static void transform_r300_vertex_SSG(struct radeon_compiler* c,
struct rc_instruction* inst)
{
/* result = sign(x)
*
* SLT tmp0, 0, x;
* SLT tmp1, x, 0;
* ADD result, tmp0, -tmp1;
*/
struct rc_dst_register dst0 = try_to_reuse_dst(c, inst);
unsigned tmp1;
/* 0 < x */
dst0 = try_to_reuse_dst(c, inst);
emit2(c, inst->Prev, RC_OPCODE_SLT, 0,
dst0,
builtin_zero,
inst->U.I.SrcReg[0]);
/* x < 0 */
tmp1 = rc_find_free_temporary(c);
emit2(c, inst->Prev, RC_OPCODE_SLT, 0,
dstregtmpmask(tmp1, inst->U.I.DstReg.WriteMask),
inst->U.I.SrcReg[0],
builtin_zero);
/* Either both are zero, or one of them is one and the other is zero. */
/* result = tmp0 - tmp1 */
emit2(c, inst->Prev, RC_OPCODE_ADD, 0,
inst->U.I.DstReg,
srcreg(RC_FILE_TEMPORARY, dst0.Index),
negate(srcreg(RC_FILE_TEMPORARY, tmp1)));
rc_remove_instruction(inst);
}
static void transform_r300_vertex_CMP(struct radeon_compiler* c,
struct rc_instruction* inst)
{
/* There is no decent CMP available, so let's rig one up.
* CMP is defined as dst = src0 < 0.0 ? src1 : src2
* The following sequence consumes zero to two temps and two extra slots
* (the second temp and the second slot is consumed by transform_LRP),
* but should be equivalent:
*
* SLT tmp0, src0, 0.0
* LRP dst, tmp0, src1, src2
*
* Yes, I know, I'm a mad scientist. ~ C. & M. */
struct rc_dst_register dst = try_to_reuse_dst(c, inst);
/* SLT tmp0, src0, 0.0 */
emit2(c, inst->Prev, RC_OPCODE_SLT, 0,
dst,
inst->U.I.SrcReg[0], builtin_zero);
/* LRP dst, tmp0, src1, src2 */
transform_LRP(c,
emit3(c, inst->Prev, RC_OPCODE_LRP, 0,
inst->U.I.DstReg,
srcreg(RC_FILE_TEMPORARY, dst.Index), inst->U.I.SrcReg[1], inst->U.I.SrcReg[2]));
rc_remove_instruction(inst);
}
static void transform_FLR(struct radeon_compiler* c,
struct rc_instruction* inst)
{
struct rc_dst_register dst = try_to_reuse_dst(c, inst);
emit1(c, inst->Prev, RC_OPCODE_FRC, 0, dst, inst->U.I.SrcReg[0]);
emit2(c, inst->Prev, RC_OPCODE_ADD, inst->U.I.SaturateMode, inst->U.I.DstReg,
inst->U.I.SrcReg[0], negate(srcreg(RC_FILE_TEMPORARY, dst.Index)));
rc_remove_instruction(inst);
}
static void transform_SNE(struct radeon_compiler* c,
struct rc_instruction* inst)
{
struct rc_dst_register dst = try_to_reuse_dst(c, inst);
emit2(c, inst->Prev, RC_OPCODE_ADD, 0, dst, inst->U.I.SrcReg[0], negate(inst->U.I.SrcReg[1]));
emit3(c, inst->Prev, RC_OPCODE_CMP, inst->U.I.SaturateMode, inst->U.I.DstReg,
negate(absolute(srcreg(RC_FILE_TEMPORARY, dst.Index))), builtin_one, builtin_zero);
rc_remove_instruction(inst);
}
static void transform_SLE(struct radeon_compiler* c,
struct rc_instruction* inst)
{
struct rc_dst_register dst = try_to_reuse_dst(c, inst);
emit2(c, inst->Prev, RC_OPCODE_ADD, 0, dst, negate(inst->U.I.SrcReg[0]), inst->U.I.SrcReg[1]);
emit3(c, inst->Prev, RC_OPCODE_CMP, &inst->U.I, inst->U.I.DstReg,
srcreg(RC_FILE_TEMPORARY, dst.Index), builtin_zero, builtin_one);
rc_remove_instruction(inst);
}
static void transform_POW(struct radeon_compiler* c,
struct rc_instruction* inst)
{
struct rc_dst_register tempdst = try_to_reuse_dst(c, inst);
struct rc_src_register tempsrc = srcreg(RC_FILE_TEMPORARY, tempdst.Index);
tempdst.WriteMask = RC_MASK_W;
tempsrc.Swizzle = RC_SWIZZLE_WWWW;
emit1(c, inst->Prev, RC_OPCODE_LG2, 0, tempdst, swizzle_xxxx(inst->U.I.SrcReg[0]));
emit2(c, inst->Prev, RC_OPCODE_MUL, 0, tempdst, tempsrc, swizzle_xxxx(inst->U.I.SrcReg[1]));
emit1(c, inst->Prev, RC_OPCODE_EX2, inst->U.I.SaturateMode, inst->U.I.DstReg, tempsrc);
rc_remove_instruction(inst);
}
