/*
GLfloat iz = 1.0 / dir.z;
GLfloat ac = dir.x * iz;
GLfloat bc = dir.y * iz;
offset = ctx->Polygon.OffsetUnits * DEPTH_SCALE;
offset += MAX2( abs(ac), abs(bc) ) * ctx->Polygon.OffsetFactor;
offset *= MRD;
*/
static void compute_offset( struct brw_clip_compile *c )
{
struct brw_compile *p = &c->func;
struct brw_reg off = c->reg.offset;
struct brw_reg dir = c->reg.dir;
brw_math_invert(p, get_element(off, 2), get_element(dir, 2));
brw_MUL(p, vec2(off), dir, get_element(off, 2));
brw_CMP(p,
vec1(brw_null_reg()),
BRW_CONDITIONAL_GE,
brw_abs(get_element(off, 0)),
brw_abs(get_element(off, 1)));
brw_SEL(p, vec1(off), brw_abs(get_element(off, 0)), brw_abs(get_element(off, 1)));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_MUL(p, vec1(off), off, brw_imm_f(c->key.offset_factor));
brw_ADD(p, vec1(off), off, brw_imm_f(c->key.offset_units));
}
static void emit_mad(struct brw_wm_compile *c,
struct prog_instruction *inst)
{
struct brw_compile *p = &c->func;
GLuint mask = inst->DstReg.WriteMask;
struct brw_reg dst, src0, src1, src2;
int i;
for (i = 0; i < 4; i++) {
if (mask & (1<<i)) {
dst = get_dst_reg(c, inst, i, 1);
src0 = get_src_reg(c, &inst->SrcReg[0], i, 1);
src1 = get_src_reg(c, &inst->SrcReg[1], i, 1);
src2 = get_src_reg(c, &inst->SrcReg[2], i, 1);
brw_MUL(p, dst, src0, src1);
brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0);
brw_ADD(p, dst, dst, src2);
brw_set_saturate(p, 0);
}
}
}
void emit_lrp(struct brw_compile *p,
const struct brw_reg *dst,
GLuint mask,
const struct brw_reg *arg0,
const struct brw_reg *arg1,
const struct brw_reg *arg2)
{
GLuint i;
/* Uses dst as a temporary:
*/
for (i = 0; i < 4; i++) {
if (mask & (1<<i)) {
/* Can I use the LINE instruction for this?
*/
brw_ADD(p, dst[i], negate(arg0[i]), brw_imm_f(1.0));
brw_MUL(p, brw_null_reg(), dst[i], arg2[i]);
brw_set_saturate(p, (mask & SATURATE) ? 1 : 0);
brw_MAC(p, dst[i], arg0[i], arg1[i]);
brw_set_saturate(p, 0);
}
}
}
static void emit_lrp(struct brw_wm_compile *c,
struct prog_instruction *inst)
{
struct brw_compile *p = &c->func;
GLuint mask = inst->DstReg.WriteMask;
struct brw_reg dst, tmp1, tmp2, src0, src1, src2;
int i;
for (i = 0; i < 4; i++) {
if (mask & (1<<i)) {
dst = get_dst_reg(c, inst, i, 1);
src0 = get_src_reg(c, &inst->SrcReg[0], i, 1);
src1 = get_src_reg(c, &inst->SrcReg[1], i, 1);
if (src1.nr == dst.nr) {
tmp1 = alloc_tmp(c);
brw_MOV(p, tmp1, src1);
} else
tmp1 = src1;
src2 = get_src_reg(c, &inst->SrcReg[2], i, 1);
if (src2.nr == dst.nr) {
tmp2 = alloc_tmp(c);
brw_MOV(p, tmp2, src2);
} else
tmp2 = src2;
brw_ADD(p, dst, negate(src0), brw_imm_f(1.0));
brw_MUL(p, brw_null_reg(), dst, tmp2);
brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0);
brw_MAC(p, dst, src0, tmp1);
brw_set_saturate(p, 0);
}
release_tmps(c);
}
}
void
vec4_generator::generate_oword_dual_block_offsets(struct brw_reg m1,
struct brw_reg index)
{
int second_vertex_offset;
if (brw->gen >= 6)
second_vertex_offset = 1;
else
second_vertex_offset = 16;
m1 = retype(m1, BRW_REGISTER_TYPE_D);
/* Set up M1 (message payload). Only the block offsets in M1.0 and
* M1.4 are used, and the rest are ignored.
*/
struct brw_reg m1_0 = suboffset(vec1(m1), 0);
struct brw_reg m1_4 = suboffset(vec1(m1), 4);
struct brw_reg index_0 = suboffset(vec1(index), 0);
struct brw_reg index_4 = suboffset(vec1(index), 4);
brw_push_insn_state(p);
brw_set_mask_control(p, BRW_MASK_DISABLE);
brw_set_access_mode(p, BRW_ALIGN_1);
brw_MOV(p, m1_0, index_0);
if (index.file == BRW_IMMEDIATE_VALUE) {
index_4.dw1.ud += second_vertex_offset;
brw_MOV(p, m1_4, index_4);
} else {
brw_ADD(p, m1_4, index_4, brw_imm_d(second_vertex_offset));
}
brw_pop_insn_state(p);
}
/**
* Loads the clip distance for a vertex into `dst`, and ends with
* a comparison of it to zero with the condition `cond`.
*
* - If using a fixed plane, the distance is dot(hpos, plane).
* - If using a user clip plane, the distance is directly available in the vertex.
*/
static inline void
load_clip_distance(struct brw_clip_compile *c, struct brw_indirect vtx,
struct brw_reg dst, GLuint hpos_offset, int cond)
{
struct brw_codegen *p = &c->func;
dst = vec4(dst);
brw_AND(p, vec1(brw_null_reg()), 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);
{
struct brw_indirect temp_ptr = brw_indirect(7, 0);
brw_ADD(p, get_addr_reg(temp_ptr), get_addr_reg(vtx), c->reg.clipdistance_offset);
brw_MOV(p, vec1(dst), deref_1f(temp_ptr, 0));
}
brw_ELSE(p);
{
brw_MOV(p, dst, deref_4f(vtx, hpos_offset));
brw_DP4(p, dst, dst, c->reg.plane_equation);
}
brw_ENDIF(p);
brw_CMP(p, brw_null_reg(), cond, vec1(dst), brw_imm_f(0.0f));
}
/* 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);
}
/* Interpolate between two vertices and put the result into a0.0.
