/*
* Copyright © 2012 Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
* IN THE SOFTWARE.
*/
/** @file brw_fs_fp.cpp
*
* Implementation of the compiler for GL_ARB_fragment_program shaders on top
* of the GLSL compiler backend.
*/
#include "brw_context.h"
#include "brw_fs.h"
void
fs_visitor::emit_fp_alu1(enum opcode opcode,
const struct prog_instruction *fpi,
fs_reg dst, fs_reg src)
{
for (int i = 0; i < 4; i++) {
if (fpi->DstReg.WriteMask & (1 << i))
emit(opcode, offset(dst, i), offset(src, i));
}
}
void
fs_visitor::emit_fp_alu2(enum opcode opcode,
const struct prog_instruction *fpi,
fs_reg dst, fs_reg src0, fs_reg src1)
{
for (int i = 0; i < 4; i++) {
if (fpi->DstReg.WriteMask & (1 << i))
emit(opcode, offset(dst, i),
offset(src0, i), offset(src1, i));
}
}
void
fs_visitor::emit_fp_minmax(const prog_instruction *fpi,
fs_reg dst, fs_reg src0, fs_reg src1)
{
enum brw_conditional_mod conditionalmod;
if (fpi->Opcode == OPCODE_MIN)
conditionalmod = BRW_CONDITIONAL_L;
else
conditionalmod = BRW_CONDITIONAL_GE;
for (int i = 0; i < 4; i++) {
if (fpi->DstReg.WriteMask & (1 << i)) {
emit_minmax(conditionalmod, offset(dst, i),
offset(src0, i), offset(src1, i));
}
}
}
void
fs_visitor::emit_fp_sop(enum brw_conditional_mod conditional_mod,
const struct prog_instruction *fpi,
fs_reg dst, fs_reg src0, fs_reg src1,
fs_reg one)
{
for (int i = 0; i < 4; i++) {
if (fpi->DstReg.WriteMask & (1 << i)) {
fs_inst *inst;
emit(CMP(reg_null_d, offset(src0, i), offset(src1, i),
conditional_mod));
inst = emit(BRW_OPCODE_SEL, offset(dst, i), one, fs_reg(0.0f));
inst->predicate = BRW_PREDICATE_NORMAL;
}
}
}
void
fs_visitor::emit_fp_scalar_write(const struct prog_instruction *fpi,
fs_reg dst, fs_reg src)
{
for (int i = 0; i < 4; i++) {
if (fpi->DstReg.WriteMask & (1 << i))
emit(MOV(offset(dst, i), src));
}
}
void
fs_visitor::emit_fp_scalar_math(enum opcode opcode,
const struct prog_instruction *fpi,
fs_reg dst, fs_reg src)
{
fs_reg temp = vgrf(glsl_type::float_type);
emit_math(opcode, temp, src);
emit_fp_scalar_write(fpi, dst, temp);
}
void
fs_visitor::emit_fragment_program_code()
{
setup_fp_regs();
/* Keep a reg with 1.0 around, for reuse by emit_fp_sop so that it can just
* be:
*
* sel.f0 dst 1.0 0.0
*
* instead of
*
* mov dst 0.0
* mov.f0 dst 1.0
*/
fs_reg one = vgrf(glsl_type::float_type);
emit(MOV(one, fs_reg(1.0f)));
for (unsigned int insn = 0; insn < prog->NumInstructions; insn++) {
const struct prog_instruction *fpi = &prog->Instructions[insn];
base_ir = fpi;
fs_reg dst;
fs_reg src[3];
/* We always emit into a temporary destination register to avoid
* aliasing issues.
*/
dst = vgrf(glsl_type::vec4_type);
for (int i = 0; i < 3; i++)
src[i] = get_fp_src_reg(&fpi->SrcReg[i]);
switch (fpi->Opcode) {
case OPCODE_ABS:
src[0].abs = true;
src[0].negate = false;
emit_fp_alu1(BRW_OPCODE_MOV, fpi, dst, src[0]);
break;
case OPCODE_ADD:
emit_fp_alu2(BRW_OPCODE_ADD, fpi, dst, src[0], src[1]);
break;
case OPCODE_CMP:
for (int i = 0; i < 4; i++) {
if (fpi->DstReg.WriteMask & (1 << i)) {
fs_inst *inst;
emit(CMP(reg_null_f, offset(src[0], i), fs_reg(0.0f),
BRW_CONDITIONAL_L));
inst = emit(BRW_OPCODE_SEL, offset(dst, i),
offset(src[1], i), offset(src[2], i));
inst->predicate = BRW_PREDICATE_NORMAL;
}
}
break;
case OPCODE_COS:
emit_fp_scalar_math(SHADER_OPCODE_COS, fpi, dst, src[0]);
break;
case OPCODE_DP2:
case OPCODE_DP3:
case OPCODE_DP4:
case OPCODE_DPH: {
fs_reg mul = vgrf(glsl_type::float_type);
fs_reg acc = vgrf(glsl_type::float_type);
int count;
switch (fpi->Opcode) {
case OPCODE_DP2: count = 2; break;
case OPCODE_DP3: count = 3; break;
case OPCODE_DP4: count = 4; break;
case OPCODE_DPH: count = 3; break;
default: unreachable("not reached");
}
emit(MUL(acc, offset(src[0], 0), offset(src[1], 0)));
for (int i = 1; i < count; i++) {
emit(MUL(mul, offset(src[0], i), offset(src[1], i)));
emit(ADD(acc, acc, mul));
}
if (fpi->Opcode == OPCODE_DPH)
emit(ADD(acc, acc, offset(src[1], 3)));
emit_fp_scalar_write(fpi, dst, acc);
break;
}
case OPCODE_DST:
if (fpi->DstReg.WriteMask & WRITEMASK_X)
emit(MOV(dst, fs_reg(1.0f)));
if (fpi->DstReg.WriteMask & WRITEMASK_Y) {
emit(MUL(offset(dst, 1),
offset(src[0], 1), offset(src[1], 1)));
}
if (fpi->DstReg.WriteMask & WRITEMASK_Z)
emit(MOV(offset(dst, 2), offset(src[0], 2)));
if (fpi->DstReg.WriteMask & WRITEMASK_W)
emit(MOV(offset(dst, 3), offset(src[1], 3)));
break;
case OPCODE_EX2:
emit_fp_scalar_math(SHADER_OPCODE_EXP2, fpi, dst, src[0]);
break;
case OPCODE_FLR:
emit_fp_alu1(BRW_OPCODE_RNDD, fpi, dst, src[0]);
break;
case OPCODE_FRC:
emit_fp_alu1(BRW_OPCODE_FRC, fpi, dst, src[0]);
break;
case OPCODE_KIL: {
for (int i = 0; i < 4; i++) {
/* In most cases the argument to a KIL will be something like
* TEMP[0].wwww, so there's no point in checking whether .w is < 0
* 4 times in a row.
*/
if (i > 0 &&
GET_SWZ(fpi->SrcReg[0].Swizzle, i) ==
GET_SWZ(fpi->SrcReg[0].Swizzle, i - 1) &&
((fpi->SrcReg[0].Negate >> i) & 1) ==
((fpi->SrcReg[0].Negate >> (i - 1)) & 1)) {
continue;
}
/* Emit an instruction that's predicated on the current
* undiscarded pixels, and updates just those pixels to be
* turned off.
*/
fs_inst *cmp = emit(CMP(reg_null_f, offset(src[0], i),
fs_reg(0.0f), BRW_CONDITIONAL_GE));
cmp->predicate = BRW_PREDICATE_NORMAL;
cmp->flag_subreg = 1;
if (devinfo->gen >= 6)
emit_discard_jump();
}
break;
}
case OPCODE_LG2:
emit_fp_scalar_math(SHADER_OPCODE_LOG2, fpi, dst, src[0]);
break;
case OPCODE_LIT:
/* From the ARB_fragment_program spec:
*
* tmp = VectorLoad(op0);
* if (tmp.x < 0) tmp.x = 0;
* if (tmp.y < 0) tmp.y = 0;
* if (tmp.w < -(128.0-epsilon)) tmp.w = -(128.0-epsilon);
* else if (tmp.w > 128-epsilon) tmp.w = 128-epsilon;
* result.x = 1.0;
* result.y = tmp.x;
* result.z = (tmp.x > 0) ? RoughApproxPower(tmp.y, tmp.w) : 0.0;
* result.w = 1.0;
*
* Note that we don't do the clamping to +/- 128. We didn't in
* brw_wm_emit.c either.
*/
if (fpi->DstReg.WriteMask & WRITEMASK_X)
emit(MOV(offset(dst, 0), fs_reg(1.0f)));
if (fpi->DstReg.WriteMask & WRITEMASK_YZ) {
fs_inst *inst;
emit(CMP(reg_null_f, offset(src[0], 0), fs_reg(0.0f),
BRW_CONDITIONAL_LE));
if (fpi->DstReg.WriteMask & WRITEMASK_Y) {
emit(MOV(offset(dst, 1), offset(src[0], 0)));
inst = emit(MOV(offset(dst, 1), fs_reg(0.0f)));
inst->predicate = BRW_PREDICATE_NORMAL;
}
if (fpi->DstReg.WriteMask & WRITEMASK_Z) {
emit_math(SHADER_OPCODE_POW, offset(dst, 2),
offset(src[0], 1), offset(src[0], 3));
inst = emit(MOV(offset(dst, 2), fs_reg(0.0f)));
inst->predicate = BRW_PREDICATE_NORMAL;
}
}
if (fpi->DstReg.WriteMask & WRITEMASK_W)
emit(MOV(offset(dst, 3), fs_reg(1.0f)));
break;
case OPCODE_LRP:
for (int i = 0; i < 4; i++) {
if (fpi->DstReg.WriteMask & (1 << i)) {
fs_reg a = offset(src[0], i);
fs_reg y = offset(src[1], i);
fs_reg x = offset(src[2], i);
emit_lrp(offset(dst, i), x, y, a);
}
}
break;
case OPCODE_MAD:
for (int i = 0; i < 4; i++) {
if (fpi->DstReg.WriteMask & (1 << i)) {
if (devinfo->gen >= 6) {
emit(MAD(offset(dst, i), offset(src[2], i),
offset(src[1], i), offset(src[0], i)));
} else {
fs_reg temp = vgrf(glsl_type::float_type);
emit(MUL(temp, offset(src[0], i), offset(src[1], i)));
emit(ADD(offset(dst, i), temp, offset(src[2], i)));
}
}
}
break;
case OPCODE_MAX:
emit_fp_minmax(fpi, dst, src[0], src[1]);
break;
case OPCODE_MOV:
emit_fp_alu1(BRW_OPCODE_MOV, fpi, dst, src[0]);
break;
case OPCODE_MIN:
emit_fp_minmax(fpi, dst, src[0], src[1]);
break;
case OPCODE_MUL:
emit_fp_alu2(BRW_OPCODE_MUL, fpi, dst, src[0], src[1]);
break;
case OPCODE_POW: {
fs_reg temp = vgrf(glsl_type::float_type);
emit_math(SHADER_OPCODE_POW, temp, src[0], src[1]);
emit_fp_scalar_write(fpi, dst, temp);
break;
}
case OPCODE_RCP:
emit_fp_scalar_math(SHADER_OPCODE_RCP, fpi, dst, src[0]);
break;
case OPCODE_RSQ:
emit_fp_scalar_math(SHADER_OPCODE_RSQ, fpi, dst, src[0]);
break;
case OPCODE_SCS:
if (fpi->DstReg.WriteMask & WRITEMASK_X) {
emit_math(SHADER_OPCODE_COS, offset(dst, 0),
offset(src[0], 0));
}
if (fpi->DstReg.WriteMask & WRITEMASK_Y) {
emit_math(SHADER_OPCODE_SIN, offset(dst, 1),
offset(src[0], 1));
}
break;
case OPCODE_SGE:
emit_fp_sop(BRW_CONDITIONAL_GE, fpi, dst, src[0], src[1], one);
break;
case OPCODE_SIN:
emit_fp_scalar_math(SHADER_OPCODE_SIN, fpi, dst, src[0]);
break;
case OPCODE_SLT:
emit_fp_sop(BRW_CONDITIONAL_L, fpi, dst, src[0], src[1], one);
break;
case OPCODE_SUB: {
fs_reg neg_src1 = src[1];
neg_src1.negate = !src[1].negate;
emit_fp_alu2(BRW_OPCODE_ADD, fpi, dst, src[0], neg_src1);
break;
}
case OPCODE_TEX:
case OPCODE_TXB:
case OPCODE_TXP: {
ir_texture_opcode op;
fs_reg lod;
fs_reg dpdy;
fs_reg coordinate = src[0];
fs_reg shadow_c;
fs_reg sample_index;
fs_reg texel_offset; /* No offsets; leave as BAD_FILE. */
switch (fpi->Opcode) {
case OPCODE_TEX:
op = ir_tex;
break;
case OPCODE_TXP: {
op = ir_tex;
coordinate = vgrf(glsl_type::vec3_type);
fs_reg invproj = vgrf(glsl_type::float_type);
emit_math(SHADER_OPCODE_RCP, invproj, offset(src[0], 3));
for (int i = 0; i < 3; i++) {
emit(MUL(offset(coordinate, i),
offset(src[0], i), invproj));
}
break;
}
case OPCODE_TXB:
op = ir_txb;
lod = offset(src[0], 3);
break;
default:
unreachable("not reached");
}
int coord_components;
switch (fpi->TexSrcTarget) {
case TEXTURE_1D_INDEX:
coord_components = 1;
break;
case TEXTURE_2D_INDEX:
case TEXTURE_1D_ARRAY_INDEX:
case TEXTURE_RECT_INDEX:
case TEXTURE_EXTERNAL_INDEX:
coord_components = 2;
break;
case TEXTURE_3D_INDEX:
case TEXTURE_2D_ARRAY_INDEX:
coord_components = 3;
break;
case TEXTURE_CUBE_INDEX: {
coord_components = 3;
fs_reg temp = vgrf(glsl_type::float_type);
fs_reg cubecoord = vgrf(glsl_type::vec3_type);
fs_reg abscoord = coordinate;
abscoord.negate = false;
abscoord.abs = true;
emit_minmax(BRW_CONDITIONAL_GE, temp,
offset(abscoord, 0), offset(abscoord, 1));
emit_minmax(BRW_CONDITIONAL_GE, temp,
temp, offset(abscoord, 2));
emit_math(SHADER_OPCODE_RCP, temp, temp);
for (int i = 0; i < 3; i++) {
emit(MUL(offset(cubecoord, i),
offset(coordinate, i), temp));
}
coordinate = cubecoord;
break;
}
default:
unreachable("not reached");
}
if (fpi->TexShadow)
shadow_c = offset(coordinate, 2);
emit_texture(op, glsl_type::vec4_type, coordinate, coord_components,
shadow_c, lod, dpdy, 0, sample_index,
reg_undef, /* offset */
reg_undef, /* mcs */
0, /* gather component */
false, /* is cube array */
fpi->TexSrcTarget == TEXTURE_RECT_INDEX,
fpi->TexSrcUnit, fs_reg(fpi->TexSrcUnit),
fpi->TexSrcUnit);
dst = this->result;
break;
}
case OPCODE_SWZ:
/* Note that SWZ's extended swizzles are handled in the general
* get_src_reg() code.
*/
emit_fp_alu1(BRW_OPCODE_MOV, fpi, dst, src[0]);
break;
case OPCODE_XPD:
for (int i = 0; i < 3; i++) {
if (fpi->DstReg.WriteMask & (1 << i)) {
int i1 = (i + 1) % 3;
int i2 = (i + 2) % 3;
fs_reg temp = vgrf(glsl_type::float_type);
fs_reg neg_src1_1 = offset(src[1], i1);
neg_src1_1.negate = !neg_src1_1.negate;
emit(MUL(temp, offset(src[0], i2), neg_src1_1));
emit(MUL(offset(dst, i),
offset(src[0], i1), offset(src[1], i2)));
emit(ADD(offset(dst, i), offset(dst, i), temp));
}
}
break;
case OPCODE_END:
break;
default:
_mesa_problem(ctx, "Unsupported opcode %s in fragment program\n",
_mesa_opcode_string(fpi->Opcode));
}
/* To handle saturates, we emit a MOV with a saturate bit, which
* optimization should fold into the preceding instructions when safe.
*/
if (_mesa_num_inst_dst_regs(fpi->Opcode) != 0) {
fs_reg real_dst = get_fp_dst_reg(&fpi->DstReg);
for (int i = 0; i < 4; i++) {
if (fpi->DstReg.WriteMask & (1 << i)) {
fs_inst *inst = emit(MOV(offset(real_dst, i),
offset(dst, i)));
inst->saturate = fpi->SaturateMode;
}
}
}
}
/* Epilogue:
*
* Fragment depth has this strange convention of being the .z component of
* a vec4. emit_fb_write() wants to see a float value, instead.
*/
this->current_annotation = "result.depth write";
if (frag_depth.file != BAD_FILE) {
fs_reg temp = vgrf(glsl_type::float_type);
emit(MOV(temp, offset(frag_depth, 2)));
frag_depth = temp;
}
}
void
fs_visitor::setup_fp_regs()
{
/* PROGRAM_TEMPORARY */
int num_temp = prog->NumTemporaries;
fp_temp_regs = rzalloc_array(mem_ctx, fs_reg, num_temp);
for (int i = 0; i < num_temp; i++)
fp_temp_regs[i] = vgrf(glsl_type::vec4_type);
/* PROGRAM_STATE_VAR etc. */
if (dispatch_width == 8) {
for (unsigned p = 0;
p < prog->Parameters->NumParameters; p++) {
for (unsigned int i = 0; i < 4; i++) {
stage_prog_data->param[uniforms++] =
&prog->Parameters->ParameterValues[p][i];
}
}
}
fp_input_regs = rzalloc_array(mem_ctx, fs_reg, VARYING_SLOT_MAX);
for (int i = 0; i < VARYING_SLOT_MAX; i++) {
if (prog->InputsRead & BITFIELD64_BIT(i)) {
this->current_annotation = ralloc_asprintf(ctx, "interpolate input %d",
i);
switch (i) {
case VARYING_SLOT_POS:
{
assert(stage == MESA_SHADER_FRAGMENT);
gl_fragment_program *fp = (gl_fragment_program*) prog;
fp_input_regs[i] =
*emit_fragcoord_interpolation(fp->PixelCenterInteger,
fp->OriginUpperLeft);
}
break;
case VARYING_SLOT_FACE:
fp_input_regs[i] = *emit_frontfacing_interpolation();
break;
default:
fp_input_regs[i] = vgrf(glsl_type::vec4_type);
emit_general_interpolation(fp_input_regs[i], "fp_input",
glsl_type::vec4_type,
INTERP_QUALIFIER_NONE,
i, false, false);
if (i == VARYING_SLOT_FOGC) {
emit(MOV(offset(fp_input_regs[i], 1), fs_reg(0.0f)));
emit(MOV(offset(fp_input_regs[i], 2), fs_reg(0.0f)));
emit(MOV(offset(fp_input_regs[i], 3), fs_reg(1.0f)));
}
break;
}
this->current_annotation = NULL;
}
}
}
fs_reg
fs_visitor::get_fp_dst_reg(const prog_dst_register *dst)
{
assert(stage == MESA_SHADER_FRAGMENT);
brw_wm_prog_key *key = (brw_wm_prog_key*) this->key;
switch (dst->File) {
case PROGRAM_TEMPORARY:
return fp_temp_regs[dst->Index];
case PROGRAM_OUTPUT:
if (dst->Index == FRAG_RESULT_DEPTH) {
if (frag_depth.file == BAD_FILE)
frag_depth = vgrf(glsl_type::vec4_type);
return frag_depth;
} else if (dst->Index == FRAG_RESULT_COLOR) {
if (outputs[0].file == BAD_FILE) {
outputs[0] = vgrf(glsl_type::vec4_type);
output_components[0] = 4;
/* Tell emit_fb_writes() to smear fragment.color across all the
* color attachments.
*/
for (int i = 1; i < key->nr_color_regions; i++) {
outputs[i] = outputs[0];
output_components[i] = output_components[0];
}
}
return outputs[0];
} else {
int output_index = dst->Index - FRAG_RESULT_DATA0;
if (outputs[output_index].file == BAD_FILE) {
outputs[output_index] = vgrf(glsl_type::vec4_type);
}
output_components[output_index] = 4;
return outputs[output_index];
}
case PROGRAM_UNDEFINED:
return fs_reg();
default:
_mesa_problem(ctx, "bad dst register file: %s\n",
_mesa_register_file_name((gl_register_file)dst->File));
return vgrf(glsl_type::vec4_type);
}
}
fs_reg
fs_visitor::get_fp_src_reg(const prog_src_register *src)
{
struct gl_program_parameter_list *plist = prog->Parameters;
fs_reg result;
assert(!src->Abs);
switch (src->File) {
case PROGRAM_UNDEFINED:
return fs_reg();
case PROGRAM_TEMPORARY:
result = fp_temp_regs[src->Index];
break;
case PROGRAM_INPUT:
result = fp_input_regs[src->Index];
break;
case PROGRAM_STATE_VAR:
case PROGRAM_UNIFORM:
case PROGRAM_CONSTANT:
/* We actually want to look at the type in the Parameters list for this,
* because this lets us upload constant builtin uniforms, as actual
* constants.
*/
switch (plist->Parameters[src->Index].Type) {
case PROGRAM_CONSTANT: {
result = vgrf(glsl_type::vec4_type);
for (int i = 0; i < 4; i++) {
emit(MOV(offset(result, i),
fs_reg(plist->ParameterValues[src->Index][i].f)));
}
break;
}
case PROGRAM_STATE_VAR:
case PROGRAM_UNIFORM:
result = fs_reg(UNIFORM, src->Index * 4);
break;
default:
_mesa_problem(ctx, "bad uniform src register file: %s\n",
_mesa_register_file_name((gl_register_file)src->File));
return vgrf(glsl_type::vec4_type);
}
break;
default:
_mesa_problem(ctx, "bad src register file: %s\n",
_mesa_register_file_name((gl_register_file)src->File));
return vgrf(glsl_type::vec4_type);
}
if (src->Swizzle != SWIZZLE_NOOP || src->Negate) {
fs_reg unswizzled = result;
result = vgrf(glsl_type::vec4_type);
for (int i = 0; i < 4; i++) {
bool negate = src->Negate & (1 << i);
/* The ZERO, ONE, and Negate options are only used for OPCODE_SWZ,
* but it costs us nothing to support it.
*/
int src_swiz = GET_SWZ(src->Swizzle, i);
if (src_swiz == SWIZZLE_ZERO) {
emit(MOV(offset(result, i), fs_reg(0.0f)));
} else if (src_swiz == SWIZZLE_ONE) {
emit(MOV(offset(result, i),
negate ? fs_reg(-1.0f) : fs_reg(1.0f)));
} else {
fs_reg src = offset(unswizzled, src_swiz);
if (negate)
src.negate = !src.negate;
emit(MOV(offset(result, i), src));
}
}
}
return result;
}