static void transform_LRP(struct radeon_compiler* c,
struct rc_instruction* inst)
{
struct rc_dst_register dst = try_to_reuse_dst(c, inst);
emit2(c, inst->Prev, RC_OPCODE_ADD, 0,
dst,
inst->U.I.SrcReg[1], negate(inst->U.I.SrcReg[2]));
emit3(c, inst->Prev, RC_OPCODE_MAD, &inst->U.I,
inst->U.I.DstReg,
inst->U.I.SrcReg[0], srcreg(RC_FILE_TEMPORARY, dst.Index), inst->U.I.SrcReg[2]);
rc_remove_instruction(inst);
}
static void transform_XPD(struct radeon_compiler* c,
struct rc_instruction* inst)
{
struct rc_dst_register dst = try_to_reuse_dst(c, inst);
emit2(c, inst->Prev, RC_OPCODE_MUL, 0, dst,
swizzle(inst->U.I.SrcReg[0], RC_SWIZZLE_Z, RC_SWIZZLE_X, RC_SWIZZLE_Y, RC_SWIZZLE_W),
swizzle(inst->U.I.SrcReg[1], RC_SWIZZLE_Y, RC_SWIZZLE_Z, RC_SWIZZLE_X, RC_SWIZZLE_W));
emit3(c, inst->Prev, RC_OPCODE_MAD, inst->U.I.SaturateMode, inst->U.I.DstReg,
swizzle(inst->U.I.SrcReg[0], RC_SWIZZLE_Y, RC_SWIZZLE_Z, RC_SWIZZLE_X, RC_SWIZZLE_W),
swizzle(inst->U.I.SrcReg[1], RC_SWIZZLE_Z, RC_SWIZZLE_X, RC_SWIZZLE_Y, RC_SWIZZLE_W),
negate(srcreg(RC_FILE_TEMPORARY, dst.Index)));
rc_remove_instruction(inst);
}
static void transform_CLAMP(struct radeon_compiler *c,
struct rc_instruction *inst)
{
/* CLAMP dst, src, min, max
* into:
* MIN tmp, src, max
* MAX dst, tmp, min
*/
struct rc_dst_register dst = try_to_reuse_dst(c, inst);
emit2(c, inst->Prev, RC_OPCODE_MIN, 0, dst,
inst->U.I.SrcReg[0], inst->U.I.SrcReg[2]);
emit2(c, inst->Prev, RC_OPCODE_MAX, inst->U.I.SaturateMode, inst->U.I.DstReg,
srcreg(RC_FILE_TEMPORARY, dst.Index), inst->U.I.SrcReg[1]);
rc_remove_instruction(inst);
}
static void transform_TRUNC(struct radeon_compiler* c,
struct rc_instruction* inst)
{
/* Definition of trunc:
* trunc(x) = (abs(x) - fract(abs(x))) * sgn(x)
*
* The multiplication by sgn(x) can be simplified using CMP:
* y * sgn(x) = (x < 0 ? -y : y)
*/
struct rc_dst_register dst = try_to_reuse_dst(c, inst);
emit1(c, inst->Prev, RC_OPCODE_FRC, 0, dst, absolute(inst->U.I.SrcReg[0]));
emit2(c, inst->Prev, RC_OPCODE_ADD, 0, dst, absolute(inst->U.I.SrcReg[0]),
negate(srcreg(RC_FILE_TEMPORARY, dst.Index)));
emit3(c, inst->Prev, RC_OPCODE_CMP, &inst->U.I, inst->U.I.DstReg, inst->U.I.SrcReg[0],
negate(srcreg(RC_FILE_TEMPORARY, dst.Index)), srcreg(RC_FILE_TEMPORARY, dst.Index));
rc_remove_instruction(inst);
}
static void transform_CEIL(struct radeon_compiler* c,
struct rc_instruction* inst)
{
/* Assuming:
* ceil(x) = -floor(-x)
*
* After inlining floor:
* ceil(x) = -(-x-frac(-x))
*
* After simplification:
* ceil(x) = x+frac(-x)
*/
struct rc_dst_register dst = try_to_reuse_dst(c, inst);
emit1(c, inst->Prev, RC_OPCODE_FRC, 0, dst, negate(inst->U.I.SrcReg[0]));
emit2(c, inst->Prev, RC_OPCODE_ADD, inst->U.I.SaturateMode, inst->U.I.DstReg,
inst->U.I.SrcReg[0], srcreg(RC_FILE_TEMPORARY, dst.Index));
rc_remove_instruction(inst);
}
static void transform_r300_vertex_fix_LIT(struct radeon_compiler* c,
struct rc_instruction* inst)
{
struct rc_dst_register dst = try_to_reuse_dst(c, inst);
unsigned constant_swizzle;
int constant = rc_constants_add_immediate_scalar(&c->Program.Constants,
0.0000000000000000001,
&constant_swizzle);
/* MOV dst, src */
dst.WriteMask = RC_MASK_XYZW;
emit1(c, inst->Prev, RC_OPCODE_MOV, 0,
dst,
inst->U.I.SrcReg[0]);
/* MAX dst.y, src, 0.00...001 */
emit2(c, inst->Prev, RC_OPCODE_MAX, 0,
dstregtmpmask(dst.Index, RC_MASK_Y),
srcreg(RC_FILE_TEMPORARY, dst.Index),
srcregswz(RC_FILE_CONSTANT, constant, constant_swizzle));
inst->U.I.SrcReg[0] = srcreg(RC_FILE_TEMPORARY, dst.Index);
}