Google Git
Sign in
fuchsia / third_party / mesa / a748a57e636985e7a30db44be8b935beddb92560 / . / src / gallium / auxiliary / tgsi / tgsi_exec.c
blob: 5e440353e36646e9a488402803b318fb0a5a932a [file] [log] [blame]
/**************************************************************************
*
* Copyright 2007-2008 VMware, Inc.
* All Rights Reserved.
* Copyright 2009-2010 VMware, Inc. All rights Reserved.
*
* 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, sub license, 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 NON-INFRINGEMENT.
* IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS 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.
*
**************************************************************************/
/**
* TGSI interpreter/executor.
*
* Flow control information:
*
* Since we operate on 'quads' (4 pixels or 4 vertices in parallel)
* flow control statements (IF/ELSE/ENDIF, LOOP/ENDLOOP) require special
* care since a condition may be true for some quad components but false
* for other components.
*
* We basically execute all statements (even if they're in the part of
* an IF/ELSE clause that's "not taken") and use a special mask to
* control writing to destination registers. This is the ExecMask.
* See store_dest().
*
* The ExecMask is computed from three other masks (CondMask, LoopMask and
* ContMask) which are controlled by the flow control instructions (namely:
* (IF/ELSE/ENDIF, LOOP/ENDLOOP and CONT).
*
*
* Authors:
* Michal Krol
* Brian Paul
*/
#include "pipe/p_compiler.h"
#include "pipe/p_state.h"
#include "pipe/p_shader_tokens.h"
#include "tgsi/tgsi_dump.h"
#include "tgsi/tgsi_parse.h"
#include "tgsi/tgsi_util.h"
#include "tgsi_exec.h"
#include "util/compiler.h"
#include "util/half_float.h"
#include "util/u_memory.h"
#include "util/u_math.h"
#include "util/rounding.h"
#define DEBUG_EXECUTION 0
#define TILE_TOP_LEFT 0
#define TILE_TOP_RIGHT 1
#define TILE_BOTTOM_LEFT 2
#define TILE_BOTTOM_RIGHT 3
union tgsi_double_channel {
double d[TGSI_QUAD_SIZE];
unsigned u[TGSI_QUAD_SIZE][2];
uint64_t u64[TGSI_QUAD_SIZE];
int64_t i64[TGSI_QUAD_SIZE];
} ALIGN16;
struct ALIGN16 tgsi_double_vector {
union tgsi_double_channel xy;
union tgsi_double_channel zw;
};
static void
micro_abs(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
dst->f[0] = fabsf(src->f[0]);
dst->f[1] = fabsf(src->f[1]);
dst->f[2] = fabsf(src->f[2]);
dst->f[3] = fabsf(src->f[3]);
}
static void
micro_arl(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
dst->i[0] = (int)floorf(src->f[0]);
dst->i[1] = (int)floorf(src->f[1]);
dst->i[2] = (int)floorf(src->f[2]);
dst->i[3] = (int)floorf(src->f[3]);
}
static void
micro_arr(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
dst->i[0] = (int)floorf(src->f[0] + 0.5f);
dst->i[1] = (int)floorf(src->f[1] + 0.5f);
dst->i[2] = (int)floorf(src->f[2] + 0.5f);
dst->i[3] = (int)floorf(src->f[3] + 0.5f);
}
static void
micro_ceil(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
dst->f[0] = ceilf(src->f[0]);
dst->f[1] = ceilf(src->f[1]);
dst->f[2] = ceilf(src->f[2]);
dst->f[3] = ceilf(src->f[3]);
}
static void
micro_cmp(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src0,
const union tgsi_exec_channel *src1,
const union tgsi_exec_channel *src2)
{
dst->f[0] = src0->f[0] < 0.0f ? src1->f[0] : src2->f[0];
dst->f[1] = src0->f[1] < 0.0f ? src1->f[1] : src2->f[1];
dst->f[2] = src0->f[2] < 0.0f ? src1->f[2] : src2->f[2];
dst->f[3] = src0->f[3] < 0.0f ? src1->f[3] : src2->f[3];
}
static void
micro_cos(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
dst->f[0] = cosf(src->f[0]);
dst->f[1] = cosf(src->f[1]);
dst->f[2] = cosf(src->f[2]);
dst->f[3] = cosf(src->f[3]);
}
static void
micro_d2f(union tgsi_exec_channel *dst,
const union tgsi_double_channel *src)
{
dst->f[0] = (float)src->d[0];
dst->f[1] = (float)src->d[1];
