void brw_emit_anyprim_setup( struct brw_sf_compile *c )
{
struct brw_compile *p = &c->func;
struct brw_context *brw = p->brw;
struct brw_reg payload_prim = brw_uw1_reg(BRW_GENERAL_REGISTER_FILE, 1, 0);
struct brw_reg payload_attr = get_element_ud(brw_vec1_reg(BRW_GENERAL_REGISTER_FILE, 1, 0), 0);
struct brw_reg primmask;
int jmp;
struct brw_reg v1_null_ud = vec1(retype(brw_null_reg(), BRW_REGISTER_TYPE_UD));
c->nr_verts = 3;
alloc_regs(c);
primmask = retype(get_element(c->tmp, 0), BRW_REGISTER_TYPE_UD);
brw_MOV(p, primmask, brw_imm_ud(1));
brw_SHL(p, primmask, primmask, payload_prim);
brw_AND(p, v1_null_ud, primmask, brw_imm_ud((1<<_3DPRIM_TRILIST) |
(1<<_3DPRIM_TRISTRIP) |
(1<<_3DPRIM_TRIFAN) |
(1<<_3DPRIM_TRISTRIP_REVERSE) |
(1<<_3DPRIM_POLYGON) |
(1<<_3DPRIM_RECTLIST) |
(1<<_3DPRIM_TRIFAN_NOSTIPPLE)));
brw_inst_set_cond_modifier(brw, brw_last_inst, BRW_CONDITIONAL_Z);
jmp = brw_JMPI(p, brw_imm_d(0), BRW_PREDICATE_NORMAL) - p->store;
brw_emit_tri_setup(c, false);
brw_land_fwd_jump(p, jmp);
brw_AND(p, v1_null_ud, primmask, brw_imm_ud((1<<_3DPRIM_LINELIST) |
(1<<_3DPRIM_LINESTRIP) |
(1<<_3DPRIM_LINELOOP) |
(1<<_3DPRIM_LINESTRIP_CONT) |
(1<<_3DPRIM_LINESTRIP_BF) |
(1<<_3DPRIM_LINESTRIP_CONT_BF)));
brw_inst_set_cond_modifier(brw, brw_last_inst, BRW_CONDITIONAL_Z);
jmp = brw_JMPI(p, brw_imm_d(0), BRW_PREDICATE_NORMAL) - p->store;
brw_emit_line_setup(c, false);
brw_land_fwd_jump(p, jmp);
brw_AND(p, v1_null_ud, payload_attr, brw_imm_ud(1<<BRW_SPRITE_POINT_ENABLE));
brw_inst_set_cond_modifier(brw, brw_last_inst, BRW_CONDITIONAL_Z);
jmp = brw_JMPI(p, brw_imm_d(0), BRW_PREDICATE_NORMAL) - p->store;
brw_emit_point_sprite_setup(c, false);
brw_land_fwd_jump(p, jmp);
brw_emit_point_setup( c, false );
}
/* Post-fragment-program processing. Send the results to the
* framebuffer.
* \param arg0 the fragment color
* \param arg1 the pass-through depth value
* \param arg2 the shader-computed depth value
*/
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;
struct intel_context *intel = &brw->intel;
GLuint nr = 2;
GLuint channel;
/* Reserve a space for AA - may not be needed:
*/
if (c->aa_dest_stencil_reg)
nr += 1;
/* 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);
if (c->key.clamp_fragment_color)
brw_set_saturate(p, 1);
for (channel = 0; channel < 4; channel++) {
if (intel->gen >= 6) {
/* gen6 SIMD16 single source DP write looks like:
* m + 0: r0
* m + 1: r1
* m + 2: g0
* m + 3: g1
* m + 4: b0
* m + 5: b1
* m + 6: a0
* m + 7: a1
*/
if (c->dispatch_width == 16) {
brw_MOV(p, brw_message_reg(nr + channel * 2), arg0[channel]);
} else {
brw_MOV(p, brw_message_reg(nr + channel), arg0[channel]);
}
} else if (c->dispatch_width == 16 && brw->has_compr4) {
/* pre-gen6 SIMD16 single source DP write looks like:
* m + 0: r0
* m + 1: g0
* m + 2: b0
* m + 3: a0
* m + 4: r1
* m + 5: g1
* m + 6: b1
* m + 7: a1
*
* 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 + BRW_MRF_COMPR4),
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]);
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]));
}
}
}
brw_set_saturate(p, 0);
/* skip over the regs populated above:
*/
if (c->dispatch_width == 16)
nr += 8;
else
nr += 4;
brw_pop_insn_state(p);
if (c->source_depth_to_render_target)
{
if (c->computes_depth)
brw_MOV(p, brw_message_reg(nr), arg2[2]);
else
brw_MOV(p, brw_message_reg(nr), arg1[1]); /* ? */
nr += 2;
}
if (c->dest_depth_reg)
{
GLuint comp = c->dest_depth_reg / 2;
GLuint off = c->dest_depth_reg % 2;
if (off != 0) {
brw_push_insn_state(p);
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_MOV(p, brw_message_reg(nr), offset(arg1[comp],1));
/* 2nd half? */
brw_MOV(p, brw_message_reg(nr+1), arg1[comp+1]);
brw_pop_insn_state(p);
}
else {
brw_MOV(p, brw_message_reg(nr), arg1[comp]);
}
nr += 2;
}
if (intel->gen >= 6) {
/* Load the message header. There's no implied move from src0
* to the base mrf on gen6.
