return reg;
}
+
/* Payload R0:
*
* R0.0 -- pixel mask, one bit for each of 4 pixels in 4 tiles,
* R1.8 -- ?
*/
-
-static void emit_pixel_xy(struct brw_compile *p,
- const struct brw_reg *dst,
- GLuint mask)
+void emit_pixel_xy(struct brw_wm_compile *c,
+ const struct brw_reg *dst,
+ GLuint mask)
{
+ struct brw_compile *p = &c->func;
struct brw_reg r1 = brw_vec1_grf(1, 0);
struct brw_reg r1_uw = retype(r1, BRW_REGISTER_TYPE_UW);
+ struct brw_reg dst0_uw, dst1_uw;
+ brw_push_insn_state(p);
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
+ if (c->dispatch_width == 16) {
+ dst0_uw = vec16(retype(dst[0], BRW_REGISTER_TYPE_UW));
+ dst1_uw = vec16(retype(dst[1], BRW_REGISTER_TYPE_UW));
+ } else {
+ dst0_uw = vec8(retype(dst[0], BRW_REGISTER_TYPE_UW));
+ dst1_uw = vec8(retype(dst[1], BRW_REGISTER_TYPE_UW));
+ }
+
/* Calculate pixel centers by adding 1 or 0 to each of the
* micro-tile coordinates passed in r1.
*/
if (mask & WRITEMASK_X) {
brw_ADD(p,
- vec16(retype(dst[0], BRW_REGISTER_TYPE_UW)),
+ dst0_uw,
stride(suboffset(r1_uw, 4), 2, 4, 0),
brw_imm_v(0x10101010));
}
if (mask & WRITEMASK_Y) {
brw_ADD(p,
- vec16(retype(dst[1], BRW_REGISTER_TYPE_UW)),
+ dst1_uw,
stride(suboffset(r1_uw,5), 2, 4, 0),
brw_imm_v(0x11001100));
}
-
- brw_set_compression_control(p, BRW_COMPRESSION_COMPRESSED);
+ brw_pop_insn_state(p);
}
-
-static void emit_delta_xy(struct brw_compile *p,
- const struct brw_reg *dst,
- GLuint mask,
- const struct brw_reg *arg0)
+void emit_delta_xy(struct brw_compile *p,
+ const struct brw_reg *dst,
+ GLuint mask,
+ const struct brw_reg *arg0)
{
struct brw_reg r1 = brw_vec1_grf(1, 0);
}
}
-static void emit_wpos_xy(struct brw_wm_compile *c,
- const struct brw_reg *dst,
- GLuint mask,
- const struct brw_reg *arg0)
+void emit_wpos_xy(struct brw_wm_compile *c,
+ const struct brw_reg *dst,
+ GLuint mask,
+ const struct brw_reg *arg0)
{
struct brw_compile *p = &c->func;
}
-static void emit_pixel_w( struct brw_compile *p,
- const struct brw_reg *dst,
- GLuint mask,
- const struct brw_reg *arg0,
- const struct brw_reg *deltas)
+void emit_pixel_w(struct brw_wm_compile *c,
+ const struct brw_reg *dst,
+ GLuint mask,
+ const struct brw_reg *arg0,
+ const struct brw_reg *deltas)
{
+ struct brw_compile *p = &c->func;
+
/* Don't need this if all you are doing is interpolating color, for
* instance.
