requested origin differ (the latter condition is
+ * stored in the 'invert' variable).
+ *
+ * For height = 100 (i = integer, h = half-integer, l = lower, u = upper):
+ *
+ * center shift only:
+ * i -> h: +0.5
+ * h -> i: -0.5
+ *
+ * inversion only:
+ * l,i -> u,i: ( 0.0 + 1.0) * -1 + 100 = 99
+ * l,h -> u,h: ( 0.5 + 0.0) * -1 + 100 = 99.5
+ * u,i -> l,i: (99.0 + 1.0) * -1 + 100 = 0
+ * u,h -> l,h: (99.5 + 0.0) * -1 + 100 = 0.5
+ *
+ * inversion and center shift:
+ * l,i -> u,h: ( 0.0 + 0.5) * -1 + 100 = 99.5
+ * l,h -> u,i: ( 0.5 + 0.5) * -1 + 100 = 99
+ * u,i -> l,h: (99.0 + 0.5) * -1 + 100 = 0.5
+ * u,h -> l,i: (99.5 + 0.5) * -1 + 100 = 0
+ */
if (fp->OriginUpperLeft) {
/* Fragment shader wants origin in upper-left */
if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT)) {
if (fp->PixelCenterInteger) {
/* Fragment shader wants pixel center integer */
- if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER))
+ if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER)) {
/* the driver supports pixel center integer */
+ adjY[1] = 1.0f;
ureg_property_fs_coord_pixel_center(ureg, TGSI_FS_COORD_PIXEL_CENTER_INTEGER);
- else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER))
+ }
+ else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER)) {
/* the driver supports pixel center half integer, need to bias X,Y */
- emit_adjusted_wpos(t, program, 0.5f, invert ? 0.5f : -0.5f);
+ adjX = -0.5f;
+ adjY[0] = -0.5f;
+ adjY[1] = 0.5f;
+ }
else
assert(0);
}
}
else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER)) {
/* the driver supports pixel center integer, need to bias X,Y */
+ adjX = adjY[0] = adjY[1] = 0.5f;
ureg_property_fs_coord_pixel_center(ureg, TGSI_FS_COORD_PIXEL_CENTER_INTEGER);
- emit_adjusted_wpos(t, program, 0.5f, invert ? -0.5f : 0.5f);
}
else
assert(0);
/* we invert after adjustment so that we avoid the MOV to temporary,
* and reuse the adjustment ADD instead */
- emit_wpos_inversion(t, program, invert);
+ emit_wpos_adjustment(t, program, invert, adjX, adjY);
}
const GLuint outputMapping[],
const ubyte outputSemanticName[],
const ubyte outputSemanticIndex[],
- boolean passthrough_edgeflags )
+ boolean passthrough_edgeflags,
+ boolean clamp_color)
{
struct st_translate translate, *t;
unsigned i;
t->inputMapping = inputMapping;
t->outputMapping = outputMapping;
t->ureg = ureg;
- t->pointSizeOutIndex = -1;
- t->prevInstWrotePointSize = GL_FALSE;
/*_mesa_print_program(program);*/
t->outputs[i] = ureg_DECL_output( ureg,
outputSemanticName[i],
outputSemanticIndex[i] );
- if ((outputSemanticName[i] == TGSI_SEMANTIC_PSIZE) && program->Id) {
- /* Writing to the point size result register requires special
- * handling to implement clamping.
- */
- static const gl_state_index pointSizeClampState[STATE_LENGTH]
- = { STATE_INTERNAL, STATE_POINT_SIZE_IMPL_CLAMP, 0, 0, 0 };
- /* XXX: note we are modifying the incoming shader here! Need to
- * do this before emitting the constant decls below, or this
- * will be missed:
- */
- unsigned pointSizeClampConst =
- _mesa_add_state_reference(program->Parameters,
- pointSizeClampState);
- struct ureg_dst psizregtemp = ureg_DECL_temporary( ureg );
- t->pointSizeConst = ureg_DECL_constant( ureg, pointSizeClampConst );
- t->pointSizeResult = t->outputs[i];
- t->pointSizeOutIndex = i;
- t->outputs[i] = psizregtemp;
- }
}
if (passthrough_edgeflags)
emit_edgeflags( t, program );
if (sysInputs & (1 << i)) {
unsigned semName = mesa_sysval_to_semantic[i];
t->systemValues[i] = ureg_DECL_system_value(ureg, numSys, semName, 0);
+ if (semName == TGSI_SEMANTIC_INSTANCEID ||
+ semName == TGSI_SEMANTIC_VERTEXID) {
+ /* From Gallium perspective, these system values are always
+ * integer, and require native integer support. However, if
+ * native integer is supported on the vertex stage but not the
+ * pixel stage (e.g, i915g + draw), Mesa will generate IR that
+ * assumes these system values are floats. To resolve the
+ * inconsistency, we insert a U2F.
+ */
+ struct st_context *st = st_context(ctx);
+ struct pipe_screen *pscreen = st->pipe->screen;
+ assert(procType == TGSI_PROCESSOR_VERTEX);
+ assert(pscreen->get_shader_param(pscreen, PIPE_SHADER_VERTEX, PIPE_SHADER_CAP_INTEGERS));
+ if (!ctx->Const.NativeIntegers) {
+ struct ureg_dst temp = ureg_DECL_local_temporary(t->ureg);
+ ureg_U2F( t->ureg, ureg_writemask(temp, TGSI_WRITEMASK_X), t->systemValues[i]);
+ t->systemValues[i] = ureg_scalar(ureg_src(temp), 0);
+ }
+ }
numSys++;
sysInputs &= ~(1 << i);
}
else
t->constants[i] =
ureg_DECL_immediate( ureg,
- program->Parameters->ParameterValues[i],
+ (const float*) program->Parameters->ParameterValues[i],
4 );
break;
default:
*/
for (i = 0; i < program->NumInstructions; i++) {
set_insn_start( t, ureg_get_instruction_number( ureg ));
- compile_instruction( t, &program->Instructions[i] );
-
- if (t->prevInstWrotePointSize && program->Id) {
- /* The previous instruction wrote to the (fake) vertex point size
- * result register. Now we need to clamp that value to the min/max
- * point size range, putting the result into the real point size
- * register.
- * Note that we can't do this easily at the end of program due to
- * possible early return.
- */
- set_insn_start( t, ureg_get_instruction_number( ureg ));
- ureg_MAX( t->ureg,
- ureg_writemask(t->outputs[t->pointSizeOutIndex], WRITEMASK_X),
- ureg_src(t->outputs[t->pointSizeOutIndex]),
- ureg_swizzle(t->pointSizeConst, 1,1,1,1));
- ureg_MIN( t->ureg, ureg_writemask(t->pointSizeResult, WRITEMASK_X),
- ureg_src(t->outputs[t->pointSizeOutIndex]),
- ureg_swizzle(t->pointSizeConst, 2,2,2,2));
- }
- t->prevInstWrotePointSize = GL_FALSE;
+ compile_instruction( t, &program->Instructions[i], clamp_color );
}
/* Fix up all emitted labels:
projects / mesa.git / blobdiff