*/
brw_push_insn_state(p);
brw_set_mask_control(p, BRW_MASK_DISABLE);
brw_MOV(p, retype(brw_message_reg(0), BRW_REGISTER_TYPE_UD),
retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UD));
brw_pop_insn_state(p);
if (target != 0) {
brw_MOV(p, retype(brw_vec1_reg(BRW_MESSAGE_REGISTER_FILE,
0,
2), BRW_REGISTER_TYPE_UD),
brw_imm_ud(target));
}
}
if (!c->runtime_check_aads_emit) {
if (c->aa_dest_stencil_reg)
emit_aa(c, arg1, 2);
fire_fb_write(c, 0, nr, target, eot);
}
else {
struct brw_reg v1_null_ud = vec1(retype(brw_null_reg(), BRW_REGISTER_TYPE_UD));
struct brw_reg ip = brw_ip_reg();
struct brw_instruction *jmp;
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_set_conditionalmod(p, BRW_CONDITIONAL_Z);
brw_AND(p,
v1_null_ud,
get_element_ud(brw_vec8_grf(1,0), 6),
brw_imm_ud(1<<26));
jmp = brw_JMPI(p, ip, ip, brw_imm_w(0));
{
emit_aa(c, arg1, 2);
fire_fb_write(c, 0, nr, target, eot);
/* note - thread killed in subroutine */
}
brw_land_fwd_jump(p, jmp);
/* ELSE: Shuffle up one register to fill in the hole left for AA:
*/
fire_fb_write(c, 1, nr-1, target, eot);
}
}
void brw_emit_anyprim_setup( struct brw_sf_compile *c )
{
struct brw_compile *p = &c->func;
struct brw_reg ip = brw_ip_reg();
struct brw_reg payload_prim = brw_uw1_reg(BRW_GENERAL_REGISTER_FILE, 1, 0);
struct brw_reg payload_attr = get_element_ud(brw_vec1_reg(BRW_GENERAL_REGISTER_FILE, 1, 0), 0);
struct brw_reg primmask;
struct brw_instruction *jmp;
struct brw_reg v1_null_ud = vec1(retype(brw_null_reg(), BRW_REGISTER_TYPE_UD));
GLuint saveflag;
c->nr_verts = 3;
alloc_regs(c);
primmask = retype(get_element(c->tmp, 0), BRW_REGISTER_TYPE_UD);
brw_MOV(p, primmask, brw_imm_ud(1));
brw_SHL(p, primmask, primmask, payload_prim);
brw_set_conditionalmod(p, BRW_CONDITIONAL_Z);
brw_AND(p, v1_null_ud, primmask, brw_imm_ud((1<<_3DPRIM_TRILIST) |
(1<<_3DPRIM_TRISTRIP) |
(1<<_3DPRIM_TRIFAN) |
(1<<_3DPRIM_TRISTRIP_REVERSE) |
(1<<_3DPRIM_POLYGON) |
(1<<_3DPRIM_RECTLIST) |
(1<<_3DPRIM_TRIFAN_NOSTIPPLE)));
jmp = brw_JMPI(p, ip, ip, brw_imm_d(0));
{
saveflag = p->flag_value;
brw_push_insn_state(p);
brw_emit_tri_setup( c, GL_FALSE );
brw_pop_insn_state(p);
p->flag_value = saveflag;
/* note - thread killed in subroutine, so must
* restore the flag which is changed when building
* the subroutine. fix #13240
*/
}
brw_land_fwd_jump(p, jmp);
brw_set_conditionalmod(p, BRW_CONDITIONAL_Z);
brw_AND(p, v1_null_ud, primmask, brw_imm_ud((1<<_3DPRIM_LINELIST) |
(1<<_3DPRIM_LINESTRIP) |
(1<<_3DPRIM_LINELOOP) |
(1<<_3DPRIM_LINESTRIP_CONT) |
(1<<_3DPRIM_LINESTRIP_BF) |
(1<<_3DPRIM_LINESTRIP_CONT_BF)));
jmp = brw_JMPI(p, ip, ip, brw_imm_d(0));
{
saveflag = p->flag_value;
brw_push_insn_state(p);
brw_emit_line_setup( c, GL_FALSE );
brw_pop_insn_state(p);
p->flag_value = saveflag;
/* note - thread killed in subroutine */
}
brw_land_fwd_jump(p, jmp);
brw_set_conditionalmod(p, BRW_CONDITIONAL_Z);
brw_AND(p, v1_null_ud, payload_attr, brw_imm_ud(1<<BRW_SPRITE_POINT_ENABLE));
jmp = brw_JMPI(p, ip, ip, brw_imm_d(0));
{
saveflag = p->flag_value;
brw_push_insn_state(p);
brw_emit_point_sprite_setup( c, GL_FALSE );
brw_pop_insn_state(p);
p->flag_value = saveflag;
}
brw_land_fwd_jump(p, jmp);
brw_emit_point_setup( c, GL_FALSE );
}
void
gen8_vec4_generator::generate_tex(vec4_instruction *ir, struct brw_reg dst)
{
int msg_type = 0;
switch (ir->opcode) {
case SHADER_OPCODE_TEX:
case SHADER_OPCODE_TXL:
if (ir->shadow_compare) {
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE_LOD_COMPARE;
} else {
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE_LOD;
}
break;
case SHADER_OPCODE_TXD:
if (ir->shadow_compare) {
msg_type = HSW_SAMPLER_MESSAGE_SAMPLE_DERIV_COMPARE;
} else {
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE_DERIVS;
}
break;
case SHADER_OPCODE_TXF:
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE_LD;
break;
case SHADER_OPCODE_TXF_CMS:
msg_type = GEN7_SAMPLER_MESSAGE_SAMPLE_LD2DMS;
break;
case SHADER_OPCODE_TXF_MCS:
msg_type = GEN7_SAMPLER_MESSAGE_SAMPLE_LD_MCS;
break;
case SHADER_OPCODE_TXS:
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE_RESINFO;
break;
case SHADER_OPCODE_TG4:
if (ir->shadow_compare) {
msg_type = GEN7_SAMPLER_MESSAGE_SAMPLE_GATHER4_C;
} else {
msg_type = GEN7_SAMPLER_MESSAGE_SAMPLE_GATHER4;
}
break;
case SHADER_OPCODE_TG4_OFFSET:
if (ir->shadow_compare) {
msg_type = GEN7_SAMPLER_MESSAGE_SAMPLE_GATHER4_PO_C;
} else {
msg_type = GEN7_SAMPLER_MESSAGE_SAMPLE_GATHER4_PO;
}
break;
default:
assert(!"should not get here: invalid VS texture opcode");
break;
}
if (ir->header_present) {
MOV_RAW(retype(brw_message_reg(ir->base_mrf), BRW_REGISTER_TYPE_UD),
retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UD));
default_state.access_mode = BRW_ALIGN_1;
if (ir->texture_offset) {
/* Set the offset bits in DWord 2. */
MOV_RAW(retype(brw_vec1_reg(MRF, ir->base_mrf, 2),
BRW_REGISTER_TYPE_UD),
brw_imm_ud(ir->texture_offset));
}
if (ir->sampler >= 16) {
/* The "Sampler Index" field can only store values between 0 and 15.
* However, we can add an offset to the "Sampler State Pointer"
* field, effectively selecting a different set of 16 samplers.
*
* The "Sampler State Pointer" needs to be aligned to a 32-byte
* offset, and each sampler state is only 16-bytes, so we can't
* exclusively use the offset - we have to use both.
*/
gen8_instruction *add =
ADD(get_element_ud(brw_message_reg(ir->base_mrf), 3),
get_element_ud(brw_vec8_grf(0, 0), 3),
brw_imm_ud(16 * (ir->sampler / 16) *
sizeof(gen7_sampler_state)));
gen8_set_mask_control(add, BRW_MASK_DISABLE);
}
default_state.access_mode = BRW_ALIGN_16;
}
uint32_t surf_index =
prog_data->base.binding_table.texture_start + ir->sampler;
gen8_instruction *inst = next_inst(BRW_OPCODE_SEND);
gen8_set_dst(brw, inst, dst);
gen8_set_src0(brw, inst, brw_message_reg(ir->base_mrf));
gen8_set_sampler_message(brw, inst,
surf_index,
ir->sampler % 16,
msg_type,
1,
ir->mlen,
ir->header_present,
BRW_SAMPLER_SIMD_MODE_SIMD4X2);
mark_surface_used(surf_index);
}
void
vec4_generator::generate_tex(vec4_instruction *inst,
struct brw_reg dst,
struct brw_reg src)
{
int msg_type = -1;
if (intel->gen >= 5) {
switch (inst->opcode) {
case SHADER_OPCODE_TEX:
case SHADER_OPCODE_TXL:
if (inst->shadow_compare) {
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE_LOD_COMPARE;
} else {
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE_LOD;
}
break;
case SHADER_OPCODE_TXD:
/* There is no sample_d_c message; comparisons are done manually. */
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE_DERIVS;
break;
case SHADER_OPCODE_TXF:
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE_LD;
break;
case SHADER_OPCODE_TXS:
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE_RESINFO;
break;
default:
assert(!"should not get here: invalid VS texture opcode");
break;
}
} else {
switch (inst->opcode) {
case SHADER_OPCODE_TEX:
case SHADER_OPCODE_TXL:
if (inst->shadow_compare) {
msg_type = BRW_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_LOD_COMPARE;
assert(inst->mlen == 3);
} else {
msg_type = BRW_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_LOD;
assert(inst->mlen == 2);
}
break;
case SHADER_OPCODE_TXD:
/* There is no sample_d_c message; comparisons are done manually. */
msg_type = BRW_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_GRADIENTS;
assert(inst->mlen == 4);
break;
case SHADER_OPCODE_TXF:
msg_type = BRW_SAMPLER_MESSAGE_SIMD4X2_LD;
assert(inst->mlen == 2);
break;
case SHADER_OPCODE_TXS:
msg_type = BRW_SAMPLER_MESSAGE_SIMD4X2_RESINFO;
assert(inst->mlen == 2);
break;
default:
assert(!"should not get here: invalid VS texture opcode");
break;
}
}
assert(msg_type != -1);
/* Load the message header if present. If there's a texture offset, we need
* to set it up explicitly and load the offset bitfield. Otherwise, we can
* use an implied move from g0 to the first message register.
*/
if (inst->texture_offset) {
/* Explicitly set up the message header by copying g0 to the MRF. */
brw_MOV(p, retype(brw_message_reg(inst->base_mrf), BRW_REGISTER_TYPE_UD),
retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UD));
/* Then set the offset bits in DWord 2. */
brw_set_access_mode(p, BRW_ALIGN_1);
brw_MOV(p,
retype(brw_vec1_reg(BRW_MESSAGE_REGISTER_FILE, inst->base_mrf, 2),
BRW_REGISTER_TYPE_UD),
brw_imm_uw(inst->texture_offset));
brw_set_access_mode(p, BRW_ALIGN_16);
} else if (inst->header_present) {
/* Set up an implied move from g0 to the MRF. */
src = brw_vec8_grf(0, 0);
}
uint32_t return_format;
switch (dst.type) {
case BRW_REGISTER_TYPE_D:
return_format = BRW_SAMPLER_RETURN_FORMAT_SINT32;
break;
case BRW_REGISTER_TYPE_UD:
return_format = BRW_SAMPLER_RETURN_FORMAT_UINT32;
break;
default:
return_format = BRW_SAMPLER_RETURN_FORMAT_FLOAT32;
break;
}
brw_SAMPLE(p,
dst,
inst->base_mrf,
src,
SURF_INDEX_VS_TEXTURE(inst->sampler),
inst->sampler,
WRITEMASK_XYZW,
msg_type,
1, /* response length */
inst->mlen,
inst->header_present,
BRW_SAMPLER_SIMD_MODE_SIMD4X2,
return_format);
}
