#include "brw_context.h"
#include "brw_wm.h"
+static GLboolean can_do_pln(struct intel_context *intel,
+ const struct brw_reg *deltas)
+{
+ struct brw_context *brw = brw_context(&intel->ctx);
+
+ if (!brw->has_pln)
+ return GL_FALSE;
+
+ if (deltas[1].nr != deltas[0].nr + 1)
+ return GL_FALSE;
+
+ if (intel->gen < 6 && ((deltas[0].nr & 1) != 0))
+ return GL_FALSE;
+
+ return GL_TRUE;
+}
+
/* Not quite sure how correct this is - need to understand horiz
* vs. vertical strides a little better.
*/
return reg;
}
+/* Return the SrcReg index of the channels that can be immediate float operands
+ * instead of usage of PROGRAM_CONSTANT values through push/pull.
+ */
+GLboolean
+brw_wm_arg_can_be_immediate(enum prog_opcode opcode, int arg)
+{
+ int opcode_array[] = {
+ [OPCODE_ADD] = 2,
+ [OPCODE_CMP] = 3,
+ [OPCODE_DP3] = 2,
+ [OPCODE_DP4] = 2,
+ [OPCODE_DPH] = 2,
+ [OPCODE_MAX] = 2,
+ [OPCODE_MIN] = 2,
+ [OPCODE_MOV] = 1,
+ [OPCODE_MUL] = 2,
+ [OPCODE_SEQ] = 2,
+ [OPCODE_SGE] = 2,
+ [OPCODE_SGT] = 2,
+ [OPCODE_SLE] = 2,
+ [OPCODE_SLT] = 2,
+ [OPCODE_SNE] = 2,
+ [OPCODE_XPD] = 2,
+ };
+
+ /* These opcodes get broken down in a way that allow two
+ * args to be immediates.
+ */
+ if (opcode == OPCODE_MAD || opcode == OPCODE_LRP) {
+ if (arg == 1 || arg == 2)
+ return GL_TRUE;
+ }
+
+ if (opcode > ARRAY_SIZE(opcode_array))
+ return GL_FALSE;
-/* Payload R0:
+ return arg == opcode_array[opcode] - 1;
+}
+
+/**
+ * Computes the screen-space x,y position of the pixels.
+ *
+ * This will be used by emit_delta_xy() or emit_wpos_xy() for
+ * interpolation of attributes..
+ *
+ * Payload R0:
*
* R0.0 -- pixel mask, one bit for each of 4 pixels in 4 tiles,
* corresponding to each of the 16 execution channels.
* R1.7 -- ?
* R1.8 -- ?
*/
-
void emit_pixel_xy(struct brw_wm_compile *c,
const struct brw_reg *dst,
GLuint mask)
brw_pop_insn_state(p);
}
-
+/**
+ * Computes the screen-space x,y distance of the pixels from the start
+ * vertex.
+ *
+ * This will be used in linterp or pinterp with the start vertex value
+ * and the Cx, Cy, and C0 coefficients passed in from the setup engine
+ * to produce interpolated attribute values.
+ */
void emit_delta_xy(struct brw_compile *p,
const struct brw_reg *dst,
GLuint mask,
{
struct brw_reg r1 = brw_vec1_grf(1, 0);
- /* Calc delta X,Y by subtracting origin in r1 from the pixel
- * centers.
- */
- if (mask & WRITEMASK_X) {
- brw_ADD(p,
- dst[0],
- retype(arg0[0], BRW_REGISTER_TYPE_UW),
- negate(r1));
- }
+ if (mask == 0)
+ return;
- if (mask & WRITEMASK_Y) {
- brw_ADD(p,
- dst[1],
- retype(arg0[1], BRW_REGISTER_TYPE_UW),
- negate(suboffset(r1,1)));
+ assert(mask == WRITEMASK_XY);
- }
+ /* Calc delta X,Y by subtracting origin in r1 from the pixel
+ * centers produced by emit_pixel_xy().
