void brw_clip_init_clipmask( struct brw_clip_compile *c )
{
struct brw_codegen *p = &c->func;
struct brw_reg incoming = get_element_ud(c->reg.R0, 2);
/* Shift so that lowest outcode bit is rightmost:
*/
brw_SHR(p, c->reg.planemask, incoming, brw_imm_ud(26));
if (c->key.nr_userclip) {
struct brw_reg tmp = retype(vec1(get_tmp(c)), BRW_REGISTER_TYPE_UD);
/* Rearrange userclip outcodes so that they come directly after
* the fixed plane bits.
*/
if (p->devinfo->gen == 5 || p->devinfo->is_g4x)
brw_AND(p, tmp, incoming, brw_imm_ud(0xff<<14));
else
brw_AND(p, tmp, incoming, brw_imm_ud(0x3f<<14));
brw_SHR(p, tmp, tmp, brw_imm_ud(8));
brw_OR(p, c->reg.planemask, c->reg.planemask, tmp);
release_tmp(c, tmp);
}
}
void
vec4_generator::generate_gs_get_instance_id(struct brw_reg dst)
{
/* We want to right shift R0.0 & R0.1 by GEN7_GS_PAYLOAD_INSTANCE_ID_SHIFT
* and store into dst.0 & dst.4. So generate the instruction:
*
* shr(8) dst<1> R0<1,4,0> GEN7_GS_PAYLOAD_INSTANCE_ID_SHIFT { align1 WE_normal 1Q }
*/
brw_push_insn_state(p);
brw_set_access_mode(p, BRW_ALIGN_1);
dst = retype(dst, BRW_REGISTER_TYPE_UD);
struct brw_reg r0(retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UD));
brw_SHR(p, dst, stride(r0, 1, 4, 0),
brw_imm_ud(GEN7_GS_PAYLOAD_INSTANCE_ID_SHIFT));
brw_pop_insn_state(p);
}
void
vec4_generator::generate_unpack_flags(vec4_instruction *inst,
struct brw_reg dst)
{
brw_push_insn_state(p);
brw_set_mask_control(p, BRW_MASK_DISABLE);
brw_set_access_mode(p, BRW_ALIGN_1);
struct brw_reg flags = brw_flag_reg(0, 0);
struct brw_reg dst_0 = suboffset(vec1(dst), 0);
struct brw_reg dst_4 = suboffset(vec1(dst), 4);
brw_AND(p, dst_0, flags, brw_imm_ud(0x0f));
brw_AND(p, dst_4, flags, brw_imm_ud(0xf0));
brw_SHR(p, dst_4, dst_4, brw_imm_ud(4));
brw_pop_insn_state(p);
}
/* Use mesa's clipping algorithms, translated to GEN4 assembly.
*/
void brw_clip_tri( struct brw_clip_compile *c )
{
struct brw_compile *p = &c->func;
struct brw_indirect vtx = brw_indirect(0, 0);
struct brw_indirect vtxPrev = brw_indirect(1, 0);
struct brw_indirect vtxOut = brw_indirect(2, 0);
struct brw_indirect plane_ptr = brw_indirect(3, 0);
struct brw_indirect inlist_ptr = brw_indirect(4, 0);
struct brw_indirect outlist_ptr = brw_indirect(5, 0);
struct brw_indirect freelist_ptr = brw_indirect(6, 0);
struct brw_instruction *plane_loop;
struct brw_instruction *plane_active;
struct brw_instruction *vertex_loop;
struct brw_instruction *next_test;
struct brw_instruction *prev_test;
brw_MOV(p, get_addr_reg(vtxPrev), brw_address(c->reg.vertex[2]) );
brw_MOV(p, get_addr_reg(plane_ptr), brw_clip_plane0_address(c));
brw_MOV(p, get_addr_reg(inlist_ptr), brw_address(c->reg.inlist));
brw_MOV(p, get_addr_reg(outlist_ptr), brw_address(c->reg.outlist));
brw_MOV(p, get_addr_reg(freelist_ptr), brw_address(c->reg.vertex[3]) );
plane_loop = brw_DO(p, BRW_EXECUTE_1);
{
/* if (planemask & 1)
*/
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
brw_AND(p, vec1(brw_null_reg()), c->reg.planemask, brw_imm_ud(1));
plane_active = brw_IF(p, BRW_EXECUTE_1);
{
/* vtxOut = freelist_ptr++
*/
brw_MOV(p, get_addr_reg(vtxOut), get_addr_reg(freelist_ptr) );
brw_ADD(p, get_addr_reg(freelist_ptr), get_addr_reg(freelist_ptr), brw_imm_uw(c->nr_regs * REG_SIZE));
if (c->key.nr_userclip)
brw_MOV(p, c->reg.plane_equation, deref_4f(plane_ptr, 0));
else
brw_MOV(p, c->reg.plane_equation, deref_4b(plane_ptr, 0));
brw_MOV(p, c->reg.loopcount, c->reg.nr_verts);
brw_MOV(p, c->reg.nr_verts, brw_imm_ud(0));
vertex_loop = brw_DO(p, BRW_EXECUTE_1);
{
/* vtx = *input_ptr;
*/
brw_MOV(p, get_addr_reg(vtx), deref_1uw(inlist_ptr, 0));
/* IS_NEGATIVE(prev) */
brw_set_conditionalmod(p, BRW_CONDITIONAL_L);
brw_DP4(p, vec4(c->reg.dpPrev), deref_4f(vtxPrev, c->offset_hpos), c->reg.plane_equation);
prev_test = brw_IF(p, BRW_EXECUTE_1);
{
/* IS_POSITIVE(next)
*/
brw_set_conditionalmod(p, BRW_CONDITIONAL_GE);
brw_DP4(p, vec4(c->reg.dp), deref_4f(vtx, c->offset_hpos), c->reg.plane_equation);
next_test = brw_IF(p, BRW_EXECUTE_1);
{
/* Coming back in.
