static void brw_clip_test( struct brw_clip_compile *c )
{
struct brw_reg t = retype(get_tmp(c), BRW_REGISTER_TYPE_UD);
struct brw_reg t1 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD);
struct brw_reg t2 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD);
struct brw_reg t3 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD);
struct brw_reg v0 = get_tmp(c);
struct brw_reg v1 = get_tmp(c);
struct brw_reg v2 = get_tmp(c);
struct brw_indirect vt0 = brw_indirect(0, 0);
struct brw_indirect vt1 = brw_indirect(1, 0);
struct brw_indirect vt2 = brw_indirect(2, 0);
struct brw_compile *p = &c->func;
struct brw_instruction *is_outside;
struct brw_reg tmp0 = c->reg.loopcount; /* handy temporary */
brw_MOV(p, get_addr_reg(vt0), brw_address(c->reg.vertex[0]));
brw_MOV(p, get_addr_reg(vt1), brw_address(c->reg.vertex[1]));
brw_MOV(p, get_addr_reg(vt2), brw_address(c->reg.vertex[2]));
brw_MOV(p, v0, deref_4f(vt0, c->offset[VERT_RESULT_HPOS]));
brw_MOV(p, v1, deref_4f(vt1, c->offset[VERT_RESULT_HPOS]));
brw_MOV(p, v2, deref_4f(vt2, c->offset[VERT_RESULT_HPOS]));
brw_AND(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(~0x3f));
/* test nearz, xmin, ymin plane */
/* clip.xyz < -clip.w */
brw_CMP(p, t1, BRW_CONDITIONAL_L, v0, negate(get_element(v0, 3)));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_CMP(p, t2, BRW_CONDITIONAL_L, v1, negate(get_element(v1, 3)));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_CMP(p, t3, BRW_CONDITIONAL_L, v2, negate(get_element(v2, 3)));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
/* All vertices are outside of a plane, rejected */
brw_AND(p, t, t1, t2);
brw_AND(p, t, t, t3);
brw_OR(p, tmp0, get_element(t, 0), get_element(t, 1));
brw_OR(p, tmp0, tmp0, get_element(t, 2));
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
brw_AND(p, brw_null_reg(), tmp0, brw_imm_ud(0x1));
is_outside = brw_IF(p, BRW_EXECUTE_1);
{
brw_clip_kill_thread(c);
}
brw_ENDIF(p, is_outside);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
/* some vertices are inside a plane, some are outside,need to clip */
brw_XOR(p, t, t1, t2);
brw_XOR(p, t1, t2, t3);
brw_OR(p, t, t, t1);
brw_AND(p, t, t, brw_imm_ud(0x1));
brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
get_element(t, 0), brw_imm_ud(0));
brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<5)));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
get_element(t, 1), brw_imm_ud(0));
brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<3)));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
get_element(t, 2), brw_imm_ud(0));
brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<1)));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
/* test farz, xmax, ymax plane */
/* clip.xyz > clip.w */
brw_CMP(p, t1, BRW_CONDITIONAL_G, v0, get_element(v0, 3));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_CMP(p, t2, BRW_CONDITIONAL_G, v1, get_element(v1, 3));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_CMP(p, t3, BRW_CONDITIONAL_G, v2, get_element(v2, 3));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
/* All vertices are outside of a plane, rejected */
brw_AND(p, t, t1, t2);
brw_AND(p, t, t, t3);
brw_OR(p, tmp0, get_element(t, 0), get_element(t, 1));
brw_OR(p, tmp0, tmp0, get_element(t, 2));
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
brw_AND(p, brw_null_reg(), tmp0, brw_imm_ud(0x1));
is_outside = brw_IF(p, BRW_EXECUTE_1);
{
brw_clip_kill_thread(c);
}
brw_ENDIF(p, is_outside);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
/* some vertices are inside a plane, some are outside,need to clip */
brw_XOR(p, t, t1, t2);
brw_XOR(p, t1, t2, t3);
brw_OR(p, t, t, t1);