+ */
+ brw_ADD(p,
+ dst[0],
+ retype(arg0[0], BRW_REGISTER_TYPE_UW),
+ negate(r1));
+ brw_ADD(p,
+ dst[1],
+ retype(arg0[1], BRW_REGISTER_TYPE_UW),
+ negate(suboffset(r1,1)));
}
+/**
+ * Computes the pixel offset from the window origin for gl_FragCoord().
+ */
void emit_wpos_xy(struct brw_wm_compile *c,
const struct brw_reg *dst,
GLuint mask,
{
struct brw_compile *p = &c->func;
- /* Calculate the pixel offset from window bottom left into destination
- * X and Y channels.
- */
if (mask & WRITEMASK_X) {
- /* X' = X - origin */
- brw_ADD(p,
- dst[0],
- retype(arg0[0], BRW_REGISTER_TYPE_W),
- brw_imm_d(0 - c->key.origin_x));
+ if (c->fp->program.PixelCenterInteger) {
+ /* X' = X */
+ brw_MOV(p,
+ dst[0],
+ retype(arg0[0], BRW_REGISTER_TYPE_W));
+ } else {
+ /* X' = X + 0.5 */
+ brw_ADD(p,
+ dst[0],
+ retype(arg0[0], BRW_REGISTER_TYPE_W),
+ brw_imm_f(0.5));
+ }
}
if (mask & WRITEMASK_Y) {
- /* Y' = height - (Y - origin_y) = height + origin_y - Y */
- brw_ADD(p,
- dst[1],
- negate(retype(arg0[1], BRW_REGISTER_TYPE_W)),
- brw_imm_d(c->key.origin_y + c->key.drawable_height - 1));
+ if (c->fp->program.OriginUpperLeft) {
+ if (c->fp->program.PixelCenterInteger) {
+ /* Y' = Y */
+ brw_MOV(p,
+ dst[1],
+ retype(arg0[1], BRW_REGISTER_TYPE_W));
+ } else {
+ /* Y' = Y + 0.5 */
+ brw_ADD(p,
+ dst[1],
+ retype(arg0[1], BRW_REGISTER_TYPE_W),
+ brw_imm_f(0.5));
+ }
+ } else {
+ float center_offset = c->fp->program.PixelCenterInteger ? 0.0 : 0.5;
+
+ /* Y' = (height - 1) - Y + center */
+ brw_ADD(p,
+ dst[1],
+ negate(retype(arg0[1], BRW_REGISTER_TYPE_W)),
+ brw_imm_f(c->key.drawable_height - 1 + center_offset));
+ }
}
}
const struct brw_reg *deltas)
{
struct brw_compile *p = &c->func;
+ struct intel_context *intel = &p->brw->intel;
/* Don't need this if all you are doing is interpolating color, for
* instance.
/* Calc 1/w - just linterp wpos[3] optimized by putting the
* result straight into a message reg.
*/
- brw_LINE(p, brw_null_reg(), interp3, deltas[0]);
- brw_MAC(p, brw_message_reg(2), suboffset(interp3, 1), deltas[1]);
+ if (can_do_pln(intel, deltas)) {
+ brw_PLN(p, brw_message_reg(2), interp3, deltas[0]);
+ } else {
+ brw_LINE(p, brw_null_reg(), interp3, deltas[0]);
+ brw_MAC(p, brw_message_reg(2), suboffset(interp3, 1), deltas[1]);
+ }
/* Calc w */
if (c->dispatch_width == 16) {
const struct brw_reg *arg0,
const struct brw_reg *deltas)
{
+ struct intel_context *intel = &p->brw->intel;
struct brw_reg interp[4];
GLuint nr = arg0[0].nr;
GLuint i;
for (i = 0; i < 4; i++) {
if (mask & (1<<i)) {
- brw_LINE(p, brw_null_reg(), interp[i], deltas[0]);
- brw_MAC(p, dst[i], suboffset(interp[i],1), deltas[1]);
+ if (can_do_pln(intel, deltas)) {
+ brw_PLN(p, dst[i], interp[i], deltas[0]);
+ } else {
+ brw_LINE(p, brw_null_reg(), interp[i], deltas[0]);
+ brw_MAC(p, dst[i], suboffset(interp[i],1), deltas[1]);
+ }
}