* Increment a0.0 accordingly.
*
* Beware that dest_ptr can be equal to v0_ptr!
*/
void brw_clip_interp_vertex( struct brw_clip_compile *c,
struct brw_indirect dest_ptr,
struct brw_indirect v0_ptr, /* from */
struct brw_indirect v1_ptr, /* to */
struct brw_reg t0,
bool force_edgeflag)
{
struct brw_codegen *p = &c->func;
struct brw_reg t_nopersp, v0_ndc_copy;
GLuint slot;
/* Just copy the vertex header:
*/
/*
* After CLIP stage, only first 256 bits of the VUE are read
* back on Ironlake, so needn't change it
*/
brw_copy_indirect_to_indirect(p, dest_ptr, v0_ptr, 1);
/* First handle the 3D and NDC interpolation, in case we
* need noperspective interpolation. Doing it early has no
* performance impact in any case.
*/
/* Take a copy of the v0 NDC coordinates, in case dest == v0. */
if (c->has_noperspective_shading) {
GLuint offset = brw_varying_to_offset(&c->vue_map,
BRW_VARYING_SLOT_NDC);
v0_ndc_copy = get_tmp(c);
brw_MOV(p, v0_ndc_copy, deref_4f(v0_ptr, offset));
}
/* Compute the new 3D position
*
* dest_hpos = v0_hpos * (1 - t0) + v1_hpos * t0
*/
{
GLuint delta = brw_varying_to_offset(&c->vue_map, VARYING_SLOT_POS);
struct brw_reg tmp = get_tmp(c);
brw_MUL(p, vec4(brw_null_reg()), deref_4f(v1_ptr, delta), t0);
brw_MAC(p, tmp, negate(deref_4f(v0_ptr, delta)), t0);
brw_ADD(p, deref_4f(dest_ptr, delta), deref_4f(v0_ptr, delta), tmp);
release_tmp(c, tmp);
}
/* Recreate the projected (NDC) coordinate in the new vertex header */
brw_clip_project_vertex(c, dest_ptr);
/* If we have noperspective attributes,
* we need to compute the screen-space t
*/
if (c->has_noperspective_shading) {
GLuint delta = brw_varying_to_offset(&c->vue_map,
BRW_VARYING_SLOT_NDC);
struct brw_reg tmp = get_tmp(c);
t_nopersp = get_tmp(c);
/* t_nopersp = vec4(v1.xy, dest.xy) */
brw_MOV(p, t_nopersp, deref_4f(v1_ptr, delta));
brw_MOV(p, tmp, deref_4f(dest_ptr, delta));
brw_set_default_access_mode(p, BRW_ALIGN_16);
brw_MOV(p,
brw_writemask(t_nopersp, WRITEMASK_ZW),
brw_swizzle(tmp, 0, 1, 0, 1));
/* t_nopersp = vec4(v1.xy, dest.xy) - v0.xyxy */
brw_ADD(p, t_nopersp, t_nopersp,
negate(brw_swizzle(v0_ndc_copy, 0, 1, 0, 1)));
/* Add the absolute values of the X and Y deltas so that if
* the points aren't in the same place on the screen we get
* nonzero values to divide.
*
* After that, we have vert1 - vert0 in t_nopersp.x and
* vertnew - vert0 in t_nopersp.y
*
* t_nopersp = vec2(|v1.x -v0.x| + |v1.y -v0.y|,
* |dest.x-v0.x| + |dest.y-v0.y|)
*/
brw_ADD(p,
brw_writemask(t_nopersp, WRITEMASK_XY),
brw_abs(brw_swizzle(t_nopersp, 0, 2, 0, 0)),
brw_abs(brw_swizzle(t_nopersp, 1, 3, 0, 0)));
brw_set_default_access_mode(p, BRW_ALIGN_1);
/* If the points are in the same place, just substitute a
* value to avoid divide-by-zero
*/
brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_EQ,
vec1(t_nopersp),
brw_imm_f(0));
brw_IF(p, BRW_EXECUTE_1);
brw_MOV(p, t_nopersp, brw_imm_vf4(brw_float_to_vf(1.0),
brw_float_to_vf(0.0),
brw_float_to_vf(0.0),
brw_float_to_vf(0.0)));
brw_ENDIF(p);
/* Now compute t_nopersp = t_nopersp.y/t_nopersp.x and broadcast it. */
brw_math_invert(p, get_element(t_nopersp, 0), get_element(t_nopersp, 0));
brw_MUL(p, vec1(t_nopersp), vec1(t_nopersp),
vec1(suboffset(t_nopersp, 1)));
brw_set_default_access_mode(p, BRW_ALIGN_16);
brw_MOV(p, t_nopersp, brw_swizzle(t_nopersp, 0, 0, 0, 0));
brw_set_default_access_mode(p, BRW_ALIGN_1);
release_tmp(c, tmp);
release_tmp(c, v0_ndc_copy);
}
/* Now we can iterate over each attribute
* (could be done in pairs?)
*/
for (slot = 0; slot < c->vue_map.num_slots; slot++) {
int varying = c->vue_map.slot_to_varying[slot];
GLuint delta = brw_vue_slot_to_offset(slot);
/* HPOS, NDC already handled above */
if (varying == VARYING_SLOT_POS || varying == BRW_VARYING_SLOT_NDC)
continue;
if (varying == VARYING_SLOT_EDGE) {
if (force_edgeflag)
brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(1));
else
brw_MOV(p, deref_4f(dest_ptr, delta), deref_4f(v0_ptr, delta));
} else if (varying == VARYING_SLOT_PSIZ) {
/* PSIZ doesn't need interpolation because it isn't used by the
* fragment shader.
*/
} else if (varying < VARYING_SLOT_MAX) {
/* This is a true vertex result (and not a special value for the VUE
* header), so interpolate:
*
* New = attr0 + t*attr1 - t*attr0
*
* Unless the attribute is flat shaded -- in which case just copy
* from one of the sources (doesn't matter which; already copied from pv)
*/
GLuint interp = c->key.interpolation_mode.mode[slot];
if (interp != INTERP_QUALIFIER_FLAT) {
struct brw_reg tmp = get_tmp(c);
struct brw_reg t =
interp == INTERP_QUALIFIER_NOPERSPECTIVE ? t_nopersp : t0;
brw_MUL(p,
vec4(brw_null_reg()),
deref_4f(v1_ptr, delta),
t);
brw_MAC(p,
tmp,
negate(deref_4f(v0_ptr, delta)),
t);
brw_ADD(p,
deref_4f(dest_ptr, delta),
deref_4f(v0_ptr, delta),
tmp);
release_tmp(c, tmp);
}
else {
brw_MOV(p,
deref_4f(dest_ptr, delta),
deref_4f(v0_ptr, delta));
}
}
}
if (c->vue_map.num_slots % 2) {
GLuint delta = brw_vue_slot_to_offset(c->vue_map.num_slots);
brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(0));
}
if (c->has_noperspective_shading)
release_tmp(c, t_nopersp);
}
/**
* 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_tex(vec4_instruction *inst,
struct brw_reg dst,
struct brw_reg src)
{
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->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:
assert(!"should not get here: invalid vec4 texture opcode");
break;
}
} 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:
assert(!"should not get here: invalid vec4 texture opcode");
break;
}
}
assert(msg_type != -1);
/* 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_mask_control(p, BRW_MASK_DISABLE);
brw_MOV(p, header, retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UD));
brw_set_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));
}
if (inst->sampler >= 16) {
/* The "Sampler Index" field can only store values between 0 and 15.
* However, we can add an offset to the "Sampler State Pointer"
* field, effectively selecting a different set of 16 samplers.
*
* The "Sampler State Pointer" needs to be aligned to a 32-byte
* offset, and each sampler state is only 16-bytes, so we can't
* exclusively use the offset - we have to use both.
*/
assert(brw->is_haswell); /* field only exists on Haswell */
brw_ADD(p,
get_element_ud(header, 3),
get_element_ud(brw_vec8_grf(0, 0), 3),
brw_imm_ud(16 * (inst->sampler / 16) *
sizeof(gen7_sampler_state)));
}
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 surface_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) + inst->sampler;
brw_SAMPLE(p,
dst,
inst->base_mrf,
src,
surface_index,
inst->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, surface_index);
}
/* Interpolate between two vertices and put the result into a0.0.
* Increment a0.0 accordingly.
*/
void brw_clip_interp_vertex( struct brw_clip_compile *c,
struct brw_indirect dest_ptr,
struct brw_indirect v0_ptr, /* from */
struct brw_indirect v1_ptr, /* to */
struct brw_reg t0,
boolean force_edgeflag)
{
#if 0
struct brw_compile *p = &c->func;
struct brw_reg tmp = get_tmp(c);
unsigned i;
/* Just copy the vertex header:
*/
brw_copy_indirect_to_indirect(p, dest_ptr, v0_ptr, 1);
/* Iterate over each attribute (could be done in pairs?)
*/
for (i = 0; i < c->nr_attrs; i++) {
unsigned delta = i*16 + 32;
if (delta == c->offset[VERT_RESULT_EDGE]) {
if (force_edgeflag)
brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(1));
else
brw_MOV(p, deref_4f(dest_ptr, delta), deref_4f(v0_ptr, delta));
}
else {
/* Interpolate:
*
* New = attr0 + t*attr1 - t*attr0
*/
brw_MUL(p,
vec4(brw_null_reg()),
deref_4f(v1_ptr, delta),
t0);
brw_MAC(p,
tmp,
negate(deref_4f(v0_ptr, delta)),
t0);
brw_ADD(p,
deref_4f(dest_ptr, delta),
deref_4f(v0_ptr, delta),
tmp);
}
}
if (i & 1) {
unsigned delta = i*16 + 32;
brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(0));
}
release_tmp(c, tmp);
/* Recreate the projected (NDC) coordinate in the new vertex
* header:
*/
brw_clip_project_vertex(c, dest_ptr );
#else
#warning "disabled"
#endif
}
/**
* Apply an additive offset to DWORD 2 of c->reg.header.
*
* This is used to set/unset the "PrimStart" and "PrimEnd" flags appropriately
* for each vertex.
*/
static void brw_gs_offset_header_dw2(struct brw_gs_compile *c, int offset)
{
struct brw_compile *p = &c->func;
brw_ADD(p, get_element_d(c->reg.header, 2), get_element_d(c->reg.header, 2),
brw_imm_d(offset));
}
void brw_emit_tri_setup(struct brw_sf_compile *c, bool allocate)
{
struct brw_compile *p = &c->func;
GLuint i;
c->flag_value = 0xff;
c->nr_verts = 3;
if (allocate)
alloc_regs(c);
invert_det(c);
copy_z_inv_w(c);
if (c->key.do_twoside_color)
do_twoside_color(c);
if (c->has_flat_shading)
do_flatshade_triangle(c);
for (i = 0; i < c->nr_setup_regs; i++)
{
/* Pair of incoming attributes:
*/
struct brw_reg a0 = offset(c->vert[0], i);
struct brw_reg a1 = offset(c->vert[1], i);
struct brw_reg a2 = offset(c->vert[2], i);
GLushort pc, pc_persp, pc_linear;
bool last = calculate_masks(c, i, &pc, &pc_persp, &pc_linear);
if (pc_persp)
{
set_predicate_control_flag_value(p, c, pc_persp);
brw_MUL(p, a0, a0, c->inv_w[0]);
brw_MUL(p, a1, a1, c->inv_w[1]);
brw_MUL(p, a2, a2, c->inv_w[2]);
}
/* Calculate coefficients for interpolated values:
*/
if (pc_linear)
{
set_predicate_control_flag_value(p, c, pc_linear);
brw_ADD(p, c->a1_sub_a0, a1, negate(a0));
brw_ADD(p, c->a2_sub_a0, a2, negate(a0));
/* calculate dA/dx
*/
brw_MUL(p, brw_null_reg(), c->a1_sub_a0, c->dy2);
brw_MAC(p, c->tmp, c->a2_sub_a0, negate(c->dy0));
brw_MUL(p, c->m1Cx, c->tmp, c->inv_det);
/* calculate dA/dy
*/
brw_MUL(p, brw_null_reg(), c->a2_sub_a0, c->dx0);
brw_MAC(p, c->tmp, c->a1_sub_a0, negate(c->dx2));
brw_MUL(p, c->m2Cy, c->tmp, c->inv_det);
}
{
set_predicate_control_flag_value(p, c, pc);
/* start point for interpolation
*/
brw_MOV(p, c->m3C0, a0);
/* Copy m0..m3 to URB. m0 is implicitly copied from r0 in
* the send instruction:
*/
brw_urb_WRITE(p,
brw_null_reg(),
0,
brw_vec8_grf(0, 0), /* r0, will be copied to m0 */
last ? BRW_URB_WRITE_EOT_COMPLETE
: BRW_URB_WRITE_NO_FLAGS,
4, /* msg len */
0, /* response len */
i*4, /* offset */
BRW_URB_SWIZZLE_TRANSPOSE); /* XXX: Swizzle control "SF to windower" */
}
}
brw_set_default_predicate_control(p, BRW_PREDICATE_NONE);
}
/***********************************************************************
* Output clipped polygon as an unfilled primitive:
*/
static void emit_lines(struct brw_clip_compile *c,
bool do_offset)
{
struct brw_compile *p = &c->func;
const struct brw_context *brw = p->brw;
struct brw_indirect v0 = brw_indirect(0, 0);
struct brw_indirect v1 = brw_indirect(1, 0);
struct brw_indirect v0ptr = brw_indirect(2, 0);
struct brw_indirect v1ptr = brw_indirect(3, 0);
/* Need a seperate loop for offset:
*/
if (do_offset) {
brw_MOV(p, c->reg.loopcount, c->reg.nr_verts);
brw_MOV(p, get_addr_reg(v0ptr), brw_address(c->reg.inlist));
brw_DO(p, BRW_EXECUTE_1);
{
brw_MOV(p, get_addr_reg(v0), deref_1uw(v0ptr, 0));
brw_ADD(p, get_addr_reg(v0ptr), get_addr_reg(v0ptr), brw_imm_uw(2));
apply_one_offset(c, v0);
brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
brw_inst_set_cond_modifier(brw, brw_last_inst, BRW_CONDITIONAL_G);
}
brw_WHILE(p);
brw_inst_set_pred_control(brw, brw_last_inst, BRW_PREDICATE_NORMAL);
}
/* v1ptr = &inlist[nr_verts]
* *v1ptr = v0
*/
brw_MOV(p, c->reg.loopcount, c->reg.nr_verts);
brw_MOV(p, get_addr_reg(v0ptr), brw_address(c->reg.inlist));
brw_ADD(p, get_addr_reg(v1ptr), get_addr_reg(v0ptr), retype(c->reg.nr_verts, BRW_REGISTER_TYPE_UW));
brw_ADD(p, get_addr_reg(v1ptr), get_addr_reg(v1ptr), retype(c->reg.nr_verts, BRW_REGISTER_TYPE_UW));
brw_MOV(p, deref_1uw(v1ptr, 0), deref_1uw(v0ptr, 0));
brw_DO(p, BRW_EXECUTE_1);
{
brw_MOV(p, get_addr_reg(v0), deref_1uw(v0ptr, 0));
brw_MOV(p, get_addr_reg(v1), deref_1uw(v0ptr, 2));
brw_ADD(p, get_addr_reg(v0ptr), get_addr_reg(v0ptr), brw_imm_uw(2));
/* draw edge if edgeflag != 0 */
brw_CMP(p,
vec1(brw_null_reg()), BRW_CONDITIONAL_NZ,
deref_1f(v0, brw_varying_to_offset(&c->vue_map,
VARYING_SLOT_EDGE)),
brw_imm_f(0));
brw_IF(p, BRW_EXECUTE_1);
{
brw_clip_emit_vue(c, v0, BRW_URB_WRITE_ALLOCATE_COMPLETE,
(_3DPRIM_LINESTRIP << URB_WRITE_PRIM_TYPE_SHIFT)
| URB_WRITE_PRIM_START);
brw_clip_emit_vue(c, v1, BRW_URB_WRITE_ALLOCATE_COMPLETE,
(_3DPRIM_LINESTRIP << URB_WRITE_PRIM_TYPE_SHIFT)
| URB_WRITE_PRIM_END);
}
brw_ENDIF(p);
brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
brw_inst_set_cond_modifier(brw, brw_last_inst, BRW_CONDITIONAL_NZ);
}
brw_WHILE(p);
brw_inst_set_pred_control(brw, brw_last_inst, BRW_PREDICATE_NORMAL);
}
/**
* Read float[4] constant(s) from VS constant buffer.
* For relative addressing, two float[4] constants will be read into 'dest'.
* Otherwise, one float[4] constant will be read into the lower half of 'dest'.
*/
void brw_dp_READ_4_vs(struct brw_compile *p,
struct brw_reg dest,
GLuint oword,
GLboolean relAddr,
struct brw_reg addrReg,
GLuint location,
GLuint bind_table_index)
{
GLuint msg_reg_nr = 1;
assert(oword < 2);
/*
printf("vs const read msg, location %u, msg_reg_nr %d\n",
location, msg_reg_nr);
*/
/* Setup MRF[1] with location/offset into const buffer */
{
struct brw_reg b;
brw_push_insn_state(p);
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_set_mask_control(p, BRW_MASK_DISABLE);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
/*brw_set_access_mode(p, BRW_ALIGN_16);*/
/* XXX I think we're setting all the dwords of MRF[1] to 'location'.
* when the docs say only dword[2] should be set. Hmmm. But it works.
*/
b = brw_message_reg(msg_reg_nr);
b = retype(b, BRW_REGISTER_TYPE_UD);
/*b = get_element_ud(b, 2);*/
if (relAddr) {
brw_ADD(p, b, addrReg, brw_imm_ud(location));
}
else {
brw_MOV(p, b, brw_imm_ud(location));
}
brw_pop_insn_state(p);
}
{
struct brw_instruction *insn = next_insn(p, BRW_OPCODE_SEND);
insn->header.predicate_control = BRW_PREDICATE_NONE;
insn->header.compression_control = BRW_COMPRESSION_NONE;
insn->header.destreg__conditionalmod = msg_reg_nr;
insn->header.mask_control = BRW_MASK_DISABLE;
/*insn->header.access_mode = BRW_ALIGN_16;*/
brw_set_dest(insn, dest);
brw_set_src0(insn, brw_null_reg());
brw_set_dp_read_message(p->brw,
insn,
bind_table_index,
oword, /* 0 = lower Oword, 1 = upper Oword */
BRW_DATAPORT_READ_MESSAGE_OWORD_BLOCK_READ, /* msg_type */
0, /* source cache = data cache */
1, /* msg_length */
1, /* response_length (1 Oword) */
0); /* eot */
}
}
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;
}
/* Interpolate between two vertices and put the result into a0.0.
* Increment a0.0 accordingly.
*/
void brw_clip_interp_vertex( struct brw_clip_compile *c,
struct brw_indirect dest_ptr,
struct brw_indirect v0_ptr, /* from */
struct brw_indirect v1_ptr, /* to */
struct brw_reg t0,
bool force_edgeflag)
{
struct brw_compile *p = &c->func;
struct brw_reg tmp = get_tmp(c);
GLuint slot;
/* Just copy the vertex header:
*/
/*
* After CLIP stage, only first 256 bits of the VUE are read
* back on Ironlake, so needn't change it
*/
brw_copy_indirect_to_indirect(p, dest_ptr, v0_ptr, 1);
/* Iterate over each attribute (could be done in pairs?)
*/
for (slot = 0; slot < c->vue_map.num_slots; slot++) {
int varying = c->vue_map.slot_to_varying[slot];
GLuint delta = brw_vue_slot_to_offset(slot);
if (varying == VARYING_SLOT_EDGE) {
if (force_edgeflag)
brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(1));
else
brw_MOV(p, deref_4f(dest_ptr, delta), deref_4f(v0_ptr, delta));
} else if (varying == VARYING_SLOT_PSIZ ||
varying == VARYING_SLOT_CLIP_DIST0 ||
varying == VARYING_SLOT_CLIP_DIST1) {
/* PSIZ doesn't need interpolation because it isn't used by the
* fragment shader. CLIP_DIST0 and CLIP_DIST1 don't need
* intepolation because on pre-GEN6, these are just placeholder VUE
* slots that don't perform any action.
*/
} else if (varying < VARYING_SLOT_MAX) {
/* This is a true vertex result (and not a special value for the VUE
* header), so interpolate:
*
* New = attr0 + t*attr1 - t*attr0
*/
brw_MUL(p,
vec4(brw_null_reg()),
deref_4f(v1_ptr, delta),
t0);
brw_MAC(p,
tmp,
negate(deref_4f(v0_ptr, delta)),
t0);
brw_ADD(p,
deref_4f(dest_ptr, delta),
deref_4f(v0_ptr, delta),
tmp);
}
}
if (c->vue_map.num_slots % 2) {
GLuint delta = brw_vue_slot_to_offset(c->vue_map.num_slots);
brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(0));
}
release_tmp(c, tmp);
/* Recreate the projected (NDC) coordinate in the new vertex
* header:
*/
brw_clip_project_vertex(c, dest_ptr );
}
/***********************************************************************
* Output clipped polygon as an unfilled primitive:
*/
static void emit_lines(struct brw_clip_compile *c,
GLboolean do_offset)
{
struct brw_compile *p = &c->func;
struct brw_instruction *loop;
struct brw_instruction *draw_edge;
struct brw_indirect v0 = brw_indirect(0, 0);
struct brw_indirect v1 = brw_indirect(1, 0);
struct brw_indirect v0ptr = brw_indirect(2, 0);
struct brw_indirect v1ptr = brw_indirect(3, 0);
/* Need a seperate loop for offset:
*/
if (do_offset) {
brw_MOV(p, c->reg.loopcount, c->reg.nr_verts);
brw_MOV(p, get_addr_reg(v0ptr), brw_address(c->reg.inlist));
loop = brw_DO(p, BRW_EXECUTE_1);
{
brw_MOV(p, get_addr_reg(v0), deref_1uw(v0ptr, 0));
brw_ADD(p, get_addr_reg(v0ptr), get_addr_reg(v0ptr), brw_imm_uw(2));
apply_one_offset(c, v0);
brw_set_conditionalmod(p, BRW_CONDITIONAL_G);
brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
}
brw_WHILE(p, loop);
}
/* v1ptr = &inlist[nr_verts]
* *v1ptr = v0
*/
brw_MOV(p, c->reg.loopcount, c->reg.nr_verts);
brw_MOV(p, get_addr_reg(v0ptr), brw_address(c->reg.inlist));
brw_ADD(p, get_addr_reg(v1ptr), get_addr_reg(v0ptr), retype(c->reg.nr_verts, BRW_REGISTER_TYPE_UW));
brw_ADD(p, get_addr_reg(v1ptr), get_addr_reg(v1ptr), retype(c->reg.nr_verts, BRW_REGISTER_TYPE_UW));
brw_MOV(p, deref_1uw(v1ptr, 0), deref_1uw(v0ptr, 0));
loop = brw_DO(p, BRW_EXECUTE_1);
{
brw_MOV(p, get_addr_reg(v0), deref_1uw(v0ptr, 0));
brw_MOV(p, get_addr_reg(v1), deref_1uw(v0ptr, 2));
brw_ADD(p, get_addr_reg(v0ptr), get_addr_reg(v0ptr), brw_imm_uw(2));
/* draw edge if edgeflag != 0 */
brw_CMP(p,
vec1(brw_null_reg()), BRW_CONDITIONAL_NZ,
deref_1f(v0, c->offset[VERT_RESULT_EDGE]),
brw_imm_f(0));
draw_edge = brw_IF(p, BRW_EXECUTE_1);
{
brw_clip_emit_vue(c, v0, 1, 0, (_3DPRIM_LINESTRIP << 2) | R02_PRIM_START);
brw_clip_emit_vue(c, v1, 1, 0, (_3DPRIM_LINESTRIP << 2) | R02_PRIM_END);
}
brw_ENDIF(p, draw_edge);
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
}
brw_WHILE(p, loop);
}
/* Interpolate between two vertices and put the result into a0.0.
* Increment a0.0 accordingly.
*/
void brw_clip_interp_vertex( struct brw_clip_compile *c,
struct brw_indirect dest_ptr,
struct brw_indirect v0_ptr, /* from */
struct brw_indirect v1_ptr, /* to */
struct brw_reg t0,
GLboolean force_edgeflag)
{
struct brw_compile *p = &c->func;
struct brw_reg tmp = get_tmp(c);
GLuint i;
/* Just copy the vertex header:
*/
/*
* After CLIP stage, only first 256 bits of the VUE are read
* back on IGDNG, so needn't change it
*/
brw_copy_indirect_to_indirect(p, dest_ptr, v0_ptr, 1);
/* Iterate over each attribute (could be done in pairs?)