*/
brw_ADD(p, c->reg.t, c->reg.dpPrev, negate(c->reg.dp));
brw_math_invert(p, c->reg.t, c->reg.t);
brw_MUL(p, c->reg.t, c->reg.t, c->reg.dpPrev);
/* If (vtxOut == 0) vtxOut = vtxPrev
*/
brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_EQ, get_addr_reg(vtxOut), brw_imm_uw(0) );
brw_MOV(p, get_addr_reg(vtxOut), get_addr_reg(vtxPrev) );
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_clip_interp_vertex(c, vtxOut, vtxPrev, vtx, c->reg.t, GL_FALSE);
/* *outlist_ptr++ = vtxOut;
* nr_verts++;
* vtxOut = 0;
*/
brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxOut));
brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short)));
brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1));
brw_MOV(p, get_addr_reg(vtxOut), brw_imm_uw(0) );
}
brw_ENDIF(p, next_test);
}
prev_test = brw_ELSE(p, prev_test);
{
/* *outlist_ptr++ = vtxPrev;
* nr_verts++;
*/
brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxPrev));
brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short)));
brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1));
/* IS_NEGATIVE(next)
*/
brw_set_conditionalmod(p, BRW_CONDITIONAL_L);
brw_DP4(p, vec4(c->reg.dp), deref_4f(vtx, c->offset_hpos), c->reg.plane_equation);
next_test = brw_IF(p, BRW_EXECUTE_1);
{
/* Going out of bounds. Avoid division by zero as we
* know dp != dpPrev from DIFFERENT_SIGNS, above.
*/
brw_ADD(p, c->reg.t, c->reg.dp, negate(c->reg.dpPrev));
brw_math_invert(p, c->reg.t, c->reg.t);
brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp);
/* If (vtxOut == 0) vtxOut = vtx
*/
brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_EQ, get_addr_reg(vtxOut), brw_imm_uw(0) );
brw_MOV(p, get_addr_reg(vtxOut), get_addr_reg(vtx) );
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_clip_interp_vertex(c, vtxOut, vtx, vtxPrev, c->reg.t, GL_TRUE);
/* *outlist_ptr++ = vtxOut;
* nr_verts++;
* vtxOut = 0;
*/
brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxOut));
brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short)));
brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1));
brw_MOV(p, get_addr_reg(vtxOut), brw_imm_uw(0) );
}
brw_ENDIF(p, next_test);
}
brw_ENDIF(p, prev_test);
/* vtxPrev = vtx;
* inlist_ptr++;
*/
brw_MOV(p, get_addr_reg(vtxPrev), get_addr_reg(vtx));
brw_ADD(p, get_addr_reg(inlist_ptr), get_addr_reg(inlist_ptr), brw_imm_uw(sizeof(short)));
/* while (--loopcount != 0)
*/
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
}
brw_WHILE(p, vertex_loop);
/* vtxPrev = *(outlist_ptr-1) OR: outlist[nr_verts-1]
* inlist = outlist
* inlist_ptr = &inlist[0]
* outlist_ptr = &outlist[0]
*/
brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_w(-2));
brw_MOV(p, get_addr_reg(vtxPrev), deref_1uw(outlist_ptr, 0));
brw_MOV(p, brw_vec8_grf(c->reg.inlist.nr, 0), brw_vec8_grf(c->reg.outlist.nr, 0));
brw_MOV(p, get_addr_reg(inlist_ptr), brw_address(c->reg.inlist));
brw_MOV(p, get_addr_reg(outlist_ptr), brw_address(c->reg.outlist));
}
brw_ENDIF(p, plane_active);
/* plane_ptr++;
*/
brw_ADD(p, get_addr_reg(plane_ptr), get_addr_reg(plane_ptr), brw_clip_plane_stride(c));
/* nr_verts >= 3
*/
brw_CMP(p,
vec1(brw_null_reg()),
BRW_CONDITIONAL_GE,
c->reg.nr_verts,
brw_imm_ud(3));
/* && (planemask>>=1) != 0
*/
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
brw_SHR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(1));
}
brw_WHILE(p, plane_loop);
}
/**
* Generate assembly for a Vec4 IR instruction.
*
* \param instruction The Vec4 IR instruction to generate code for.
* \param dst The destination register.
* \param src An array of up to three source registers.