}
}
const struct brw_reg *deltas,
const struct brw_reg *w)
{
+ struct intel_context *intel = &p->brw->intel;
struct brw_reg interp[4];
GLuint nr = arg0[0].nr;
GLuint i;
for (i = 0; i < 4; i++) {
if (mask & (1<<i)) {
- brw_LINE(p, brw_null_reg(), interp[i], deltas[0]);
- brw_MAC(p, dst[i], suboffset(interp[i],1), deltas[1]);
+ if (can_do_pln(intel, deltas)) {
+ brw_PLN(p, dst[i], interp[i], deltas[0]);
+ } else {
+ brw_LINE(p, brw_null_reg(), interp[i], deltas[0]);
+ brw_MAC(p, dst[i], suboffset(interp[i],1), deltas[1]);
+ }
}
}
for (i = 0; i < 4; i++) {
emit_sop(p, dst, mask, BRW_CONDITIONAL_NEQ, arg0, arg1);
}
-static void emit_cmp( struct brw_compile *p,
- const struct brw_reg *dst,
- GLuint mask,
- const struct brw_reg *arg0,
- const struct brw_reg *arg1,
- const struct brw_reg *arg2 )
+void emit_cmp(struct brw_compile *p,
+ const struct brw_reg *dst,
+ GLuint mask,
+ const struct brw_reg *arg0,
+ const struct brw_reg *arg1,
+ const struct brw_reg *arg2)
{
GLuint i;
for (i = 0; i < 4; i++) {
if (mask & (1<<i)) {
- brw_set_saturate(p, (mask & SATURATE) ? 1 : 0);
- brw_MOV(p, dst[i], arg2[i]);
- brw_set_saturate(p, 0);
-
brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, arg0[i], brw_imm_f(0));
brw_set_saturate(p, (mask & SATURATE) ? 1 : 0);
- brw_MOV(p, dst[i], arg1[i]);
+ brw_SEL(p, dst[i], arg1[i], arg2[i]);
brw_set_saturate(p, 0);
brw_set_predicate_control_flag_value(p, 0xff);
}
for (i = 0; i < 4; i++) {
if (mask & (1<<i)) {
- brw_set_saturate(p, (mask & SATURATE) ? 1 : 0);
- brw_MOV(p, dst[i], arg0[i]);
- brw_set_saturate(p, 0);
-
- brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, arg0[i], arg1[i]);
+ brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_GE, arg0[i], arg1[i]);
brw_set_saturate(p, (mask & SATURATE) ? 1 : 0);
- brw_MOV(p, dst[i], arg1[i]);
+ brw_SEL(p, dst[i], arg0[i], arg1[i]);
brw_set_saturate(p, 0);
brw_set_predicate_control_flag_value(p, 0xff);
}
for (i = 0; i < 4; i++) {
if (mask & (1<<i)) {
- brw_set_saturate(p, (mask & SATURATE) ? 1 : 0);
- brw_MOV(p, dst[i], arg1[i]);
- brw_set_saturate(p, 0);
-
brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, arg0[i], arg1[i]);
brw_set_saturate(p, (mask & SATURATE) ? 1 : 0);
- brw_MOV(p, dst[i], arg0[i]);
+ brw_SEL(p, dst[i], arg0[i], arg1[i]);
brw_set_saturate(p, 0);
brw_set_predicate_control_flag_value(p, 0xff);
}
{
GLuint i;
- assert(!(mask & WRITEMASK_W) == WRITEMASK_X);
+ assert((mask & WRITEMASK_W) != WRITEMASK_W);
for (i = 0 ; i < 3; i++) {
if (mask & (1<<i)) {
GLboolean shadow)
{
struct brw_compile *p = &c->func;
+ struct intel_context *intel = &p->brw->intel;
struct brw_reg dst_retyped;
GLuint cur_mrf = 2, response_length;
GLuint i, nr_texcoords;
}
/* Pre-Ironlake, the 8-wide sampler always took u,v,r. */
- if (!BRW_IS_IGDNG(p->brw) && c->dispatch_width == 8)
+ if (!intel->is_ironlake && c->dispatch_width == 8)