brw_AND(p, t, t, brw_imm_ud(0x1));
brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
get_element(t, 0), brw_imm_ud(0));
brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<4)));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
get_element(t, 1), brw_imm_ud(0));
brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<2)));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
get_element(t, 2), brw_imm_ud(0));
brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<0)));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
release_tmps(c);
}
void
gen8_vec4_generator::generate_vec4_instruction(vec4_instruction *instruction,
struct brw_reg dst,
struct brw_reg *src)
{
vec4_instruction *ir = (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(ir->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 (ir->opcode) {
case BRW_OPCODE_MOV:
MOV(dst, src[0]);
break;
case BRW_OPCODE_ADD:
ADD(dst, src[0], src[1]);
break;
case BRW_OPCODE_MUL:
MUL(dst, src[0], src[1]);
break;
case BRW_OPCODE_MACH:
MACH(dst, src[0], src[1]);
break;
case BRW_OPCODE_MAD:
MAD(dst, src[0], src[1], src[2]);
break;
case BRW_OPCODE_FRC:
FRC(dst, src[0]);
break;
case BRW_OPCODE_RNDD:
RNDD(dst, src[0]);
break;
case BRW_OPCODE_RNDE:
RNDE(dst, src[0]);
break;
case BRW_OPCODE_RNDZ:
RNDZ(dst, src[0]);
break;
case BRW_OPCODE_AND:
AND(dst, src[0], src[1]);
break;
case BRW_OPCODE_OR:
OR(dst, src[0], src[1]);
break;
case BRW_OPCODE_XOR:
XOR(dst, src[0], src[1]);
break;
case BRW_OPCODE_NOT:
NOT(dst, src[0]);
break;
case BRW_OPCODE_ASR:
ASR(dst, src[0], src[1]);
break;
case BRW_OPCODE_SHR:
SHR(dst, src[0], src[1]);
break;
case BRW_OPCODE_SHL:
SHL(dst, src[0], src[1]);
break;
case BRW_OPCODE_CMP:
CMP(dst, ir->conditional_mod, src[0], src[1]);
break;
case BRW_OPCODE_SEL:
SEL(dst, src[0], src[1]);
break;
case BRW_OPCODE_DPH:
DPH(dst, src[0], src[1]);
break;
case BRW_OPCODE_DP4:
DP4(dst, src[0], src[1]);
break;
case BRW_OPCODE_DP3:
DP3(dst, src[0], src[1]);
break;
case BRW_OPCODE_DP2:
DP2(dst, src[0], src[1]);
break;
case BRW_OPCODE_F32TO16:
/* Emulate the Gen7 zeroing bug. */
MOV(retype(dst, BRW_REGISTER_TYPE_UD), brw_imm_ud(0u));
MOV(retype(dst, BRW_REGISTER_TYPE_HF), src[0]);
break;
case BRW_OPCODE_F16TO32:
MOV(dst, retype(src[0], BRW_REGISTER_TYPE_HF));
break;
case BRW_OPCODE_LRP:
LRP(dst, src[0], src[1], src[2]);
break;
case BRW_OPCODE_BFREV:
/* BFREV only supports UD type for src and dst. */
BFREV(retype(dst, BRW_REGISTER_TYPE_UD),
retype(src[0], BRW_REGISTER_TYPE_UD));
break;
case BRW_OPCODE_FBH:
/* FBH only supports UD type for dst. */
FBH(retype(dst, BRW_REGISTER_TYPE_UD), src[0]);
break;
case BRW_OPCODE_FBL:
/* FBL only supports UD type for dst. */
FBL(retype(dst, BRW_REGISTER_TYPE_UD), src[0]);
break;
case BRW_OPCODE_CBIT:
/* CBIT only supports UD type for dst. */
CBIT(retype(dst, BRW_REGISTER_TYPE_UD), src[0]);
break;
case BRW_OPCODE_ADDC:
ADDC(dst, src[0], src[1]);
break;
case BRW_OPCODE_SUBB:
SUBB(dst, src[0], src[1]);
break;
case BRW_OPCODE_BFE:
BFE(dst, src[0], src[1], src[2]);
break;
case BRW_OPCODE_BFI1:
BFI1(dst, src[0], src[1]);
break;
case BRW_OPCODE_BFI2:
BFI2(dst, src[0], src[1], src[2]);
break;
case BRW_OPCODE_IF:
IF(ir->predicate);
break;
case BRW_OPCODE_ELSE:
ELSE();
break;
case BRW_OPCODE_ENDIF:
ENDIF();
break;
case BRW_OPCODE_DO:
DO();
break;
case BRW_OPCODE_BREAK:
BREAK();
break;
case BRW_OPCODE_CONTINUE:
CONTINUE();
break;
case BRW_OPCODE_WHILE:
WHILE();
break;
case SHADER_OPCODE_RCP:
MATH(BRW_MATH_FUNCTION_INV, dst, src[0]);
break;
case SHADER_OPCODE_RSQ:
MATH(BRW_MATH_FUNCTION_RSQ, dst, src[0]);
break;
case SHADER_OPCODE_SQRT:
MATH(BRW_MATH_FUNCTION_SQRT, dst, src[0]);
break;
case SHADER_OPCODE_EXP2:
MATH(BRW_MATH_FUNCTION_EXP, dst, src[0]);
break;
case SHADER_OPCODE_LOG2:
MATH(BRW_MATH_FUNCTION_LOG, dst, src[0]);
break;
case SHADER_OPCODE_SIN:
MATH(BRW_MATH_FUNCTION_SIN, dst, src[0]);
break;
case SHADER_OPCODE_COS:
MATH(BRW_MATH_FUNCTION_COS, dst, src[0]);
break;
case SHADER_OPCODE_POW:
MATH(BRW_MATH_FUNCTION_POW, dst, src[0], src[1]);
break;
case SHADER_OPCODE_INT_QUOTIENT:
MATH(BRW_MATH_FUNCTION_INT_DIV_QUOTIENT, dst, src[0], src[1]);
break;
case SHADER_OPCODE_INT_REMAINDER:
MATH(BRW_MATH_FUNCTION_INT_DIV_REMAINDER, 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(ir, dst);
break;
case VS_OPCODE_URB_WRITE:
generate_urb_write(ir, true);
break;
case SHADER_OPCODE_GEN4_SCRATCH_READ:
generate_scratch_read(ir, dst, src[0]);
break;
case SHADER_OPCODE_GEN4_SCRATCH_WRITE:
generate_scratch_write(ir, dst, src[0], src[1]);
break;
case VS_OPCODE_PULL_CONSTANT_LOAD:
case VS_OPCODE_PULL_CONSTANT_LOAD_GEN7:
generate_pull_constant_load(ir, dst, src[0], src[1]);
break;
case GS_OPCODE_URB_WRITE:
generate_urb_write(ir, false);
break;
case GS_OPCODE_THREAD_END:
generate_gs_thread_end(ir);
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 SHADER_OPCODE_SHADER_TIME_ADD:
assert(!"XXX: Missing Gen8 vec4 support for INTEL_DEBUG=shader_time");
break;
case SHADER_OPCODE_UNTYPED_ATOMIC:
assert(!"XXX: Missing Gen8 vec4 support for UNTYPED_ATOMIC");
break;
case SHADER_OPCODE_UNTYPED_SURFACE_READ:
assert(!"XXX: Missing Gen8 vec4 support for UNTYPED_SURFACE_READ");
break;
case VS_OPCODE_UNPACK_FLAGS_SIMD4X2:
assert(!"VS_OPCODE_UNPACK_FLAGS_SIMD4X2 should not be used on Gen8+.");
break;
default:
if (ir->opcode < (int) ARRAY_SIZE(opcode_descs)) {
_mesa_problem(ctx, "Unsupported opcode in `%s' in VS\n",
opcode_descs[ir->opcode].name);
} else {
_mesa_problem(ctx, "Unsupported opcode %d in VS", ir->opcode);
}
abort();
}
}
/* Post-fragment-program processing. Send the results to the
* framebuffer.
*/
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)
{
struct brw_compile *p = &c->func;
GLuint nr = 2;
GLuint channel;
/* Reserve a space for AA - may not be needed:
*/
if (c->key.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);
for (channel = 0; channel < 4; channel++) {
/* 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->key.source_depth_to_render_target)
{
if (c->key.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->key.dest_depth_reg)
{
GLuint comp = c->key.dest_depth_reg / 2;
GLuint off = c->key.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 (!c->key.runtime_check_aads_emit) {
if (c->key.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);
}
}
/* 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[VERT_RESULT_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[VERT_RESULT_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[VERT_RESULT_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);
}
void brw_emit_point_sprite_setup(struct brw_sf_compile *c, bool allocate)
{
struct brw_compile *p = &c->func;
GLuint i;
c->nr_verts = 1;
if (allocate)
alloc_regs(c);
copy_z_inv_w(c);
for (i = 0; i < c->nr_setup_regs; i++)
{
struct brw_reg a0 = offset(c->vert[0], i);
GLushort pc, pc_persp, pc_linear, pc_coord_replace;
bool last = calculate_masks(c, i, &pc, &pc_persp, &pc_linear);
pc_coord_replace = calculate_point_sprite_mask(c, i);
pc_persp &= ~pc_coord_replace;
if (pc_persp) {
brw_set_predicate_control_flag_value(p, pc_persp);
brw_MUL(p, a0, a0, c->inv_w[0]);
}
/* Point sprite coordinate replacement: A texcoord with this
* enabled gets replaced with the value (x, y, 0, 1) where x and
* y vary from 0 to 1 across the horizontal and vertical of the
* point.
*/
if (pc_coord_replace) {
brw_set_predicate_control_flag_value(p, pc_coord_replace);
/* Caculate 1.0/PointWidth */
brw_math(&c->func,
c->tmp,
BRW_MATH_FUNCTION_INV,
BRW_MATH_SATURATE_NONE,
0,
c->dx0,
BRW_MATH_DATA_SCALAR,
BRW_MATH_PRECISION_FULL);
brw_set_access_mode(p, BRW_ALIGN_16);
/* dA/dx, dA/dy */
brw_MOV(p, c->m1Cx, brw_imm_f(0.0));
brw_MOV(p, c->m2Cy, brw_imm_f(0.0));
brw_MOV(p, brw_writemask(c->m1Cx, WRITEMASK_X), c->tmp);
if (c->key.sprite_origin_lower_left) {
brw_MOV(p, brw_writemask(c->m2Cy, WRITEMASK_Y), negate(c->tmp));
} else {
brw_MOV(p, brw_writemask(c->m2Cy, WRITEMASK_Y), c->tmp);
}
/* attribute constant offset */
brw_MOV(p, c->m3C0, brw_imm_f(0.0));
if (c->key.sprite_origin_lower_left) {
brw_MOV(p, brw_writemask(c->m3C0, WRITEMASK_YW), brw_imm_f(1.0));
} else {
brw_MOV(p, brw_writemask(c->m3C0, WRITEMASK_W), brw_imm_f(1.0));
}
brw_set_access_mode(p, BRW_ALIGN_1);
}
if (pc & ~pc_coord_replace) {
brw_set_predicate_control_flag_value(p, pc & ~pc_coord_replace);
brw_MOV(p, c->m1Cx, brw_imm_ud(0));
brw_MOV(p, c->m2Cy, brw_imm_ud(0));
brw_MOV(p, c->m3C0, a0); /* constant value */
}
brw_set_predicate_control_flag_value(p, pc);
/* Copy m0..m3 to URB. */
brw_urb_WRITE(p,
brw_null_reg(),
0,
brw_vec8_grf(0, 0),
0, /* allocate */
1, /* used */
4, /* msg len */
0, /* response len */
last, /* eot */
last, /* writes complete */
i*4, /* urb destination offset */
BRW_URB_SWIZZLE_TRANSPOSE);
}
}
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;
int 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)) - p->store;
{
saveflag = p->flag_value;
brw_push_insn_state(p);
brw_emit_tri_setup( c, 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)) - p->store;
{
saveflag = p->flag_value;
brw_push_insn_state(p);
brw_emit_line_setup( c, 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)) - p->store;
{
saveflag = p->flag_value;
brw_push_insn_state(p);
brw_emit_point_sprite_setup( c, false );
brw_pop_insn_state(p);
p->flag_value = saveflag;
}
brw_land_fwd_jump(p, jmp);
brw_emit_point_setup( c, false );
}
void brw_emit_unfilled_clip( struct brw_clip_compile *c )
{
struct brw_compile *p = &c->func;
struct brw_instruction *do_clip;
c->need_direction = ((c->key.offset_ccw || c->key.offset_cw) ||
(c->key.fill_ccw != c->key.fill_cw) ||
c->key.fill_ccw == CLIP_CULL ||
c->key.fill_cw == CLIP_CULL ||
c->key.copy_bfc_cw ||
c->key.copy_bfc_ccw);
brw_clip_tri_alloc_regs(c, 3 + c->key.nr_userclip + 6);
brw_clip_tri_init_vertices(c);
assert(c->offset[VERT_RESULT_EDGE]);
if (c->key.fill_ccw == CLIP_CULL &&
c->key.fill_cw == CLIP_CULL) {
brw_clip_kill_thread(c);
return;
}
merge_edgeflags(c);
/* Need to use the inlist indirection here:
*/
if (c->need_direction)
compute_tri_direction(c);
if (c->key.fill_ccw == CLIP_CULL ||
c->key.fill_cw == CLIP_CULL)
cull_direction(c);
if (c->key.offset_ccw ||
c->key.offset_cw)
compute_offset(c);
if (c->key.copy_bfc_ccw ||
c->key.copy_bfc_cw)
copy_bfc(c);
/* Need to do this whether we clip or not:
*/
if (c->key.do_flat_shading)
brw_clip_tri_flat_shade(c);
brw_clip_init_clipmask(c);
brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_NZ, c->reg.planemask, brw_imm_ud(0));
do_clip = brw_IF(p, BRW_EXECUTE_1);
{
brw_clip_init_planes(c);
brw_clip_tri(c);
check_nr_verts(c);
}
brw_ENDIF(p, do_clip);
emit_unfilled_primitives(c);
brw_clip_kill_thread(c);
}
void
vec4_gs_visitor::gs_emit_vertex(int stream_id)
{
this->current_annotation = "emit vertex: safety check";
/* Haswell and later hardware ignores the "Render Stream Select" bits
* from the 3DSTATE_STREAMOUT packet when the SOL stage is disabled,
* and instead sends all primitives down the pipeline for rasterization.
* If the SOL stage is enabled, "Render Stream Select" is honored and
* primitives bound to non-zero streams are discarded after stream output.
*
* Since the only purpose of primives sent to non-zero streams is to
* be recorded by transform feedback, we can simply discard all geometry
* bound to these streams when transform feedback is disabled.
*/
if (stream_id > 0 && !nir->info.has_transform_feedback_varyings)
return;
/* If we're outputting 32 control data bits or less, then we can wait
* until the shader is over to output them all. Otherwise we need to
* output them as we go. Now is the time to do it, since we're about to
* output the vertex_count'th vertex, so it's guaranteed that the
* control data bits associated with the (vertex_count - 1)th vertex are
* correct.
*/
if (c->control_data_header_size_bits > 32) {
this->current_annotation = "emit vertex: emit control data bits";
/* Only emit control data bits if we've finished accumulating a batch
* of 32 bits. This is the case when:
*
* (vertex_count * bits_per_vertex) % 32 == 0
*
* (in other words, when the last 5 bits of vertex_count *
* bits_per_vertex are 0). Assuming bits_per_vertex == 2^n for some
* integer n (which is always the case, since bits_per_vertex is
* always 1 or 2), this is equivalent to requiring that the last 5-n
* bits of vertex_count are 0:
*
* vertex_count & (2^(5-n) - 1) == 0
*
* 2^(5-n) == 2^5 / 2^n == 32 / bits_per_vertex, so this is
* equivalent to:
*
* vertex_count & (32 / bits_per_vertex - 1) == 0
*/
vec4_instruction *inst =
emit(AND(dst_null_ud(), this->vertex_count,
brw_imm_ud(32 / c->control_data_bits_per_vertex - 1)));
inst->conditional_mod = BRW_CONDITIONAL_Z;
emit(IF(BRW_PREDICATE_NORMAL));
{
/* If vertex_count is 0, then no control data bits have been
* accumulated yet, so we skip emitting them.
*/
emit(CMP(dst_null_ud(), this->vertex_count, brw_imm_ud(0u),
BRW_CONDITIONAL_NEQ));
emit(IF(BRW_PREDICATE_NORMAL));
emit_control_data_bits();
emit(BRW_OPCODE_ENDIF);
/* Reset control_data_bits to 0 so we can start accumulating a new
* batch.
*
* Note: in the case where vertex_count == 0, this neutralizes the
* effect of any call to EndPrimitive() that the shader may have
* made before outputting its first vertex.
*/
inst = emit(MOV(dst_reg(this->control_data_bits), brw_imm_ud(0u)));
inst->force_writemask_all = true;
}
emit(BRW_OPCODE_ENDIF);
}
this->current_annotation = "emit vertex: vertex data";
emit_vertex();
/* In stream mode we have to set control data bits for all vertices
* unless we have disabled control data bits completely (which we do
* do for GL_POINTS outputs that don't use streams).
*/
if (c->control_data_header_size_bits > 0 &&
gs_prog_data->control_data_format ==
GEN7_GS_CONTROL_DATA_FORMAT_GSCTL_SID) {
this->current_annotation = "emit vertex: Stream control data bits";
set_stream_control_data_bits(stream_id);
}
this->current_annotation = NULL;
}
/**
* Write out a batch of 32 control data bits from the control_data_bits
* register to the URB.
*
* The current value of the vertex_count register determines which DWORD in
* the URB receives the control data bits. The control_data_bits register is
* assumed to contain the correct data for the vertex that was most recently
* output, and all previous vertices that share the same DWORD.
*
* This function takes care of ensuring that if no vertices have been output
* yet, no control bits are emitted.
*/
void
vec4_gs_visitor::emit_control_data_bits()
{
assert(c->control_data_bits_per_vertex != 0);
/* Since the URB_WRITE_OWORD message operates with 128-bit (vec4 sized)
* granularity, we need to use two tricks to ensure that the batch of 32
* control data bits is written to the appropriate DWORD in the URB. To
* select which vec4 we are writing to, we use the "slot {0,1} offset"
* fields of the message header. To select which DWORD in the vec4 we are
* writing to, we use the channel mask fields of the message header. To
* avoid penalizing geometry shaders that emit a small number of vertices
* with extra bookkeeping, we only do each of these tricks when
* c->prog_data.control_data_header_size_bits is large enough to make it
* necessary.
*
* Note: this means that if we're outputting just a single DWORD of control
* data bits, we'll actually replicate it four times since we won't do any
* channel masking. But that's not a problem since in this case the
* hardware only pays attention to the first DWORD.
*/
enum brw_urb_write_flags urb_write_flags = BRW_URB_WRITE_OWORD;
if (c->control_data_header_size_bits > 32)
urb_write_flags = urb_write_flags | BRW_URB_WRITE_USE_CHANNEL_MASKS;
if (c->control_data_header_size_bits > 128)
urb_write_flags = urb_write_flags | BRW_URB_WRITE_PER_SLOT_OFFSET;
/* If we are using either channel masks or a per-slot offset, then we
* need to figure out which DWORD we are trying to write to, using the
* formula:
*
* dword_index = (vertex_count - 1) * bits_per_vertex / 32
*
* Since bits_per_vertex is a power of two, and is known at compile
* time, this can be optimized to:
*
* dword_index = (vertex_count - 1) >> (6 - log2(bits_per_vertex))
*/
src_reg dword_index(this, glsl_type::uint_type);
if (urb_write_flags) {
src_reg prev_count(this, glsl_type::uint_type);
emit(ADD(dst_reg(prev_count), this->vertex_count,
brw_imm_ud(0xffffffffu)));
unsigned log2_bits_per_vertex =
_mesa_fls(c->control_data_bits_per_vertex);
emit(SHR(dst_reg(dword_index), prev_count,
brw_imm_ud(6 - log2_bits_per_vertex)));
}
/* Start building the URB write message. The first MRF gets a copy of
* R0.
*/
int base_mrf = 1;
dst_reg mrf_reg(MRF, base_mrf);
src_reg r0(retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UD));
vec4_instruction *inst = emit(MOV(mrf_reg, r0));
inst->force_writemask_all = true;
if (urb_write_flags & BRW_URB_WRITE_PER_SLOT_OFFSET) {
/* Set the per-slot offset to dword_index / 4, to that we'll write to
* the appropriate OWORD within the control data header.
*/
src_reg per_slot_offset(this, glsl_type::uint_type);
emit(SHR(dst_reg(per_slot_offset), dword_index, brw_imm_ud(2u)));
emit(GS_OPCODE_SET_WRITE_OFFSET, mrf_reg, per_slot_offset,
brw_imm_ud(1u));
}
if (urb_write_flags & BRW_URB_WRITE_USE_CHANNEL_MASKS) {
/* Set the channel masks to 1 << (dword_index % 4), so that we'll
* write to the appropriate DWORD within the OWORD. We need to do
* this computation with force_writemask_all, otherwise garbage data
* from invocation 0 might clobber the mask for invocation 1 when
* GS_OPCODE_PREPARE_CHANNEL_MASKS tries to OR the two masks
* together.
*/
src_reg channel(this, glsl_type::uint_type);
inst = emit(AND(dst_reg(channel), dword_index, brw_imm_ud(3u)));
inst->force_writemask_all = true;
src_reg one(this, glsl_type::uint_type);
inst = emit(MOV(dst_reg(one), brw_imm_ud(1u)));
inst->force_writemask_all = true;
src_reg channel_mask(this, glsl_type::uint_type);
inst = emit(SHL(dst_reg(channel_mask), one, channel));
inst->force_writemask_all = true;
emit(GS_OPCODE_PREPARE_CHANNEL_MASKS, dst_reg(channel_mask),
channel_mask);
emit(GS_OPCODE_SET_CHANNEL_MASKS, mrf_reg, channel_mask);
}
/* Store the control data bits in the message payload and send it. */
dst_reg mrf_reg2(MRF, base_mrf + 1);
inst = emit(MOV(mrf_reg2, this->control_data_bits));
inst->force_writemask_all = true;
inst = emit(GS_OPCODE_URB_WRITE);
inst->urb_write_flags = urb_write_flags;
/* We need to increment Global Offset by 256-bits to make room for
* Broadwell's extra "Vertex Count" payload at the beginning of the
* URB entry. Since this is an OWord message, Global Offset is counted
* in 128-bit units, so we must set it to 2.
*/
if (devinfo->gen >= 8 && gs_prog_data->static_vertex_count == -1)
inst->offset = 2;
inst->base_mrf = base_mrf;
inst->mlen = 2;
}
void brw_SAMPLE(struct brw_compile *p,
struct brw_reg dest,
GLuint msg_reg_nr,
struct brw_reg src0,
GLuint binding_table_index,
GLuint sampler,
GLuint writemask,
GLuint msg_type,
GLuint response_length,
GLuint msg_length,
GLboolean eot)
{
GLboolean need_stall = 0;
if(writemask == 0) {
/* _mesa_printf("%s: zero writemask??\n", __FUNCTION__); */
return;
}
/* Hardware doesn't do destination dependency checking on send
* instructions properly. Add a workaround which generates the
* dependency by other means. In practice it seems like this bug
* only crops up for texture samples, and only where registers are
* written by the send and then written again later without being
* read in between. Luckily for us, we already track that
* information and use it to modify the writemask for the
* instruction, so that is a guide for whether a workaround is
* needed.