*/
brw_MAC(p, brw_message_reg(2), suboffset(interp3, 1), deltas[1]);
/* Calc w */
- brw_math_16( p, dst[3],
- BRW_MATH_FUNCTION_INV,
- BRW_MATH_SATURATE_NONE,
- 2, brw_null_reg(),
- BRW_MATH_PRECISION_FULL);
+ if (c->dispatch_width == 16) {
+ brw_math_16(p, dst[3],
+ BRW_MATH_FUNCTION_INV,
+ BRW_MATH_SATURATE_NONE,
+ 2, brw_null_reg(),
+ BRW_MATH_PRECISION_FULL);
+ } else {
+ brw_math(p, dst[3],
+ BRW_MATH_FUNCTION_INV,
+ BRW_MATH_SATURATE_NONE,
+ 2, brw_null_reg(),
+ BRW_MATH_DATA_VECTOR,
+ BRW_MATH_PRECISION_FULL);
+ }
}
}
-
-static void emit_linterp( struct brw_compile *p,
- const struct brw_reg *dst,
- GLuint mask,
- const struct brw_reg *arg0,
- const struct brw_reg *deltas )
+void emit_linterp(struct brw_compile *p,
+ const struct brw_reg *dst,
+ GLuint mask,
+ const struct brw_reg *arg0,
+ const struct brw_reg *deltas)
{
struct brw_reg interp[4];
GLuint nr = arg0[0].nr;
}
-static void emit_pinterp( struct brw_compile *p,
- const struct brw_reg *dst,
- GLuint mask,
- const struct brw_reg *arg0,
- const struct brw_reg *deltas,
- const struct brw_reg *w)
+void emit_pinterp(struct brw_compile *p,
+ const struct brw_reg *dst,
+ GLuint mask,
+ const struct brw_reg *arg0,
+ const struct brw_reg *deltas,
+ const struct brw_reg *w)
{
struct brw_reg interp[4];
GLuint nr = arg0[0].nr;
}
-static void emit_cinterp( struct brw_compile *p,
- const struct brw_reg *dst,
- GLuint mask,
- const struct brw_reg *arg0 )
+void emit_cinterp(struct brw_compile *p,
+ const struct brw_reg *dst,
+ GLuint mask,
+ const struct brw_reg *arg0)
{
struct brw_reg interp[4];
GLuint nr = arg0[0].nr;
}
/* Sets the destination channels to 1.0 or 0.0 according to glFrontFacing. */
-static void emit_frontfacing( struct brw_compile *p,
- const struct brw_reg *dst,
- GLuint mask )
+void emit_frontfacing(struct brw_compile *p,
+ const struct brw_reg *dst,
+ GLuint mask)
{
struct brw_reg r1_6ud = retype(brw_vec1_grf(1, 6), BRW_REGISTER_TYPE_UD);
GLuint i;
}
}
-static void emit_max( struct brw_compile *p,
- const struct brw_reg *dst,
- GLuint mask,
- const struct brw_reg *arg0,
- const struct brw_reg *arg1 )
+void emit_max(struct brw_compile *p,
+ const struct brw_reg *dst,
+ GLuint mask,
+ const struct brw_reg *arg0,
+ const struct brw_reg *arg1)
{
GLuint i;
}
}
-static void emit_min( struct brw_compile *p,
- const struct brw_reg *dst,
- GLuint mask,
- const struct brw_reg *arg0,
- const struct brw_reg *arg1 )
+void emit_min(struct brw_compile *p,
+ const struct brw_reg *dst,
+ GLuint mask,
+ const struct brw_reg *arg0,
+ const struct brw_reg *arg1)
{
GLuint i;
}
brw_pop_insn_state(p);
}
-
-static void emit_tex( struct brw_wm_compile *c,
- const struct brw_wm_instruction *inst,
- struct brw_reg *dst,
- GLuint dst_flags,
- struct brw_reg *arg )
+void emit_tex(struct brw_wm_compile *c,
+ struct brw_reg *dst,
+ GLuint dst_flags,
+ struct brw_reg *arg,
+ struct brw_reg depth_payload,
+ GLuint tex_idx,
+ GLuint sampler,
+ GLboolean shadow)
{
struct brw_compile *p = &c->func;
- GLuint msgLength, responseLength;
- GLuint i, nr;
+ struct brw_reg dst_retyped;
+ GLuint cur_mrf = 2, response_length;
+ GLuint i, nr_texcoords;
GLuint emit;
GLuint msg_type;
+ GLuint mrf_per_channel;
+ GLuint simd_mode;
+
+ if (c->dispatch_width == 16) {
+ mrf_per_channel = 2;
+ response_length = 8;
+ dst_retyped = retype(vec16(dst[0]), BRW_REGISTER_TYPE_UW);
+ simd_mode = BRW_SAMPLER_SIMD_MODE_SIMD16;
+ } else {
+ mrf_per_channel = 1;
+ response_length = 4;
+ dst_retyped = retype(vec8(dst[0]), BRW_REGISTER_TYPE_UW);
+ simd_mode = BRW_SAMPLER_SIMD_MODE_SIMD8;
+ }
/* How many input regs are there?