*/
void
vec4_generator::generate_vec4_instruction(vec4_instruction *instruction,
struct brw_reg dst,
struct brw_reg *src)
{
vec4_instruction *inst = (vec4_instruction *) instruction;
if (dst.width == BRW_WIDTH_4) {
/* This happens in attribute fixups for "dual instanced" geometry
* shaders, since they use attributes that are vec4's. Since the exec
* width is only 4, it's essential that the caller set
* force_writemask_all in order to make sure the instruction is executed
* regardless of which channels are enabled.
*/
assert(inst->force_writemask_all);
/* Fix up any <8;8,1> or <0;4,1> source registers to <4;4,1> to satisfy
* the following register region restrictions (from Graphics BSpec:
* 3D-Media-GPGPU Engine > EU Overview > Registers and Register Regions
* > Register Region Restrictions)
*
* 1. ExecSize must be greater than or equal to Width.
*
* 2. If ExecSize = Width and HorzStride != 0, VertStride must be set
* to Width * HorzStride."
*/
for (int i = 0; i < 3; i++) {
if (src[i].file == BRW_GENERAL_REGISTER_FILE)
src[i] = stride(src[i], 4, 4, 1);
}
}
switch (inst->opcode) {
case BRW_OPCODE_MOV:
brw_MOV(p, dst, src[0]);
break;
case BRW_OPCODE_ADD:
brw_ADD(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_MUL:
brw_MUL(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_MACH:
brw_set_acc_write_control(p, 1);
brw_MACH(p, dst, src[0], src[1]);
brw_set_acc_write_control(p, 0);
break;
case BRW_OPCODE_MAD:
assert(brw->gen >= 6);
brw_MAD(p, dst, src[0], src[1], src[2]);
break;
case BRW_OPCODE_FRC:
brw_FRC(p, dst, src[0]);
break;
case BRW_OPCODE_RNDD:
brw_RNDD(p, dst, src[0]);
break;
case BRW_OPCODE_RNDE:
brw_RNDE(p, dst, src[0]);
break;
case BRW_OPCODE_RNDZ:
brw_RNDZ(p, dst, src[0]);
break;
case BRW_OPCODE_AND:
brw_AND(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_OR:
brw_OR(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_XOR:
brw_XOR(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_NOT:
brw_NOT(p, dst, src[0]);
break;
case BRW_OPCODE_ASR:
brw_ASR(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_SHR:
brw_SHR(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_SHL:
brw_SHL(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_CMP:
brw_CMP(p, dst, inst->conditional_mod, src[0], src[1]);
break;
case BRW_OPCODE_SEL:
brw_SEL(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_DPH:
brw_DPH(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_DP4:
brw_DP4(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_DP3:
brw_DP3(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_DP2:
brw_DP2(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_F32TO16:
assert(brw->gen >= 7);
brw_F32TO16(p, dst, src[0]);
break;
case BRW_OPCODE_F16TO32:
assert(brw->gen >= 7);
brw_F16TO32(p, dst, src[0]);
break;
case BRW_OPCODE_LRP:
assert(brw->gen >= 6);
brw_LRP(p, dst, src[0], src[1], src[2]);
break;
case BRW_OPCODE_BFREV:
assert(brw->gen >= 7);
/* BFREV only supports UD type for src and dst. */
brw_BFREV(p, retype(dst, BRW_REGISTER_TYPE_UD),
retype(src[0], BRW_REGISTER_TYPE_UD));
break;
case BRW_OPCODE_FBH:
assert(brw->gen >= 7);
/* FBH only supports UD type for dst. */
brw_FBH(p, retype(dst, BRW_REGISTER_TYPE_UD), src[0]);
break;
case BRW_OPCODE_FBL:
assert(brw->gen >= 7);
/* FBL only supports UD type for dst. */
brw_FBL(p, retype(dst, BRW_REGISTER_TYPE_UD), src[0]);
break;
case BRW_OPCODE_CBIT:
assert(brw->gen >= 7);
/* CBIT only supports UD type for dst. */
brw_CBIT(p, retype(dst, BRW_REGISTER_TYPE_UD), src[0]);
break;
case BRW_OPCODE_ADDC:
assert(brw->gen >= 7);
brw_set_acc_write_control(p, 1);
brw_ADDC(p, dst, src[0], src[1]);
brw_set_acc_write_control(p, 0);
break;
case BRW_OPCODE_SUBB:
assert(brw->gen >= 7);
brw_set_acc_write_control(p, 1);
brw_SUBB(p, dst, src[0], src[1]);
brw_set_acc_write_control(p, 0);
break;
case BRW_OPCODE_BFE:
assert(brw->gen >= 7);
brw_BFE(p, dst, src[0], src[1], src[2]);
break;
case BRW_OPCODE_BFI1:
assert(brw->gen >= 7);
brw_BFI1(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_BFI2:
assert(brw->gen >= 7);
brw_BFI2(p, dst, src[0], src[1], src[2]);
break;
case BRW_OPCODE_IF:
if (inst->src[0].file != BAD_FILE) {
/* The instruction has an embedded compare (only allowed on gen6) */
assert(brw->gen == 6);
gen6_IF(p, inst->conditional_mod, src[0], src[1]);
} else {
struct brw_instruction *brw_inst = brw_IF(p, BRW_EXECUTE_8);
brw_inst->header.predicate_control = inst->predicate;
}
break;
case BRW_OPCODE_ELSE:
brw_ELSE(p);
break;
case BRW_OPCODE_ENDIF:
brw_ENDIF(p);
break;
case BRW_OPCODE_DO:
brw_DO(p, BRW_EXECUTE_8);
break;
case BRW_OPCODE_BREAK:
brw_BREAK(p);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
break;
case BRW_OPCODE_CONTINUE:
/* FINISHME: We need to write the loop instruction support still. */
if (brw->gen >= 6)
gen6_CONT(p);
else
brw_CONT(p);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
break;
case BRW_OPCODE_WHILE:
brw_WHILE(p);
break;
case SHADER_OPCODE_RCP:
case SHADER_OPCODE_RSQ:
case SHADER_OPCODE_SQRT:
case SHADER_OPCODE_EXP2:
case SHADER_OPCODE_LOG2:
case SHADER_OPCODE_SIN:
case SHADER_OPCODE_COS:
if (brw->gen == 6) {
generate_math1_gen6(inst, dst, src[0]);
} else {
/* Also works for Gen7. */
generate_math1_gen4(inst, dst, src[0]);
}
break;
case SHADER_OPCODE_POW:
case SHADER_OPCODE_INT_QUOTIENT:
case SHADER_OPCODE_INT_REMAINDER:
if (brw->gen >= 7) {
generate_math2_gen7(inst, dst, src[0], src[1]);
} else if (brw->gen == 6) {
generate_math2_gen6(inst, dst, src[0], src[1]);
} else {
generate_math2_gen4(inst, dst, src[0], src[1]);
}
break;
case SHADER_OPCODE_TEX:
case SHADER_OPCODE_TXD:
case SHADER_OPCODE_TXF:
case SHADER_OPCODE_TXF_CMS:
case SHADER_OPCODE_TXF_MCS:
case SHADER_OPCODE_TXL:
case SHADER_OPCODE_TXS:
case SHADER_OPCODE_TG4:
case SHADER_OPCODE_TG4_OFFSET:
generate_tex(inst, dst, src[0]);
break;
case VS_OPCODE_URB_WRITE:
generate_vs_urb_write(inst);
break;
case SHADER_OPCODE_GEN4_SCRATCH_READ:
generate_scratch_read(inst, dst, src[0]);
break;
case SHADER_OPCODE_GEN4_SCRATCH_WRITE:
generate_scratch_write(inst, dst, src[0], src[1]);
break;
case VS_OPCODE_PULL_CONSTANT_LOAD:
generate_pull_constant_load(inst, dst, src[0], src[1]);
break;
case VS_OPCODE_PULL_CONSTANT_LOAD_GEN7:
generate_pull_constant_load_gen7(inst, dst, src[0], src[1]);
break;
case GS_OPCODE_URB_WRITE:
generate_gs_urb_write(inst);
break;
case GS_OPCODE_THREAD_END:
generate_gs_thread_end(inst);
break;
case GS_OPCODE_SET_WRITE_OFFSET:
generate_gs_set_write_offset(dst, src[0], src[1]);
break;
case GS_OPCODE_SET_VERTEX_COUNT:
generate_gs_set_vertex_count(dst, src[0]);
break;
case GS_OPCODE_SET_DWORD_2_IMMED:
generate_gs_set_dword_2_immed(dst, src[0]);
break;
case GS_OPCODE_PREPARE_CHANNEL_MASKS:
generate_gs_prepare_channel_masks(dst);
break;
case GS_OPCODE_SET_CHANNEL_MASKS:
generate_gs_set_channel_masks(dst, src[0]);
break;
case GS_OPCODE_GET_INSTANCE_ID:
generate_gs_get_instance_id(dst);
break;
case SHADER_OPCODE_SHADER_TIME_ADD:
brw_shader_time_add(p, src[0],
prog_data->base.binding_table.shader_time_start);
brw_mark_surface_used(&prog_data->base,
prog_data->base.binding_table.shader_time_start);
break;
case SHADER_OPCODE_UNTYPED_ATOMIC:
generate_untyped_atomic(inst, dst, src[0], src[1]);
break;
case SHADER_OPCODE_UNTYPED_SURFACE_READ:
generate_untyped_surface_read(inst, dst, src[0]);
break;
case VS_OPCODE_UNPACK_FLAGS_SIMD4X2:
generate_unpack_flags(inst, dst);
break;
default:
if (inst->opcode < (int) ARRAY_SIZE(opcode_descs)) {
_mesa_problem(&brw->ctx, "Unsupported opcode in `%s' in vec4\n",
opcode_descs[inst->opcode].name);
} else {
_mesa_problem(&brw->ctx, "Unsupported opcode %d in vec4", inst->opcode);
}
abort();
}
}
/* Line clipping, more or less following the following algorithm:
*
* for (p=0;p<MAX_PLANES;p++) {
* if (clipmask & (1 << p)) {
* GLfloat dp0 = DOTPROD( vtx0, plane[p] );
* GLfloat dp1 = DOTPROD( vtx1, plane[p] );
*
* if (dp1 < 0.0f) {
* GLfloat t = dp1 / (dp1 - dp0);
* if (t > t1) t1 = t;
* } else {
* GLfloat t = dp0 / (dp0 - dp1);
* if (t > t0) t0 = t;
* }
*
* if (t0 + t1 >= 1.0)