*/
for (i = 0; i < c->nr_attrs; i++) {
GLuint delta = i*16 + 32;
if (BRW_IS_IGDNG(p->brw))
delta = i * 16 + 32 * 3;
if (delta == c->offset[VERT_RESULT_EDGE]) {
if (force_edgeflag)
brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(1));
else
brw_MOV(p, deref_4f(dest_ptr, delta), deref_4f(v0_ptr, delta));
}
else {
/* Interpolate:
*
* New = attr0 + t*attr1 - t*attr0
*/
brw_MUL(p,
vec4(brw_null_reg()),
deref_4f(v1_ptr, delta),
t0);
brw_MAC(p,
tmp,
negate(deref_4f(v0_ptr, delta)),
t0);
brw_ADD(p,
deref_4f(dest_ptr, delta),
deref_4f(v0_ptr, delta),
tmp);
}
}
if (i & 1) {
GLuint delta = i*16 + 32;
if (BRW_IS_IGDNG(p->brw))
delta = i * 16 + 32 * 3;
brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(0));
}
release_tmp(c, tmp);
/* Recreate the projected (NDC) coordinate in the new vertex
* header:
*/
brw_clip_project_vertex(c, dest_ptr );
}
/* 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);
}
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 brw_emit_line_setup(struct brw_sf_compile *c, bool allocate)
{
struct brw_compile *p = &c->func;
GLuint i;
c->flag_value = 0xff;
c->nr_verts = 2;
if (allocate)
alloc_regs(c);
invert_det(c);
copy_z_inv_w(c);
if (c->has_flat_shading)
do_flatshade_line(c);
for (i = 0; i < c->nr_setup_regs; i++)
{
/* Pair of incoming attributes:
*/
struct brw_reg a0 = offset(c->vert[0], i);
struct brw_reg a1 = offset(c->vert[1], i);
GLushort pc, pc_persp, pc_linear;
bool last = calculate_masks(c, i, &pc, &pc_persp, &pc_linear);
if (pc_persp)
{
set_predicate_control_flag_value(p, c, pc_persp);
brw_MUL(p, a0, a0, c->inv_w[0]);
brw_MUL(p, a1, a1, c->inv_w[1]);
}
/* Calculate coefficients for position, color:
*/
if (pc_linear) {
set_predicate_control_flag_value(p, c, pc_linear);
brw_ADD(p, c->a1_sub_a0, a1, negate(a0));
brw_MUL(p, c->tmp, c->a1_sub_a0, c->dx0);
brw_MUL(p, c->m1Cx, c->tmp, c->inv_det);
brw_MUL(p, c->tmp, c->a1_sub_a0, c->dy0);
brw_MUL(p, c->m2Cy, c->tmp, c->inv_det);
}
{
set_predicate_control_flag_value(p, c, pc);
/* start point for interpolation
*/
brw_MOV(p, c->m3C0, a0);
/* Copy m0..m3 to URB.
*/
brw_urb_WRITE(p,
brw_null_reg(),
0,
brw_vec8_grf(0, 0),
last ? BRW_URB_WRITE_EOT_COMPLETE
: BRW_URB_WRITE_NO_FLAGS,
4, /* msg len */
0, /* response len */
i*4, /* urb destination offset */
BRW_URB_SWIZZLE_TRANSPOSE);
}
}
brw_set_default_predicate_control(p, BRW_PREDICATE_NONE);
}
/* 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);
}
/**
* Generate the geometry shader program used on Gen6 to perform stream output
* (transform feedback).
*/
void
gen6_sol_program(struct brw_gs_compile *c, struct brw_gs_prog_key *key,
unsigned num_verts, bool check_edge_flags)
{
struct brw_compile *p = &c->func;
c->prog_data.svbi_postincrement_value = num_verts;
brw_gs_alloc_regs(c, num_verts, true);
brw_gs_initialize_header(c);
if (key->num_transform_feedback_bindings > 0) {
unsigned vertex, binding;
struct brw_reg destination_indices_uw =
vec8(retype(c->reg.destination_indices, BRW_REGISTER_TYPE_UW));
/* Note: since we use the binding table to keep track of buffer offsets
* and stride, the GS doesn't need to keep track of a separate pointer
* into each buffer; it uses a single pointer which increments by 1 for
* each vertex. So we use SVBI0 for this pointer, regardless of whether
* transform feedback is in interleaved or separate attribs mode.
*
* Make sure that the buffers have enough room for all the vertices.
*/
brw_ADD(p, get_element_ud(c->reg.temp, 0),
get_element_ud(c->reg.SVBI, 0), brw_imm_ud(num_verts));
brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_LE,
get_element_ud(c->reg.temp, 0),
get_element_ud(c->reg.SVBI, 4));
brw_IF(p, BRW_EXECUTE_1);
/* Compute the destination indices to write to. Usually we use SVBI[0]
* + (0, 1, 2). However, for odd-numbered triangles in tristrips, the
* vertices come down the pipeline in reversed winding order, so we need
* to flip the order when writing to the transform feedback buffer. To
* ensure that flatshading accuracy is preserved, we need to write them
* in order SVBI[0] + (0, 2, 1) if we're using the first provoking
* vertex convention, and in order SVBI[0] + (1, 0, 2) if we're using
* the last provoking vertex convention.
*
* Note: since brw_imm_v can only be used in instructions in
* packed-word execution mode, and SVBI is a double-word, we need to
* first move the appropriate immediate constant ((0, 1, 2), (0, 2, 1),
* or (1, 0, 2)) to the destination_indices register, and then add SVBI
* using a separate instruction. Also, since the immediate constant is
* expressed as packed words, and we need to load double-words into
* destination_indices, we need to intersperse zeros to fill the upper
* halves of each double-word.
*/
brw_MOV(p, destination_indices_uw,
brw_imm_v(0x00020100)); /* (0, 1, 2) */
if (num_verts == 3) {
/* Get primitive type into temp register. */
brw_AND(p, get_element_ud(c->reg.temp, 0),
get_element_ud(c->reg.R0, 2), brw_imm_ud(0x1f));
/* Test if primitive type is TRISTRIP_REVERSE. We need to do this as
* an 8-wide comparison so that the conditional MOV that follows
* moves all 8 words correctly.