*/
- switch (inst->tex_idx) {
+ switch (tex_idx) {
case TEXTURE_1D_INDEX:
emit = WRITEMASK_X;
- nr = 1;
+ nr_texcoords = 1;
break;
case TEXTURE_2D_INDEX:
case TEXTURE_RECT_INDEX:
emit = WRITEMASK_XY;
- nr = 2;
+ nr_texcoords = 2;
break;
case TEXTURE_3D_INDEX:
case TEXTURE_CUBE_INDEX:
emit = WRITEMASK_XYZ;
- nr = 3;
+ nr_texcoords = 3;
break;
default:
/* unexpected target */
abort();
}
- if (inst->tex_shadow) {
- nr = 4;
- emit |= WRITEMASK_W;
- }
+ /* Pre-Ironlake, the 8-wide sampler always took u,v,r. */
+ if (!BRW_IS_IGDNG(p->brw) && c->dispatch_width == 8)
+ nr_texcoords = 3;
- msgLength = 1;
+ /* For shadow comparisons, we have to supply u,v,r. */
+ if (shadow)
+ nr_texcoords = 3;
- for (i = 0; i < nr; i++) {
- static const GLuint swz[4] = {0,1,2,2};
- if (emit & (1<<i))
- brw_MOV(p, brw_message_reg(msgLength+1), arg[swz[i]]);
+ /* Emit the texcoords. */
+ for (i = 0; i < nr_texcoords; i++) {
+ if (emit & (1<<i))
+ brw_MOV(p, brw_message_reg(cur_mrf), arg[i]);
else
- brw_MOV(p, brw_message_reg(msgLength+1), brw_imm_f(0));
- msgLength += 2;
+ brw_MOV(p, brw_message_reg(cur_mrf), brw_imm_f(0));
+ cur_mrf += mrf_per_channel;
}
- responseLength = 8; /* always */
+ /* Fill in the shadow comparison reference value. */
+ if (shadow) {
+ if (BRW_IS_IGDNG(p->brw)) {
+ /* 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;
+ } else if (c->dispatch_width == 8) {
+ /* Fill in the LOD bias value. */
+ brw_MOV(p, brw_message_reg(cur_mrf), brw_imm_f(0));
+ cur_mrf += mrf_per_channel;
+ }
+ brw_MOV(p, brw_message_reg(cur_mrf), arg[2]);
+ cur_mrf += mrf_per_channel;
+ }
if (BRW_IS_IGDNG(p->brw)) {
- if (inst->tex_shadow)
- msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_COMPARE_IGDNG;
- else
- msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_IGDNG;
+ if (shadow)
+ msg_type = BRW_SAMPLER_MESSAGE_SAMPLE_COMPARE_IGDNG;
+ else
+ msg_type = BRW_SAMPLER_MESSAGE_SAMPLE_IGDNG;
} else {
- if (inst->tex_shadow)
- msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_COMPARE;
- else
- msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE;
+ /* Note that G45 and older determines shadow compare and dispatch width
+ * from message length for most messages.
+ */
+ if (c->dispatch_width == 16 && shadow)
+ msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_COMPARE;
+ else
+ msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE;
}
- brw_SAMPLE(p,
- retype(vec16(dst[0]), BRW_REGISTER_TYPE_UW),
+ brw_SAMPLE(p,
+ dst_retyped,
1,
- retype(c->payload.depth[0].hw_reg, BRW_REGISTER_TYPE_UW),
- SURF_INDEX_TEXTURE(inst->tex_unit),
- inst->tex_unit, /* sampler */
- inst->writemask,
- msg_type,
- responseLength,
- msgLength,
- 0,
+ retype(depth_payload, BRW_REGISTER_TYPE_UW),
+ SURF_INDEX_TEXTURE(sampler),
+ sampler,
+ dst_flags & WRITEMASK_XYZW,
+ msg_type,
+ response_length,
+ cur_mrf - 1,
+ 0,
1,
- BRW_SAMPLER_SIMD_MODE_SIMD16);
+ simd_mode);
}
-static void emit_txb( struct brw_wm_compile *c,
- const struct brw_wm_instruction *inst,
- struct brw_reg *dst,
- GLuint dst_flags,
- struct brw_reg *arg )
+void emit_txb(struct brw_wm_compile *c,
+ struct brw_reg *dst,
+ GLuint dst_flags,
+ struct brw_reg *arg,
+ struct brw_reg depth_payload,
+ GLuint tex_idx,
+ GLuint sampler)
{
struct brw_compile *p = &c->func;
GLuint msgLength;
GLuint msg_type;
- /* Shadow ignored for txb.
