void brw_clip_ff_sync(struct brw_clip_compile *c)
{
if (c->need_ff_sync) {
struct brw_compile *p = &c->func;
struct brw_instruction *need_ff_sync;
brw_set_conditionalmod(p, BRW_CONDITIONAL_Z);
brw_AND(p, brw_null_reg(), c->reg.ff_sync, brw_imm_ud(0x1));
need_ff_sync = brw_IF(p, BRW_EXECUTE_1);
{
brw_OR(p, c->reg.ff_sync, c->reg.ff_sync, brw_imm_ud(0x1));
brw_ff_sync(p,
c->reg.R0,
0,
c->reg.R0,
1,
1, /* used */
1, /* msg length */
1, /* response length */
0, /* eot */
1, /* write compelete */
0, /* urb offset */
BRW_URB_SWIZZLE_NONE);
}
brw_ENDIF(p, need_ff_sync);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
}
}
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 brw_clip_ff_sync(struct brw_clip_compile *c)
{
struct brw_codegen *p = &c->func;
if (p->devinfo->gen == 5) {
brw_AND(p, brw_null_reg(), c->reg.ff_sync, brw_imm_ud(0x1));
brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_Z);
brw_IF(p, BRW_EXECUTE_1);
{
brw_OR(p, c->reg.ff_sync, c->reg.ff_sync, brw_imm_ud(0x1));
brw_ff_sync(p,
c->reg.R0,
0,
c->reg.R0,
1, /* allocate */
1, /* response length */
0 /* eot */);
}
brw_ENDIF(p);
brw_set_default_predicate_control(p, BRW_PREDICATE_NONE);
}
}
void brw_clip_ff_sync(struct brw_clip_compile *c)
{
struct brw_compile *p = &c->func;
struct brw_context *brw = p->brw;
if (brw->gen == 5) {
brw_AND(p, brw_null_reg(), c->reg.ff_sync, brw_imm_ud(0x1));
brw_last_inst->header.destreg__conditionalmod = BRW_CONDITIONAL_Z;
brw_IF(p, BRW_EXECUTE_1);
{
brw_OR(p, c->reg.ff_sync, c->reg.ff_sync, brw_imm_ud(0x1));
brw_ff_sync(p,
c->reg.R0,
0,
c->reg.R0,
1, /* allocate */
1, /* response length */
0 /* eot */);
}
brw_ENDIF(p);
brw_set_default_predicate_control(p, BRW_PREDICATE_NONE);
}
}
void brw_clip_ff_sync(struct brw_clip_compile *c)
{
struct intel_context *intel = &c->func.brw->intel;
if (intel->needs_ff_sync) {
struct brw_compile *p = &c->func;
brw_set_conditionalmod(p, BRW_CONDITIONAL_Z);
brw_AND(p, brw_null_reg(), c->reg.ff_sync, brw_imm_ud(0x1));
brw_IF(p, BRW_EXECUTE_1);
{
brw_OR(p, c->reg.ff_sync, c->reg.ff_sync, brw_imm_ud(0x1));
brw_ff_sync(p,
c->reg.R0,
0,
c->reg.R0,
1, /* allocate */
1, /* response length */
0 /* eot */);
}
brw_ENDIF(p);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
}
}
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_hpos));
brw_MOV(p, v1, deref_4f(vt1, c->offset_hpos));
brw_MOV(p, v2, deref_4f(vt2, c->offset_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);
}
/**
* 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();
}
}
void
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);
brw_push_insn_state(p);
brw_set_access_mode(p, BRW_ALIGN_1);
brw_set_mask_control(p, BRW_MASK_DISABLE);
brw_OR(p, suboffset(vec1(dst), 21), vec1(src), suboffset(vec1(src), 16));
brw_pop_insn_state(p);
}
/* 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);
}
void
vec4_generator::generate_tex(vec4_instruction *inst,
struct brw_reg dst,
struct brw_reg src,
struct brw_reg sampler_index)
{
int msg_type = -1;
if (brw->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:
if (inst->shadow_compare) {
/* Gen7.5+. Otherwise, lowered by brw_lower_texture_gradients(). */
assert(brw->gen >= 8 || brw->is_haswell);
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:
if (brw->gen >= 7)
msg_type = GEN7_SAMPLER_MESSAGE_SAMPLE_LD2DMS;
else
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE_LD;
break;
case SHADER_OPCODE_TXF_MCS:
assert(brw->gen >= 7);
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 (inst->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 (inst->shadow_compare) {
msg_type = GEN7_SAMPLER_MESSAGE_SAMPLE_GATHER4_PO_C;
} else {
msg_type = GEN7_SAMPLER_MESSAGE_SAMPLE_GATHER4_PO;
}
break;
default:
unreachable("should not get here: invalid vec4 texture opcode");
}
} 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:
unreachable("should not get here: invalid vec4 texture opcode");
}
}
assert(msg_type != -1);
assert(sampler_index.type == BRW_REGISTER_TYPE_UD);
/* 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->header_present) {
if (brw->gen < 6 && !inst->texture_offset) {
/* Set up an implied move from g0 to the MRF. */
src = brw_vec8_grf(0, 0);
} else {
struct brw_reg header =
retype(brw_message_reg(inst->base_mrf), BRW_REGISTER_TYPE_UD);
/* Explicitly set up the message header by copying g0 to the MRF. */
brw_push_insn_state(p);
brw_set_default_mask_control(p, BRW_MASK_DISABLE);
brw_MOV(p, header, retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UD));
brw_set_default_access_mode(p, BRW_ALIGN_1);
if (inst->texture_offset) {
/* Set the texel offset bits in DWord 2. */
brw_MOV(p, get_element_ud(header, 2),
brw_imm_ud(inst->texture_offset));
}
brw_adjust_sampler_state_pointer(p, header, sampler_index, dst);
brw_pop_insn_state(p);
}
}
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;
}
uint32_t base_binding_table_index = (inst->opcode == SHADER_OPCODE_TG4 ||
inst->opcode == SHADER_OPCODE_TG4_OFFSET)
? prog_data->base.binding_table.gather_texture_start
: prog_data->base.binding_table.texture_start;
if (sampler_index.file == BRW_IMMEDIATE_VALUE) {
uint32_t sampler = sampler_index.dw1.ud;
brw_SAMPLE(p,
dst,
inst->base_mrf,
src,
sampler + base_binding_table_index,
sampler % 16,
msg_type,
1, /* response length */
inst->mlen,
inst->header_present,
BRW_SAMPLER_SIMD_MODE_SIMD4X2,
return_format);
brw_mark_surface_used(&prog_data->base, sampler + base_binding_table_index);
} else {
/* Non-constant sampler index. */
/* Note: this clobbers `dst` as a temporary before emitting the send */
struct brw_reg addr = vec1(retype(brw_address_reg(0), BRW_REGISTER_TYPE_UD));
struct brw_reg temp = vec1(retype(dst, BRW_REGISTER_TYPE_UD));
struct brw_reg sampler_reg = vec1(retype(sampler_index, BRW_REGISTER_TYPE_UD));
brw_push_insn_state(p);
brw_set_default_mask_control(p, BRW_MASK_DISABLE);
brw_set_default_access_mode(p, BRW_ALIGN_1);
/* Some care required: `sampler` and `temp` may alias:
* addr = sampler & 0xff
* temp = (sampler << 8) & 0xf00
* addr = addr | temp
*/
brw_ADD(p, addr, sampler_reg, brw_imm_ud(base_binding_table_index));
brw_SHL(p, temp, sampler_reg, brw_imm_ud(8u));
brw_AND(p, temp, temp, brw_imm_ud(0x0f00));
brw_AND(p, addr, addr, brw_imm_ud(0x0ff));
brw_OR(p, addr, addr, temp);
/* a0.0 |= <descriptor> */
brw_inst *insn_or = brw_next_insn(p, BRW_OPCODE_OR);
brw_set_sampler_message(p, insn_or,
0 /* surface */,
0 /* sampler */,
msg_type,
1 /* rlen */,
inst->mlen /* mlen */,
inst->header_present /* header */,
BRW_SAMPLER_SIMD_MODE_SIMD4X2,
return_format);
brw_inst_set_exec_size(p->brw, insn_or, BRW_EXECUTE_1);
brw_inst_set_src1_reg_type(p->brw, insn_or, BRW_REGISTER_TYPE_UD);
brw_set_src0(p, insn_or, addr);
brw_set_dest(p, insn_or, addr);
/* dst = send(offset, a0.0) */
brw_inst *insn_send = brw_next_insn(p, BRW_OPCODE_SEND);
brw_set_dest(p, insn_send, dst);
brw_set_src0(p, insn_send, src);
brw_set_indirect_send_descriptor(p, insn_send, BRW_SFID_SAMPLER, addr);
brw_pop_insn_state(p);
/* visitor knows more than we do about the surface limit required,
* so has already done marking.
*/
}
}
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;
}