* return;
* }
* }
*
* interp( ctx, newvtx0, vtx0, vtx1, t0 );
* interp( ctx, newvtx1, vtx1, vtx0, t1 );
*
*/
static void clip_and_emit_line( struct brw_clip_compile *c )
{
struct brw_codegen *p = &c->func;
struct brw_indirect vtx0 = brw_indirect(0, 0);
struct brw_indirect vtx1 = brw_indirect(1, 0);
struct brw_indirect newvtx0 = brw_indirect(2, 0);
struct brw_indirect newvtx1 = brw_indirect(3, 0);
struct brw_indirect plane_ptr = brw_indirect(4, 0);
struct brw_reg v1_null_ud = retype(vec1(brw_null_reg()), BRW_REGISTER_TYPE_UD);
GLuint hpos_offset = brw_varying_to_offset(&c->vue_map, VARYING_SLOT_POS);
GLint clipdist0_offset = c->key.nr_userclip
? brw_varying_to_offset(&c->vue_map, VARYING_SLOT_CLIP_DIST0)
: 0;
brw_MOV(p, get_addr_reg(vtx0), brw_address(c->reg.vertex[0]));
brw_MOV(p, get_addr_reg(vtx1), brw_address(c->reg.vertex[1]));
brw_MOV(p, get_addr_reg(newvtx0), brw_address(c->reg.vertex[2]));
brw_MOV(p, get_addr_reg(newvtx1), brw_address(c->reg.vertex[3]));
brw_MOV(p, get_addr_reg(plane_ptr), brw_clip_plane0_address(c));
/* Note: init t0, t1 together:
*/
brw_MOV(p, vec2(c->reg.t0), brw_imm_f(0));
brw_clip_init_planes(c);
brw_clip_init_clipmask(c);
/* -ve rhw workaround */
if (p->devinfo->has_negative_rhw_bug) {
brw_AND(p, brw_null_reg(), get_element_ud(c->reg.R0, 2),
brw_imm_ud(1<<20));
brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ);
brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(0x3f));
brw_inst_set_pred_control(p->devinfo, brw_last_inst, BRW_PREDICATE_NORMAL);
}
/* Set the initial vertex source mask: The first 6 planes are the bounds
* of the view volume; the next 8 planes are the user clipping planes.
*/
brw_MOV(p, c->reg.vertex_src_mask, brw_imm_ud(0x3fc0));
/* Set the initial clipdistance offset to be 6 floats before gl_ClipDistance[0].
* We'll increment 6 times before we start hitting actual user clipping. */
brw_MOV(p, c->reg.clipdistance_offset, brw_imm_d(clipdist0_offset - 6*sizeof(float)));
brw_DO(p, BRW_EXECUTE_1);
{
/* if (planemask & 1)
*/
brw_AND(p, v1_null_ud, c->reg.planemask, brw_imm_ud(1));
brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ);
brw_IF(p, BRW_EXECUTE_1);
{
brw_AND(p, v1_null_ud, c->reg.vertex_src_mask, brw_imm_ud(1));
brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ);
brw_IF(p, BRW_EXECUTE_1);
{
/* user clip distance: just fetch the correct float from each vertex */
struct brw_indirect temp_ptr = brw_indirect(7, 0);
brw_ADD(p, get_addr_reg(temp_ptr), get_addr_reg(vtx0), c->reg.clipdistance_offset);
brw_MOV(p, c->reg.dp0, deref_1f(temp_ptr, 0));
brw_ADD(p, get_addr_reg(temp_ptr), get_addr_reg(vtx1), c->reg.clipdistance_offset);
brw_MOV(p, c->reg.dp1, deref_1f(temp_ptr, 0));
}
brw_ELSE(p);
{
/* fixed plane: fetch the hpos, dp4 against the plane. */
if (c->key.nr_userclip)
brw_MOV(p, c->reg.plane_equation, deref_4f(plane_ptr, 0));
else
brw_MOV(p, c->reg.plane_equation, deref_4b(plane_ptr, 0));
brw_DP4(p, vec4(c->reg.dp0), deref_4f(vtx0, hpos_offset), c->reg.plane_equation);
brw_DP4(p, vec4(c->reg.dp1), deref_4f(vtx1, hpos_offset), c->reg.plane_equation);
}
brw_ENDIF(p);
brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, vec1(c->reg.dp1), brw_imm_f(0.0f));
brw_IF(p, BRW_EXECUTE_1);
{
/*
* Both can be negative on GM965/G965 due to RHW workaround
* if so, this object should be rejected.
*/
if (p->devinfo->has_negative_rhw_bug) {
brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_LE, c->reg.dp0, brw_imm_f(0.0));
brw_IF(p, BRW_EXECUTE_1);
{
brw_clip_kill_thread(c);
}
brw_ENDIF(p);
}
brw_ADD(p, c->reg.t, c->reg.dp1, negate(c->reg.dp0));
brw_math_invert(p, c->reg.t, c->reg.t);
brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp1);
brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_G, c->reg.t, c->reg.t1 );
brw_MOV(p, c->reg.t1, c->reg.t);
brw_inst_set_pred_control(p->devinfo, brw_last_inst,
BRW_PREDICATE_NORMAL);
}
brw_ELSE(p);
{
/* Coming back in. We know that both cannot be negative
* because the line would have been culled in that case.