*/
brw_CMP(p, vec8(brw_null_reg()), BRW_CONDITIONAL_EQ,
get_element_ud(c->reg.temp, 0),
brw_imm_ud(_3DPRIM_TRISTRIP_REVERSE));
/* If so, then overwrite destination_indices_uw with the appropriate
* reordering.
*/
brw_MOV(p, destination_indices_uw,
brw_imm_v(key->pv_first ? 0x00010200 /* (0, 2, 1) */
: 0x00020001)); /* (1, 0, 2) */
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
}
brw_ADD(p, c->reg.destination_indices,
c->reg.destination_indices, get_element_ud(c->reg.SVBI, 0));
/* For each vertex, generate code to output each varying using the
* appropriate binding table entry.
*/
for (vertex = 0; vertex < num_verts; ++vertex) {
/* Set up the correct destination index for this vertex */
brw_MOV(p, get_element_ud(c->reg.header, 5),
get_element_ud(c->reg.destination_indices, vertex));
for (binding = 0; binding < key->num_transform_feedback_bindings;
++binding) {
unsigned char varying =
key->transform_feedback_bindings[binding];
unsigned char slot = c->vue_map.varying_to_slot[varying];
/* From the Sandybridge PRM, Volume 2, Part 1, Section 4.5.1:
*
* "Prior to End of Thread with a URB_WRITE, the kernel must
* ensure that all writes are complete by sending the final
* write as a committed write."
*/
bool final_write =
binding == key->num_transform_feedback_bindings - 1 &&
vertex == num_verts - 1;
struct brw_reg vertex_slot = c->reg.vertex[vertex];
vertex_slot.nr += slot / 2;
vertex_slot.subnr = (slot % 2) * 16;
/* gl_PointSize is stored in VARYING_SLOT_PSIZ.w. */
vertex_slot.dw1.bits.swizzle = varying == VARYING_SLOT_PSIZ
? BRW_SWIZZLE_WWWW : key->transform_feedback_swizzles[binding];
brw_set_access_mode(p, BRW_ALIGN_16);
brw_MOV(p, stride(c->reg.header, 4, 4, 1),
retype(vertex_slot, BRW_REGISTER_TYPE_UD));
brw_set_access_mode(p, BRW_ALIGN_1);
brw_svb_write(p,
final_write ? c->reg.temp : brw_null_reg(), /* dest */
1, /* msg_reg_nr */
c->reg.header, /* src0 */
SURF_INDEX_SOL_BINDING(binding), /* binding_table_index */
final_write); /* send_commit_msg */
}
}
brw_ENDIF(p);
/* Now, reinitialize the header register from R0 to restore the parts of
* the register that we overwrote while streaming out transform feedback
* data.
*/
brw_gs_initialize_header(c);
/* Finally, wait for the write commit to occur so that we can proceed to
* other things safely.
*
* From the Sandybridge PRM, Volume 4, Part 1, Section 3.3:
*
* The write commit does not modify the destination register, but
* merely clears the dependency associated with the destination
* register. Thus, a simple “mov” instruction using the register as a
* source is sufficient to wait for the write commit to occur.
*/
brw_MOV(p, c->reg.temp, c->reg.temp);
}
brw_gs_ff_sync(c, 1);
/* If RASTERIZER_DISCARD is enabled, we have nothing further to do, so
* release the URB that was just allocated, and terminate the thread.
*/
if (key->rasterizer_discard) {
brw_gs_terminate(c);
return;
}
brw_gs_overwrite_header_dw2_from_r0(c);
switch (num_verts) {
case 1:
brw_gs_offset_header_dw2(c, URB_WRITE_PRIM_START | URB_WRITE_PRIM_END);
brw_gs_emit_vue(c, c->reg.vertex[0], true);
break;
case 2:
brw_gs_offset_header_dw2(c, URB_WRITE_PRIM_START);
brw_gs_emit_vue(c, c->reg.vertex[0], false);
brw_gs_offset_header_dw2(c, URB_WRITE_PRIM_END - URB_WRITE_PRIM_START);
brw_gs_emit_vue(c, c->reg.vertex[1], true);
break;
case 3:
if (check_edge_flags) {
/* Only emit vertices 0 and 1 if this is the first triangle of the
* polygon. Otherwise they are redundant.
*/
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
brw_AND(p, retype(brw_null_reg(), BRW_REGISTER_TYPE_UD),
get_element_ud(c->reg.R0, 2),
brw_imm_ud(BRW_GS_EDGE_INDICATOR_0));
brw_IF(p, BRW_EXECUTE_1);
}
brw_gs_offset_header_dw2(c, URB_WRITE_PRIM_START);
brw_gs_emit_vue(c, c->reg.vertex[0], false);
brw_gs_offset_header_dw2(c, -URB_WRITE_PRIM_START);
brw_gs_emit_vue(c, c->reg.vertex[1], false);
if (check_edge_flags) {
brw_ENDIF(p);
/* Only emit vertex 2 in PRIM_END mode if this is the last triangle
* of the polygon. Otherwise leave the primitive incomplete because
* there are more polygon vertices coming.
*/
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
brw_AND(p, retype(brw_null_reg(), BRW_REGISTER_TYPE_UD),
get_element_ud(c->reg.R0, 2),
brw_imm_ud(BRW_GS_EDGE_INDICATOR_1));
brw_set_predicate_control(p, BRW_PREDICATE_NORMAL);
}
brw_gs_offset_header_dw2(c, URB_WRITE_PRIM_END);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_gs_emit_vue(c, c->reg.vertex[2], true);
break;
}
}
static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c)
{
#define MAX_IFSN 32
#define MAX_LOOP_DEPTH 32
struct brw_instruction *if_inst[MAX_IFSN], *loop_inst[MAX_LOOP_DEPTH];
struct brw_instruction *inst0, *inst1;
int i, if_insn = 0, loop_insn = 0;
struct brw_compile *p = &c->func;
struct brw_indirect stack_index = brw_indirect(0, 0);
c->reg_index = 0;
prealloc_reg(c);
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_MOV(p, get_addr_reg(stack_index), brw_address(c->stack));
for (i = 0; i < c->nr_fp_insns; i++) {
struct prog_instruction *inst = &c->prog_instructions[i];
struct prog_instruction *orig_inst;
if ((orig_inst = inst->Data) != 0)
orig_inst->Data = current_insn(p);
if (inst->CondUpdate)
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
else
brw_set_conditionalmod(p, BRW_CONDITIONAL_NONE);
switch (inst->Opcode) {
case WM_PIXELXY:
emit_pixel_xy(c, inst);
break;
case WM_DELTAXY:
emit_delta_xy(c, inst);
break;
case WM_PIXELW:
emit_pixel_w(c, inst);
break;
case WM_LINTERP:
emit_linterp(c, inst);
break;
case WM_PINTERP:
emit_pinterp(c, inst);
break;
case WM_CINTERP:
emit_cinterp(c, inst);
break;
case WM_WPOSXY:
emit_wpos_xy(c, inst);
break;
case WM_FB_WRITE:
emit_fb_write(c, inst);
break;
case OPCODE_ABS:
emit_abs(c, inst);
break;
case OPCODE_ADD:
emit_add(c, inst);
break;
case OPCODE_SUB:
emit_sub(c, inst);
break;
case OPCODE_FRC:
emit_frc(c, inst);
break;
case OPCODE_FLR:
emit_flr(c, inst);
break;
case OPCODE_LRP:
emit_lrp(c, inst);
break;
case OPCODE_INT:
emit_int(c, inst);
break;
case OPCODE_MOV:
emit_mov(c, inst);
break;
case OPCODE_DP3:
emit_dp3(c, inst);
break;
case OPCODE_DP4:
emit_dp4(c, inst);
break;
case OPCODE_XPD:
emit_xpd(c, inst);
break;
case OPCODE_DPH:
emit_dph(c, inst);
break;
case OPCODE_RCP:
emit_rcp(c, inst);
break;
case OPCODE_RSQ:
emit_rsq(c, inst);
break;
case OPCODE_SIN:
emit_sin(c, inst);
break;
case OPCODE_COS:
emit_cos(c, inst);
break;
case OPCODE_EX2:
emit_ex2(c, inst);
break;
case OPCODE_LG2:
emit_lg2(c, inst);
break;
case OPCODE_MAX:
emit_max(c, inst);
break;
case OPCODE_MIN:
emit_min(c, inst);
break;
case OPCODE_DDX:
emit_ddx(c, inst);
break;
case OPCODE_DDY:
emit_ddy(c, inst);
break;
case OPCODE_SLT:
emit_slt(c, inst);
break;
case OPCODE_SLE:
emit_sle(c, inst);
break;
case OPCODE_SGT:
emit_sgt(c, inst);
break;
case OPCODE_SGE:
emit_sge(c, inst);
break;
case OPCODE_SEQ:
emit_seq(c, inst);
break;
case OPCODE_SNE:
emit_sne(c, inst);
break;
case OPCODE_MUL:
emit_mul(c, inst);
break;
case OPCODE_POW:
emit_pow(c, inst);
break;
case OPCODE_MAD:
emit_mad(c, inst);
break;
case OPCODE_TEX:
emit_tex(c, inst);
break;
case OPCODE_TXB:
emit_txb(c, inst);
break;
case OPCODE_KIL_NV:
emit_kil(c);
break;
case OPCODE_IF:
assert(if_insn < MAX_IFSN);
if_inst[if_insn++] = brw_IF(p, BRW_EXECUTE_8);
break;
case OPCODE_ELSE:
if_inst[if_insn-1] = brw_ELSE(p, if_inst[if_insn-1]);
break;
case OPCODE_ENDIF:
assert(if_insn > 0);
brw_ENDIF(p, if_inst[--if_insn]);
break;
case OPCODE_BGNSUB:
case OPCODE_ENDSUB:
break;
case OPCODE_CAL:
brw_push_insn_state(p);
brw_set_mask_control(p, BRW_MASK_DISABLE);
brw_set_access_mode(p, BRW_ALIGN_1);
brw_ADD(p, deref_1ud(stack_index, 0), brw_ip_reg(), brw_imm_d(3*16));
brw_set_access_mode(p, BRW_ALIGN_16);
brw_ADD(p, get_addr_reg(stack_index),
get_addr_reg(stack_index), brw_imm_d(4));
orig_inst = inst->Data;
orig_inst->Data = &p->store[p->nr_insn];
brw_ADD(p, brw_ip_reg(), brw_ip_reg(), brw_imm_d(1*16));
brw_pop_insn_state(p);
break;
case OPCODE_RET:
brw_push_insn_state(p);
brw_set_mask_control(p, BRW_MASK_DISABLE);
brw_ADD(p, get_addr_reg(stack_index),
get_addr_reg(stack_index), brw_imm_d(-4));
brw_set_access_mode(p, BRW_ALIGN_1);
brw_MOV(p, brw_ip_reg(), deref_1ud(stack_index, 0));
brw_set_access_mode(p, BRW_ALIGN_16);
brw_pop_insn_state(p);
break;
case OPCODE_BGNLOOP:
loop_inst[loop_insn++] = brw_DO(p, BRW_EXECUTE_8);
break;
case OPCODE_BRK:
brw_BREAK(p);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
break;
case OPCODE_CONT:
brw_CONT(p);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
break;
case OPCODE_ENDLOOP:
loop_insn--;
inst0 = inst1 = brw_WHILE(p, loop_inst[loop_insn]);
/* patch all the BREAK instructions from
last BEGINLOOP */
while (inst0 > loop_inst[loop_insn]) {
inst0--;
if (inst0->header.opcode == BRW_OPCODE_BREAK) {
inst0->bits3.if_else.jump_count = inst1 - inst0 + 1;
inst0->bits3.if_else.pop_count = 0;
} else if (inst0->header.opcode == BRW_OPCODE_CONTINUE) {
inst0->bits3.if_else.jump_count = inst1 - inst0;
inst0->bits3.if_else.pop_count = 0;
}
}
break;
default:
_mesa_printf("unsupported IR in fragment shader %d\n",
inst->Opcode);
}
if (inst->CondUpdate)
brw_set_predicate_control(p, BRW_PREDICATE_NORMAL);
else
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
}
post_wm_emit(c);
for (i = 0; i < c->fp->program.Base.NumInstructions; i++)
c->fp->program.Base.Instructions[i].Data = NULL;
}