+ GLuint mrf_per_channel;
+ GLuint response_length;
+ struct brw_reg dst_retyped;
+
+ /* The G45 and older chipsets don't support 8-wide dispatch for LOD biased
+ * samples, so we'll use the 16-wide instruction, leave the second halves
+ * undefined, and trust the execution mask to keep the undefined pixels
+ * from mattering.
*/
- switch (inst->tex_idx) {
+ if (c->dispatch_width == 16 || !BRW_IS_IGDNG(p->brw)) {
+ if (BRW_IS_IGDNG(p->brw))
+ msg_type = BRW_SAMPLER_MESSAGE_SAMPLE_BIAS_IGDNG;
+ else
+ msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS;
+ mrf_per_channel = 2;
+ dst_retyped = retype(vec16(dst[0]), BRW_REGISTER_TYPE_UW);
+ response_length = 8;
+ } else {
+ msg_type = BRW_SAMPLER_MESSAGE_SAMPLE_BIAS_IGDNG;
+ mrf_per_channel = 1;
+ dst_retyped = retype(vec8(dst[0]), BRW_REGISTER_TYPE_UW);
+ response_length = 4;
+ }
+
+ /* Shadow ignored for txb. */
+ switch (tex_idx) {
case TEXTURE_1D_INDEX:
- brw_MOV(p, brw_message_reg(2), arg[0]);
- brw_MOV(p, brw_message_reg(4), brw_imm_f(0));
- brw_MOV(p, brw_message_reg(6), brw_imm_f(0));
+ brw_MOV(p, brw_message_reg(2 + 0 * mrf_per_channel), arg[0]);
+ brw_MOV(p, brw_message_reg(2 + 1 * mrf_per_channel), brw_imm_f(0));
+ brw_MOV(p, brw_message_reg(2 + 2 * mrf_per_channel), brw_imm_f(0));
break;
case TEXTURE_2D_INDEX:
case TEXTURE_RECT_INDEX:
- brw_MOV(p, brw_message_reg(2), arg[0]);
- brw_MOV(p, brw_message_reg(4), arg[1]);
- brw_MOV(p, brw_message_reg(6), brw_imm_f(0));
+ brw_MOV(p, brw_message_reg(2 + 0 * mrf_per_channel), arg[0]);
+ brw_MOV(p, brw_message_reg(2 + 1 * mrf_per_channel), arg[1]);
+ brw_MOV(p, brw_message_reg(2 + 2 * mrf_per_channel), brw_imm_f(0));
break;
case TEXTURE_3D_INDEX:
case TEXTURE_CUBE_INDEX:
- brw_MOV(p, brw_message_reg(2), arg[0]);
- brw_MOV(p, brw_message_reg(4), arg[1]);
- brw_MOV(p, brw_message_reg(6), arg[2]);
+ brw_MOV(p, brw_message_reg(2 + 0 * mrf_per_channel), arg[0]);
+ brw_MOV(p, brw_message_reg(2 + 1 * mrf_per_channel), arg[1]);
+ brw_MOV(p, brw_message_reg(2 + 2 * mrf_per_channel), arg[2]);
break;
default:
/* unexpected target */
abort();
}
- brw_MOV(p, brw_message_reg(8), arg[3]);
- msgLength = 9;
-
- if (BRW_IS_IGDNG(p->brw))
- msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS_IGDNG;
- else
- msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS;
+ brw_MOV(p, brw_message_reg(2 + 3 * mrf_per_channel), arg[3]);
+ msgLength = 2 + 4 * mrf_per_channel - 1;
brw_SAMPLE(p,
- retype(vec16(dst[0]), BRW_REGISTER_TYPE_UW),
+ dst_retyped,
1,
- retype(c->payload.depth[0].hw_reg, BRW_REGISTER_TYPE_UW),
- SURF_INDEX_TEXTURE(inst->tex_unit),
- inst->tex_unit, /* sampler */
- inst->writemask,
+ retype(depth_payload, BRW_REGISTER_TYPE_UW),
+ SURF_INDEX_TEXTURE(sampler),
+ sampler,
+ dst_flags & WRITEMASK_XYZW,
msg_type,
- 8, /* responseLength */
+ response_length,
msgLength,
0,
1,
GLuint eot )
{
struct brw_compile *p = &c->func;
-
+ struct brw_reg dst;
+
+ if (c->dispatch_width == 16)
+ dst = retype(vec16(brw_null_reg()), BRW_REGISTER_TYPE_UW);
+ else
+ dst = retype(vec8(brw_null_reg()), BRW_REGISTER_TYPE_UW);
+
/* Pass through control information:
*/
/* mov (8) m1.0<1>:ud r1.0<8;8,1>:ud { Align1 NoMask } */
/* Send framebuffer write message: */
/* send (16) null.0<1>:uw m0 r0.0<8;8,1>:uw 0x85a04000:ud { Align1 EOT } */
brw_fb_WRITE(p,
- retype(vec16(brw_null_reg()), BRW_REGISTER_TYPE_UW),
+ dst,
base_reg,
retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UW),
target,
* \param arg1 the pass-through depth value
* \param arg2 the shader-computed depth value
*/
-static void emit_fb_write( struct brw_wm_compile *c,
- struct brw_reg *arg0,
- struct brw_reg *arg1,
- struct brw_reg *arg2,
- GLuint target,
- GLuint eot)
+void emit_fb_write(struct brw_wm_compile *c,
+ struct brw_reg *arg0,
+ struct brw_reg *arg1,
+ struct brw_reg *arg2,
+ GLuint target,
+ GLuint eot)
{
struct brw_compile *p = &c->func;
+ struct brw_context *brw = p->brw;
GLuint nr = 2;
GLuint channel;
/* I don't really understand how this achieves the color interleave
* (ie RGBARGBA) in the result: [Do the saturation here]
*/
- {
- brw_push_insn_state(p);
-
- for (channel = 0; channel < 4; channel++) {
+ brw_push_insn_state(p);
+
+ for (channel = 0; channel < 4; channel++) {
+ if (c->dispatch_width == 16 && (BRW_IS_G4X(brw) || BRW_IS_IGDNG(brw))) {
+ /* 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.
+ */
+ brw_MOV(p,
+ brw_message_reg(nr + channel + (1 << 7)),
+ arg0[channel]);
+ } else {
/* mov (8) m2.0<1>:ud r28.0<8;8,1>:ud { Align1 } */
/* mov (8) m6.0<1>:ud r29.0<8;8,1>:ud { Align1 SecHalf } */
-
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_MOV(p,
brw_message_reg(nr + channel),
arg0[channel]);
-
- brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF);
- brw_MOV(p,
- brw_message_reg(nr + channel + 4),
- sechalf(arg0[channel]));
- }
- /* skip over the regs populated above:
- */
- nr += 8;
-
- brw_pop_insn_state(p);
+ if (c->dispatch_width == 16) {
+ brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF);
+ brw_MOV(p,
+ brw_message_reg(nr + channel + 4),
+ sechalf(arg0[channel]));
+ }
+ }
}
+ /* skip over the regs populated above:
+ */
+ nr += 8;
+ brw_pop_insn_state(p);
if (c->key.source_depth_to_render_target)
{
get_element_ud(brw_vec8_grf(1,0), 6),
brw_imm_ud(1<<26));
- jmp = brw_JMPI(p, ip, ip, brw_imm_d(0));
+ jmp = brw_JMPI(p, ip, ip, brw_imm_w(0));
{
emit_aa(c, arg1, 2);
fire_fb_write(c, 0, nr, target, eot);
}
}
-
/**
* Move a GPR to scratch memory.
*/
/* Generated instructions for calculating triangle interpolants:
*/
case WM_PIXELXY:
- emit_pixel_xy(p, dst, dst_flags);
+ emit_pixel_xy(c, dst, dst_flags);
break;
case WM_DELTAXY:
break;
case WM_PIXELW:
- emit_pixel_w(p, dst, dst_flags, args[0], args[1]);
+ emit_pixel_w(c, dst, dst_flags, args[0], args[1]);
break;
case WM_LINTERP:
/* Texturing operations:
*/
case OPCODE_TEX:
- emit_tex(c, inst, dst, dst_flags, args[0]);
+ emit_tex(c, dst, dst_flags, args[0], c->payload.depth[0].hw_reg,
+ inst->tex_idx, inst->tex_unit,
+ inst->tex_shadow);
break;
case OPCODE_TXB:
- emit_txb(c, inst, dst, dst_flags, args[0]);
+ emit_txb(c, dst, dst_flags, args[0], c->payload.depth[0].hw_reg,
+ inst->tex_idx, inst->tex_unit);
break;
case OPCODE_KIL: