void
gen8_vec4_generator::generate_gs_set_write_offset(struct brw_reg dst,
struct brw_reg src0,
struct brw_reg src1)
{
/* From p22 of volume 4 part 2 of the Ivy Bridge PRM (2.4.3.1 Message
* Header: M0.3):
*
* Slot 0 Offset. This field, after adding to the Global Offset field
* in the message descriptor, specifies the offset (in 256-bit units)
* from the start of the URB entry, as referenced by URB Handle 0, at
* which the data will be accessed.
*
* Similar text describes DWORD M0.4, which is slot 1 offset.
*
* Therefore, we want to multiply DWORDs 0 and 4 of src0 (the x components
* of the register for geometry shader invocations 0 and 1) by the
* immediate value in src1, and store the result in DWORDs 3 and 4 of dst.
*
* We can do this with the following EU instruction:
*
* mul(2) dst.3<1>UD src0<8;2,4>UD src1 { Align1 WE_all }
*/
default_state.access_mode = BRW_ALIGN_1;
gen8_instruction *inst =
MUL(suboffset(stride(dst, 2, 2, 1), 3), stride(src0, 8, 2, 4), src1);
gen8_set_mask_control(inst, BRW_MASK_DISABLE);
default_state.access_mode = BRW_ALIGN_16;
}
gen8_instruction *
gen8_generator::next_inst(unsigned opcode)
{
gen8_instruction *inst;
if (nr_inst + 1 > unsigned(store_size)) {
store_size <<= 1;
store = reralloc(mem_ctx, store, gen8_instruction, store_size);
assert(store);
}
next_inst_offset += 16;
inst = &store[nr_inst++];
memset(inst, 0, sizeof(gen8_instruction));
gen8_set_opcode(inst, opcode);
gen8_set_exec_size(inst, default_state.exec_size);
gen8_set_access_mode(inst, default_state.access_mode);
gen8_set_mask_control(inst, default_state.mask_control);
gen8_set_qtr_control(inst, default_state.qtr_control);
gen8_set_cond_modifier(inst, default_state.conditional_mod);
gen8_set_pred_control(inst, default_state.predicate);
gen8_set_pred_inv(inst, default_state.predicate_inverse);
gen8_set_saturate(inst, default_state.saturate);
gen8_set_flag_subreg_nr(inst, default_state.flag_subreg_nr);
return inst;
}
void
gen8_vec4_generator::generate_urb_write(vec4_instruction *ir, bool vs)
{
struct brw_reg header = brw_vec8_grf(GEN7_MRF_HACK_START + ir->base_mrf, 0);
/* Copy g0. */
if (vs)
MOV_RAW(header, brw_vec8_grf(0, 0));
gen8_instruction *inst;
if (!(ir->urb_write_flags & BRW_URB_WRITE_USE_CHANNEL_MASKS)) {
/* Enable Channel Masks in the URB_WRITE_OWORD message header */
default_state.access_mode = BRW_ALIGN_1;
inst = OR(retype(brw_vec1_grf(GEN7_MRF_HACK_START + ir->base_mrf, 5),
BRW_REGISTER_TYPE_UD),
retype(brw_vec1_grf(0, 5), BRW_REGISTER_TYPE_UD),
brw_imm_ud(0xff00));
gen8_set_mask_control(inst, BRW_MASK_DISABLE);
default_state.access_mode = BRW_ALIGN_16;
}
inst = next_inst(BRW_OPCODE_SEND);
gen8_set_urb_message(brw, inst, ir->urb_write_flags, ir->mlen, 0, ir->offset,
true);
gen8_set_dst(brw, inst, brw_null_reg());
gen8_set_src0(brw, inst, header);
}
gen8_instruction *
gen8_generator::MOV_RAW(struct brw_reg dst, struct brw_reg src0)
{
gen8_instruction *inst = next_inst(BRW_OPCODE_MOV);
gen8_set_dst(brw, inst, retype(dst, BRW_REGISTER_TYPE_UD));
gen8_set_src0(brw, inst, retype(src0, BRW_REGISTER_TYPE_UD));
gen8_set_mask_control(inst, BRW_MASK_DISABLE);
return inst;
}
void
gen8_vec4_generator::generate_gs_set_dword_2_immed(struct brw_reg dst,
struct brw_reg src)
{
assert(src.file == BRW_IMMEDIATE_VALUE);
default_state.access_mode = BRW_ALIGN_1;
gen8_instruction *inst = MOV(suboffset(vec1(dst), 2), src);
gen8_set_mask_control(inst, BRW_MASK_DISABLE);
default_state.access_mode = BRW_ALIGN_16;
}
void
gen8_vec4_generator::generate_gs_prepare_channel_masks(struct brw_reg dst)
{
/* We want to left shift just DWORD 4 (the x component belonging to the
* second geometry shader invocation) by 4 bits. So generate the
* instruction:
*
* shl(1) dst.4<1>UD dst.4<0,1,0>UD 4UD { align1 WE_all }
*/
dst = suboffset(vec1(dst), 4);
default_state.access_mode = BRW_ALIGN_1;
gen8_instruction *inst = SHL(dst, dst, brw_imm_ud(4));
gen8_set_mask_control(inst, BRW_MASK_DISABLE);
default_state.access_mode = BRW_ALIGN_16;
}
void
gen8_vec4_generator::generate_gs_thread_end(vec4_instruction *ir)
{
struct brw_reg src = brw_vec8_grf(GEN7_MRF_HACK_START + ir->base_mrf, 0);
gen8_instruction *inst;
/* Enable Channel Masks in the URB_WRITE_HWORD message header */
default_state.access_mode = BRW_ALIGN_1;
inst = OR(retype(brw_vec1_grf(GEN7_MRF_HACK_START + ir->base_mrf, 5),
BRW_REGISTER_TYPE_UD),
retype(brw_vec1_grf(0, 5), BRW_REGISTER_TYPE_UD),
brw_imm_ud(0xff00)); /* could be 0x1100 but shouldn't matter */
gen8_set_mask_control(inst, BRW_MASK_DISABLE);
default_state.access_mode = BRW_ALIGN_16;
/* mlen = 2: g0 header + vertex count */
inst = next_inst(BRW_OPCODE_SEND);
gen8_set_urb_message(brw, inst, BRW_URB_WRITE_EOT, 2, 0, 0, true);
gen8_set_dst(brw, inst, brw_null_reg());
gen8_set_src0(brw, inst, src);
}
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
gen8_vec4_generator::generate_gs_set_channel_masks(struct brw_reg dst,
struct brw_reg src)
{
/* From p21 of volume 4 part 2 of the Ivy Bridge PRM (2.4.3.1 Message
* Header: M0.5):
*
* 15 Vertex 1 DATA [3] / Vertex 0 DATA[7] Channel Mask
*
* When Swizzle Control = URB_INTERLEAVED this bit controls Vertex 1
* DATA[3], when Swizzle Control = URB_NOSWIZZLE this bit controls
* Vertex 0 DATA[7]. This bit is ANDed with the corresponding
* channel enable to determine the final channel enable. For the
* URB_READ_OWORD & URB_READ_HWORD messages, when final channel
* enable is 1 it indicates that Vertex 1 DATA [3] will be included
* in the writeback message. For the URB_WRITE_OWORD &
* URB_WRITE_HWORD messages, when final channel enable is 1 it
* indicates that Vertex 1 DATA [3] will be written to the surface.
*
* 0: Vertex 1 DATA [3] / Vertex 0 DATA[7] channel not included
* 1: Vertex DATA [3] / Vertex 0 DATA[7] channel included
*
* 14 Vertex 1 DATA [2] Channel Mask
* 13 Vertex 1 DATA [1] Channel Mask
* 12 Vertex 1 DATA [0] Channel Mask
* 11 Vertex 0 DATA [3] Channel Mask
* 10 Vertex 0 DATA [2] Channel Mask
* 9 Vertex 0 DATA [1] Channel Mask
* 8 Vertex 0 DATA [0] Channel Mask
*
* (This is from a section of the PRM that is agnostic to the particular
* type of shader being executed, so "Vertex 0" and "Vertex 1" refer to
* geometry shader invocations 0 and 1, respectively). Since we have the
* enable flags for geometry shader invocation 0 in bits 3:0 of DWORD 0,
* and the enable flags for geometry shader invocation 1 in bits 7:0 of
* DWORD 4, we just need to OR them together and store the result in bits
* 15:8 of DWORD 5.
*
* It's easier to get the EU to do this if we think of the src and dst
* registers as composed of 32 bytes each; then, we want to pick up the
* contents of bytes 0 and 16 from src, OR them together, and store them in
* byte 21.
*
* We can do that by the following EU instruction:
*
* or(1) dst.21<1>UB src<0,1,0>UB src.16<0,1,0>UB { align1 WE_all }
*
* Note: this relies on the source register having zeros in (a) bits 7:4 of
* DWORD 0 and (b) bits 3:0 of DWORD 4. We can rely on (b) because the
* source register was prepared by GS_OPCODE_PREPARE_CHANNEL_MASKS (which
* shifts DWORD 4 left by 4 bits), and we can rely on (a) because prior to
* the execution of GS_OPCODE_PREPARE_CHANNEL_MASKS, DWORDs 0 and 4 need to
* contain valid channel mask values (which are in the range 0x0-0xf).
*/
dst = retype(dst, BRW_REGISTER_TYPE_UB);
src = retype(src, BRW_REGISTER_TYPE_UB);
default_state.access_mode = BRW_ALIGN_1;
gen8_instruction *inst =
OR(suboffset(vec1(dst), 21), vec1(src), suboffset(vec1(src), 16));
gen8_set_mask_control(inst, BRW_MASK_DISABLE);
default_state.access_mode = BRW_ALIGN_16;
}