nr_texcoords = 3;
/* For shadow comparisons, we have to supply u,v,r. */
/* Fill in the shadow comparison reference value. */
if (shadow) {
- if (BRW_IS_IGDNG(p->brw)) {
+ if (intel->is_ironlake) {
/* Fill in the cube map array index value. */
brw_MOV(p, brw_message_reg(cur_mrf), brw_imm_f(0));
cur_mrf += mrf_per_channel;
cur_mrf += mrf_per_channel;
}
- if (BRW_IS_IGDNG(p->brw)) {
+ if (intel->is_ironlake) {
if (shadow)
msg_type = BRW_SAMPLER_MESSAGE_SAMPLE_COMPARE_IGDNG;
else
GLuint sampler)
{
struct brw_compile *p = &c->func;
+ struct intel_context *intel = &p->brw->intel;
GLuint msgLength;
GLuint msg_type;
GLuint mrf_per_channel;
* undefined, and trust the execution mask to keep the undefined pixels
* from mattering.
*/
- if (c->dispatch_width == 16 || !BRW_IS_IGDNG(p->brw)) {
- if (BRW_IS_IGDNG(p->brw))
+ if (c->dispatch_width == 16 || !intel->is_ironlake) {
+ if (intel->is_ironlake)
msg_type = BRW_SAMPLER_MESSAGE_SAMPLE_BIAS_IGDNG;
else
msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS;
{
struct brw_compile *p = &c->func;
struct brw_reg r0uw = retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UW);
- GLuint i;
-
- /* XXX - usually won't need 4 compares!
- */
+ GLuint i, j;
+
for (i = 0; i < 4; i++) {
+ /* Check if we've already done the comparison for this reg
+ * -- common when someone does KIL TEMP.wwww.
+ */
+ for (j = 0; j < i; j++) {
+ if (memcmp(&arg0[j], &arg0[i], sizeof(arg0[0])) == 0)
+ break;
+ }
+ if (j != i)
+ continue;
+
brw_push_insn_state(p);
brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_GE, arg0[i], brw_imm_f(0));
brw_set_predicate_control_flag_value(p, 0xff);
brw_push_insn_state(p);
brw_set_mask_control(p, BRW_MASK_DISABLE);
- brw_NOT(p, c->emit_mask_reg, brw_mask_reg(1)); //IMASK
+ brw_NOT(p, c->emit_mask_reg, brw_mask_reg(1)); /* IMASK */
brw_AND(p, r0uw, c->emit_mask_reg, r0uw);
brw_pop_insn_state(p);
}
brw_push_insn_state(p);
for (channel = 0; channel < 4; channel++) {
- if (c->dispatch_width == 16 && (BRW_IS_G4X(brw) || BRW_IS_IGDNG(brw))) {
+ if (c->dispatch_width == 16 && brw->has_compr4) {
/* By setting the high bit of the MRF register number, we indicate
* that we want COMPR4 mode - instead of doing the usual destination
* + 1 for the second half we get destination + 4.
break;
default:
- _mesa_printf("Unsupported opcode %i (%s) in fragment shader\n",
- inst->opcode, inst->opcode < MAX_OPCODE ?
- _mesa_opcode_string(inst->opcode) :
- "unknown");
+ printf("Unsupported opcode %i (%s) in fragment shader\n",
+ inst->opcode, inst->opcode < MAX_OPCODE ?
+ _mesa_opcode_string(inst->opcode) :
+ "unknown");
}
for (i = 0; i < 4; i++)
if (INTEL_DEBUG & DEBUG_WM) {
int i;
- _mesa_printf("wm-native:\n");
+ printf("wm-native:\n");
for (i = 0; i < p->nr_insn; i++)
brw_disasm(stderr, &p->store[i]);
- _mesa_printf("\n");
+ printf("\n");
}
}