*/
/* If both are positive, do nothing */
/* Only on GM965/G965 */
if (p->devinfo->has_negative_rhw_bug) {
brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_L, c->reg.dp0, brw_imm_f(0.0));
brw_IF(p, BRW_EXECUTE_1);
}
{
brw_ADD(p, c->reg.t, c->reg.dp0, negate(c->reg.dp1));
brw_math_invert(p, c->reg.t, c->reg.t);
brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp0);
brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_G, c->reg.t, c->reg.t0 );
brw_MOV(p, c->reg.t0, c->reg.t);
brw_inst_set_pred_control(p->devinfo, brw_last_inst,
BRW_PREDICATE_NORMAL);
}
if (p->devinfo->has_negative_rhw_bug) {
brw_ENDIF(p);
}
}
brw_ENDIF(p);
}
brw_ENDIF(p);
/* plane_ptr++;
*/
brw_ADD(p, get_addr_reg(plane_ptr), get_addr_reg(plane_ptr), brw_clip_plane_stride(c));
/* while (planemask>>=1) != 0
*/
brw_SHR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(1));
brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ);
brw_SHR(p, c->reg.vertex_src_mask, c->reg.vertex_src_mask, brw_imm_ud(1));
brw_inst_set_pred_control(p->devinfo, brw_last_inst, BRW_PREDICATE_NORMAL);
brw_ADD(p, c->reg.clipdistance_offset, c->reg.clipdistance_offset, brw_imm_w(sizeof(float)));
brw_inst_set_pred_control(p->devinfo, brw_last_inst, BRW_PREDICATE_NORMAL);
}
brw_WHILE(p);
brw_inst_set_pred_control(p->devinfo, brw_last_inst, BRW_PREDICATE_NORMAL);
brw_ADD(p, c->reg.t, c->reg.t0, c->reg.t1);
brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_L, c->reg.t, brw_imm_f(1.0));
brw_IF(p, BRW_EXECUTE_1);
{
brw_clip_interp_vertex(c, newvtx0, vtx0, vtx1, c->reg.t0, false);
brw_clip_interp_vertex(c, newvtx1, vtx1, vtx0, c->reg.t1, false);
brw_clip_emit_vue(c, newvtx0, BRW_URB_WRITE_ALLOCATE_COMPLETE,
(_3DPRIM_LINESTRIP << URB_WRITE_PRIM_TYPE_SHIFT)
| URB_WRITE_PRIM_START);
brw_clip_emit_vue(c, newvtx1, BRW_URB_WRITE_EOT_COMPLETE,
(_3DPRIM_LINESTRIP << URB_WRITE_PRIM_TYPE_SHIFT)
| URB_WRITE_PRIM_END);
}
brw_ENDIF(p);
brw_clip_kill_thread(c);
}
/* Use mesa's clipping algorithms, translated to GEN4 assembly.
*/
void brw_clip_tri( struct brw_clip_compile *c )
{
struct brw_codegen *p = &c->func;
struct brw_indirect vtx = brw_indirect(0, 0);
struct brw_indirect vtxPrev = brw_indirect(1, 0);
struct brw_indirect vtxOut = brw_indirect(2, 0);
struct brw_indirect plane_ptr = brw_indirect(3, 0);
struct brw_indirect inlist_ptr = brw_indirect(4, 0);
struct brw_indirect outlist_ptr = brw_indirect(5, 0);
struct brw_indirect freelist_ptr = brw_indirect(6, 0);
GLuint hpos_offset = brw_varying_to_offset(&c->vue_map, VARYING_SLOT_POS);
GLint clipdist0_offset = c->key.nr_userclip
? brw_varying_to_offset(&c->vue_map, VARYING_SLOT_CLIP_DIST0)
: 0;
brw_MOV(p, get_addr_reg(vtxPrev), brw_address(c->reg.vertex[2]) );
brw_MOV(p, get_addr_reg(plane_ptr), brw_clip_plane0_address(c));
brw_MOV(p, get_addr_reg(inlist_ptr), brw_address(c->reg.inlist));
brw_MOV(p, get_addr_reg(outlist_ptr), brw_address(c->reg.outlist));
brw_MOV(p, get_addr_reg(freelist_ptr), brw_address(c->reg.vertex[3]) );
/* Set the initial vertex source mask: The first 6 planes are the bounds
* of the view volume; the next 8 planes are the user clipping planes.
*/
brw_MOV(p, c->reg.vertex_src_mask, brw_imm_ud(0x3fc0));
/* Set the initial clipdistance offset to be 6 floats before gl_ClipDistance[0].
* We'll increment 6 times before we start hitting actual user clipping. */
brw_MOV(p, c->reg.clipdistance_offset, brw_imm_d(clipdist0_offset - 6*sizeof(float)));
brw_DO(p, BRW_EXECUTE_1);
{
/* if (planemask & 1)
*/
brw_AND(p, vec1(brw_null_reg()), c->reg.planemask, brw_imm_ud(1));
brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ);
brw_IF(p, BRW_EXECUTE_1);
{
/* vtxOut = freelist_ptr++
*/
brw_MOV(p, get_addr_reg(vtxOut), get_addr_reg(freelist_ptr) );
brw_ADD(p, get_addr_reg(freelist_ptr), get_addr_reg(freelist_ptr), brw_imm_uw(c->nr_regs * REG_SIZE));
if (c->key.nr_userclip)
brw_MOV(p, c->reg.plane_equation, deref_4f(plane_ptr, 0));
else
brw_MOV(p, c->reg.plane_equation, deref_4b(plane_ptr, 0));
brw_MOV(p, c->reg.loopcount, c->reg.nr_verts);
brw_MOV(p, c->reg.nr_verts, brw_imm_ud(0));
brw_DO(p, BRW_EXECUTE_1);
{
/* vtx = *input_ptr;
*/
brw_MOV(p, get_addr_reg(vtx), deref_1uw(inlist_ptr, 0));
load_clip_distance(c, vtxPrev, c->reg.dpPrev, hpos_offset, BRW_CONDITIONAL_L);
/* (prev < 0.0f) */
brw_IF(p, BRW_EXECUTE_1);
{
load_clip_distance(c, vtx, c->reg.dp, hpos_offset, BRW_CONDITIONAL_GE);
/* IS_POSITIVE(next)
*/
brw_IF(p, BRW_EXECUTE_1);
{
/* Coming back in.
*/
brw_ADD(p, c->reg.t, c->reg.dpPrev, negate(c->reg.dp));
brw_math_invert(p, c->reg.t, c->reg.t);
brw_MUL(p, c->reg.t, c->reg.t, c->reg.dpPrev);
/* If (vtxOut == 0) vtxOut = vtxPrev
*/
brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_EQ, get_addr_reg(vtxOut), brw_imm_uw(0) );
brw_MOV(p, get_addr_reg(vtxOut), get_addr_reg(vtxPrev));
brw_inst_set_pred_control(p->devinfo, brw_last_inst,
BRW_PREDICATE_NORMAL);
brw_clip_interp_vertex(c, vtxOut, vtxPrev, vtx, c->reg.t, false);
/* *outlist_ptr++ = vtxOut;
* nr_verts++;
* vtxOut = 0;
*/
brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxOut));
brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short)));
brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1));
brw_MOV(p, get_addr_reg(vtxOut), brw_imm_uw(0) );
}
brw_ENDIF(p);
}
brw_ELSE(p);
{
/* *outlist_ptr++ = vtxPrev;
* nr_verts++;
*/
brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxPrev));
brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short)));
brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1));
load_clip_distance(c, vtx, c->reg.dp, hpos_offset, BRW_CONDITIONAL_L);
/* (next < 0.0f)
*/
brw_IF(p, BRW_EXECUTE_1);
{
/* Going out of bounds. Avoid division by zero as we
* know dp != dpPrev from DIFFERENT_SIGNS, above.
*/
brw_ADD(p, c->reg.t, c->reg.dp, negate(c->reg.dpPrev));
brw_math_invert(p, c->reg.t, c->reg.t);
brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp);
/* If (vtxOut == 0) vtxOut = vtx
*/
brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_EQ, get_addr_reg(vtxOut), brw_imm_uw(0) );
brw_MOV(p, get_addr_reg(vtxOut), get_addr_reg(vtx));
brw_inst_set_pred_control(p->devinfo, brw_last_inst,
BRW_PREDICATE_NORMAL);
brw_clip_interp_vertex(c, vtxOut, vtx, vtxPrev, c->reg.t, true);
/* *outlist_ptr++ = vtxOut;
* nr_verts++;
* vtxOut = 0;
*/
brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxOut));
brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short)));
brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1));
brw_MOV(p, get_addr_reg(vtxOut), brw_imm_uw(0) );
}
brw_ENDIF(p);
}
brw_ENDIF(p);
/* vtxPrev = vtx;
* inlist_ptr++;
*/
brw_MOV(p, get_addr_reg(vtxPrev), get_addr_reg(vtx));
brw_ADD(p, get_addr_reg(inlist_ptr), get_addr_reg(inlist_ptr), brw_imm_uw(sizeof(short)));
/* while (--loopcount != 0)
*/
brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ);
}
brw_WHILE(p);
brw_inst_set_pred_control(p->devinfo, brw_last_inst, BRW_PREDICATE_NORMAL);
/* vtxPrev = *(outlist_ptr-1) OR: outlist[nr_verts-1]
* inlist = outlist
* inlist_ptr = &inlist[0]
* outlist_ptr = &outlist[0]
*/
brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_w(-2));
brw_MOV(p, get_addr_reg(vtxPrev), deref_1uw(outlist_ptr, 0));
brw_MOV(p, brw_vec8_grf(c->reg.inlist.nr, 0), brw_vec8_grf(c->reg.outlist.nr, 0));
brw_MOV(p, get_addr_reg(inlist_ptr), brw_address(c->reg.inlist));
brw_MOV(p, get_addr_reg(outlist_ptr), brw_address(c->reg.outlist));
}
brw_ENDIF(p);
/* plane_ptr++;
*/
brw_ADD(p, get_addr_reg(plane_ptr), get_addr_reg(plane_ptr), brw_clip_plane_stride(c));
/* nr_verts >= 3
*/
brw_CMP(p,
vec1(brw_null_reg()),
BRW_CONDITIONAL_GE,
c->reg.nr_verts,