*/
if (writemask != WRITEMASK_XYZW) {
GLuint dst_offset = 0;
GLuint i, newmask = 0, len = 0;
for (i = 0; i < 4; i++) {
if (writemask & (1<<i))
break;
dst_offset += 2;
}
for (; i < 4; i++) {
if (!(writemask & (1<<i)))
break;
newmask |= 1<<i;
len++;
}
if (newmask != writemask) {
need_stall = 1;
/* _mesa_printf("need stall %x %x\n", newmask , writemask); */
}
else {
struct brw_reg m1 = brw_message_reg(msg_reg_nr);
newmask = ~newmask & WRITEMASK_XYZW;
brw_push_insn_state(p);
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_set_mask_control(p, BRW_MASK_DISABLE);
brw_MOV(p, m1, brw_vec8_grf(0,0));
brw_MOV(p, get_element_ud(m1, 2), brw_imm_ud(newmask << 12));
brw_pop_insn_state(p);
src0 = retype(brw_null_reg(), BRW_REGISTER_TYPE_UW);
dest = offset(dest, dst_offset);
response_length = len * 2;
}
}
{
struct brw_instruction *insn = next_insn(p, BRW_OPCODE_SEND);
insn->header.predicate_control = 0; /* XXX */
insn->header.compression_control = BRW_COMPRESSION_NONE;
insn->header.destreg__conditonalmod = msg_reg_nr;
brw_set_dest(insn, dest);
brw_set_src0(insn, src0);
brw_set_sampler_message(p->brw, insn,
binding_table_index,
sampler,
msg_type,
response_length,
msg_length,
eot);
}
if (need_stall)
{
struct brw_reg reg = vec8(offset(dest, response_length-1));
/* mov (8) r9.0<1>:f r9.0<8;8,1>:f { Align1 }
*/
brw_push_insn_state(p);
brw_set_compression_control(p, GL_FALSE);
brw_MOV(p, reg, reg);
brw_pop_insn_state(p);
}
}
void
vec4_tcs_visitor::nir_emit_intrinsic(nir_intrinsic_instr *instr)
{
switch (instr->intrinsic) {
case nir_intrinsic_load_invocation_id:
emit(MOV(get_nir_dest(instr->dest, BRW_REGISTER_TYPE_UD),
invocation_id));
break;
case nir_intrinsic_load_primitive_id:
emit(TCS_OPCODE_GET_PRIMITIVE_ID,
get_nir_dest(instr->dest, BRW_REGISTER_TYPE_UD));
break;
case nir_intrinsic_load_patch_vertices_in:
emit(MOV(get_nir_dest(instr->dest, BRW_REGISTER_TYPE_D),
brw_imm_d(key->input_vertices)));
break;
case nir_intrinsic_load_per_vertex_input: {
src_reg indirect_offset = get_indirect_offset(instr);
unsigned imm_offset = instr->const_index[0];
nir_const_value *vertex_const = nir_src_as_const_value(instr->src[0]);
src_reg vertex_index =
vertex_const ? src_reg(brw_imm_ud(vertex_const->u32[0]))
: get_nir_src(instr->src[0], BRW_REGISTER_TYPE_UD, 1);
dst_reg dst = get_nir_dest(instr->dest, BRW_REGISTER_TYPE_D);
dst.writemask = brw_writemask_for_size(instr->num_components);
emit_input_urb_read(dst, vertex_index, imm_offset, indirect_offset);
break;
}
case nir_intrinsic_load_input:
unreachable("nir_lower_io should use load_per_vertex_input intrinsics");
break;
case nir_intrinsic_load_output:
case nir_intrinsic_load_per_vertex_output: {
src_reg indirect_offset = get_indirect_offset(instr);
unsigned imm_offset = instr->const_index[0];;
dst_reg dst = get_nir_dest(instr->dest, BRW_REGISTER_TYPE_D);
dst.writemask = brw_writemask_for_size(instr->num_components);
if (imm_offset == 0 && indirect_offset.file == BAD_FILE) {
dst.type = BRW_REGISTER_TYPE_F;
/* This is a read of gl_TessLevelInner[], which lives in the
* Patch URB header. The layout depends on the domain.
*/
switch (key->tes_primitive_mode) {
case GL_QUADS: {
/* DWords 3-2 (reversed); use offset 0 and WZYX swizzle. */
dst_reg tmp(this, glsl_type::vec4_type);
emit_output_urb_read(tmp, 0, src_reg());
emit(MOV(writemask(dst, WRITEMASK_XY),
swizzle(src_reg(tmp), BRW_SWIZZLE_WZYX)));
break;
}
case GL_TRIANGLES:
/* DWord 4; use offset 1 but normal swizzle/writemask. */
emit_output_urb_read(writemask(dst, WRITEMASK_X), 1, src_reg());
break;
case GL_ISOLINES:
/* All channels are undefined. */
return;
default:
unreachable("Bogus tessellation domain");
}
} else if (imm_offset == 1 && indirect_offset.file == BAD_FILE) {
dst.type = BRW_REGISTER_TYPE_F;
unsigned swiz = BRW_SWIZZLE_WZYX;
/* This is a read of gl_TessLevelOuter[], which lives in the
* high 4 DWords of the Patch URB header, in reverse order.
*/
switch (key->tes_primitive_mode) {
case GL_QUADS:
dst.writemask = WRITEMASK_XYZW;
break;
case GL_TRIANGLES:
dst.writemask = WRITEMASK_XYZ;
break;
case GL_ISOLINES:
/* Isolines are not reversed; swizzle .zw -> .xy */
swiz = BRW_SWIZZLE_ZWZW;
dst.writemask = WRITEMASK_XY;
return;
default:
unreachable("Bogus tessellation domain");
}
dst_reg tmp(this, glsl_type::vec4_type);
emit_output_urb_read(tmp, 1, src_reg());
emit(MOV(dst, swizzle(src_reg(tmp), swiz)));
} else {
emit_output_urb_read(dst, imm_offset, indirect_offset);
}
break;
}
case nir_intrinsic_store_output:
case nir_intrinsic_store_per_vertex_output: {
src_reg value = get_nir_src(instr->src[0]);
unsigned mask = instr->const_index[1];
unsigned swiz = BRW_SWIZZLE_XYZW;
src_reg indirect_offset = get_indirect_offset(instr);
unsigned imm_offset = instr->const_index[0];
if (imm_offset == 0 && indirect_offset.file == BAD_FILE) {
value.type = BRW_REGISTER_TYPE_F;
mask &= (1 << tesslevel_inner_components(key->tes_primitive_mode)) - 1;
/* This is a write to gl_TessLevelInner[], which lives in the
* Patch URB header. The layout depends on the domain.
*/
switch (key->tes_primitive_mode) {
case GL_QUADS:
/* gl_TessLevelInner[].xy lives at DWords 3-2 (reversed).
* We use an XXYX swizzle to reverse put .xy in the .wz
* channels, and use a .zw writemask.
*/
swiz = BRW_SWIZZLE4(0, 0, 1, 0);
mask = writemask_for_backwards_vector(mask);
break;
case GL_TRIANGLES:
/* gl_TessLevelInner[].x lives at DWord 4, so we set the
* writemask to X and bump the URB offset by 1.
*/
imm_offset = 1;
break;
case GL_ISOLINES:
/* Skip; gl_TessLevelInner[] doesn't exist for isolines. */
return;
default:
unreachable("Bogus tessellation domain");
}
} else if (imm_offset == 1 && indirect_offset.file == BAD_FILE) {
value.type = BRW_REGISTER_TYPE_F;
mask &= (1 << tesslevel_outer_components(key->tes_primitive_mode)) - 1;
/* This is a write to gl_TessLevelOuter[] which lives in the
* Patch URB Header at DWords 4-7. However, it's reversed, so
* instead of .xyzw we have .wzyx.
*/
if (key->tes_primitive_mode == GL_ISOLINES) {
/* Isolines .xy should be stored in .zw, in order. */
swiz = BRW_SWIZZLE4(0, 0, 0, 1);
mask <<= 2;
} else {
/* Other domains are reversed; store .wzyx instead of .xyzw. */
swiz = BRW_SWIZZLE_WZYX;
mask = writemask_for_backwards_vector(mask);
}
}
emit_urb_write(swizzle(value, swiz), mask,
imm_offset, indirect_offset);
break;
}
case nir_intrinsic_barrier: {
dst_reg header = dst_reg(this, glsl_type::uvec4_type);
emit(TCS_OPCODE_CREATE_BARRIER_HEADER, header);
emit(SHADER_OPCODE_BARRIER, dst_null_ud(), src_reg(header));
break;
}
default:
vec4_visitor::nir_emit_intrinsic(instr);
}
}