brw_imm_ud(3));
brw_set_default_predicate_control(p, BRW_PREDICATE_NORMAL);
/* && (planemask>>=1) != 0
*/
brw_SHR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(1));
brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ);
brw_SHR(p, c->reg.vertex_src_mask, c->reg.vertex_src_mask, brw_imm_ud(1));
brw_ADD(p, c->reg.clipdistance_offset, c->reg.clipdistance_offset, brw_imm_w(sizeof(float)));
}
brw_WHILE(p);
brw_set_default_predicate_control(p, BRW_PREDICATE_NONE);
}
void
vec4_generator::generate_code(exec_list *instructions)
{
int last_native_insn_offset = 0;
const char *last_annotation_string = NULL;
const void *last_annotation_ir = NULL;
if (unlikely(INTEL_DEBUG & DEBUG_VS)) {
if (shader) {
printf("Native code for vertex shader %d:\n", prog->Name);
} else {
printf("Native code for vertex program %d:\n", c->vp->program.Base.Id);
}
}
foreach_list(node, instructions) {
vec4_instruction *inst = (vec4_instruction *)node;
struct brw_reg src[3], dst;
if (unlikely(INTEL_DEBUG & DEBUG_VS)) {
if (last_annotation_ir != inst->ir) {
last_annotation_ir = inst->ir;
if (last_annotation_ir) {
printf(" ");
if (shader) {
((ir_instruction *) last_annotation_ir)->print();
} else {
const prog_instruction *vpi;
vpi = (const prog_instruction *) inst->ir;
printf("%d: ", (int)(vpi - vp->Base.Instructions));
_mesa_fprint_instruction_opt(stdout, vpi, 0,
PROG_PRINT_DEBUG, NULL);
}
printf("\n");
}
}
if (last_annotation_string != inst->annotation) {
last_annotation_string = inst->annotation;
if (last_annotation_string)
printf(" %s\n", last_annotation_string);
}
}
for (unsigned int i = 0; i < 3; i++) {
src[i] = inst->get_src(i);
}
dst = inst->get_dst();
brw_set_conditionalmod(p, inst->conditional_mod);
brw_set_predicate_control(p, inst->predicate);
brw_set_predicate_inverse(p, inst->predicate_inverse);
brw_set_saturate(p, inst->saturate);
switch (inst->opcode) {
case BRW_OPCODE_MOV:
brw_MOV(p, dst, src[0]);
break;
case BRW_OPCODE_ADD:
brw_ADD(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_MUL:
brw_MUL(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_MACH:
brw_set_acc_write_control(p, 1);
brw_MACH(p, dst, src[0], src[1]);
brw_set_acc_write_control(p, 0);
break;
case BRW_OPCODE_FRC:
brw_FRC(p, dst, src[0]);
break;
case BRW_OPCODE_RNDD:
brw_RNDD(p, dst, src[0]);
break;
case BRW_OPCODE_RNDE:
brw_RNDE(p, dst, src[0]);
break;
case BRW_OPCODE_RNDZ:
brw_RNDZ(p, dst, src[0]);
break;
case BRW_OPCODE_AND:
brw_AND(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_OR:
brw_OR(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_XOR:
brw_XOR(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_NOT:
brw_NOT(p, dst, src[0]);
break;
case BRW_OPCODE_ASR:
brw_ASR(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_SHR:
brw_SHR(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_SHL:
brw_SHL(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_CMP:
brw_CMP(p, dst, inst->conditional_mod, src[0], src[1]);
break;
case BRW_OPCODE_SEL:
brw_SEL(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_DPH:
brw_DPH(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_DP4:
brw_DP4(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_DP3:
brw_DP3(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_DP2:
brw_DP2(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_IF:
if (inst->src[0].file != BAD_FILE) {
/* The instruction has an embedded compare (only allowed on gen6) */
assert(intel->gen == 6);
gen6_IF(p, inst->conditional_mod, src[0], src[1]);
} else {
struct brw_instruction *brw_inst = brw_IF(p, BRW_EXECUTE_8);
brw_inst->header.predicate_control = inst->predicate;
}
break;
case BRW_OPCODE_ELSE:
brw_ELSE(p);
break;
case BRW_OPCODE_ENDIF:
brw_ENDIF(p);
break;
case BRW_OPCODE_DO:
brw_DO(p, BRW_EXECUTE_8);
break;
case BRW_OPCODE_BREAK:
brw_BREAK(p);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
break;
case BRW_OPCODE_CONTINUE:
/* FINISHME: We need to write the loop instruction support still. */
if (intel->gen >= 6)
gen6_CONT(p);
else
brw_CONT(p);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
break;
case BRW_OPCODE_WHILE:
brw_WHILE(p);
break;
default:
generate_vs_instruction(inst, dst, src);
break;
}
if (unlikely(INTEL_DEBUG & DEBUG_VS)) {
brw_dump_compile(p, stdout,
last_native_insn_offset, p->next_insn_offset);
}
last_native_insn_offset = p->next_insn_offset;
}