dst->f[2] = (float)src->d[2];
dst->f[3] = (float)src->d[3];
}
static void
micro_d2i(union tgsi_exec_channel *dst,
const union tgsi_double_channel *src)
{
dst->i[0] = (int)src->d[0];
dst->i[1] = (int)src->d[1];
dst->i[2] = (int)src->d[2];
dst->i[3] = (int)src->d[3];
}
static void
micro_d2u(union tgsi_exec_channel *dst,
const union tgsi_double_channel *src)
{
dst->u[0] = (unsigned)src->d[0];
dst->u[1] = (unsigned)src->d[1];
dst->u[2] = (unsigned)src->d[2];
dst->u[3] = (unsigned)src->d[3];
}
static void
micro_dabs(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->d[0] = src->d[0] >= 0.0 ? src->d[0] : -src->d[0];
dst->d[1] = src->d[1] >= 0.0 ? src->d[1] : -src->d[1];
dst->d[2] = src->d[2] >= 0.0 ? src->d[2] : -src->d[2];
dst->d[3] = src->d[3] >= 0.0 ? src->d[3] : -src->d[3];
}
static void
micro_dadd(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->d[0] = src[0].d[0] + src[1].d[0];
dst->d[1] = src[0].d[1] + src[1].d[1];
dst->d[2] = src[0].d[2] + src[1].d[2];
dst->d[3] = src[0].d[3] + src[1].d[3];
}
static void
micro_ddiv(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->d[0] = src[0].d[0] / src[1].d[0];
dst->d[1] = src[0].d[1] / src[1].d[1];
dst->d[2] = src[0].d[2] / src[1].d[2];
dst->d[3] = src[0].d[3] / src[1].d[3];
}
static void
micro_ddx(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
dst->f[0] =
dst->f[1] =
dst->f[2] =
dst->f[3] = src->f[TILE_BOTTOM_RIGHT] - src->f[TILE_BOTTOM_LEFT];
}
static void
micro_ddx_fine(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
dst->f[0] =
dst->f[1] = src->f[TILE_TOP_RIGHT] - src->f[TILE_TOP_LEFT];
dst->f[2] =
dst->f[3] = src->f[TILE_BOTTOM_RIGHT] - src->f[TILE_BOTTOM_LEFT];
}
static void
micro_ddy(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
dst->f[0] =
dst->f[1] =
dst->f[2] =
dst->f[3] = src->f[TILE_BOTTOM_LEFT] - src->f[TILE_TOP_LEFT];
}
static void
micro_ddy_fine(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
dst->f[0] =
dst->f[2] = src->f[TILE_BOTTOM_LEFT] - src->f[TILE_TOP_LEFT];
dst->f[1] =
dst->f[3] = src->f[TILE_BOTTOM_RIGHT] - src->f[TILE_TOP_RIGHT];
}
static void
micro_dmul(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->d[0] = src[0].d[0] * src[1].d[0];
dst->d[1] = src[0].d[1] * src[1].d[1];
dst->d[2] = src[0].d[2] * src[1].d[2];
dst->d[3] = src[0].d[3] * src[1].d[3];
}
static void
micro_dmax(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->d[0] = fmax(src[0].d[0], src[1].d[0]);
dst->d[1] = fmax(src[0].d[1], src[1].d[1]);
dst->d[2] = fmax(src[0].d[2], src[1].d[2]);
dst->d[3] = fmax(src[0].d[3], src[1].d[3]);
}
static void
micro_dmin(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->d[0] = fmin(src[0].d[0], src[1].d[0]);
dst->d[1] = fmin(src[0].d[1], src[1].d[1]);
dst->d[2] = fmin(src[0].d[2], src[1].d[2]);
dst->d[3] = fmin(src[0].d[3], src[1].d[3]);
}
static void
micro_dneg(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->d[0] = -src->d[0];
dst->d[1] = -src->d[1];
dst->d[2] = -src->d[2];
dst->d[3] = -src->d[3];
}
static void
micro_dslt(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->u[0][0] = src[0].d[0] < src[1].d[0] ? ~0U : 0U;
dst->u[1][0] = src[0].d[1] < src[1].d[1] ? ~0U : 0U;
dst->u[2][0] = src[0].d[2] < src[1].d[2] ? ~0U : 0U;
dst->u[3][0] = src[0].d[3] < src[1].d[3] ? ~0U : 0U;
}
static void
micro_dsne(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->u[0][0] = src[0].d[0] != src[1].d[0] ? ~0U : 0U;
dst->u[1][0] = src[0].d[1] != src[1].d[1] ? ~0U : 0U;
dst->u[2][0] = src[0].d[2] != src[1].d[2] ? ~0U : 0U;
dst->u[3][0] = src[0].d[3] != src[1].d[3] ? ~0U : 0U;
}
static void
micro_dsge(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->u[0][0] = src[0].d[0] >= src[1].d[0] ? ~0U : 0U;
dst->u[1][0] = src[0].d[1] >= src[1].d[1] ? ~0U : 0U;
dst->u[2][0] = src[0].d[2] >= src[1].d[2] ? ~0U : 0U;
dst->u[3][0] = src[0].d[3] >= src[1].d[3] ? ~0U : 0U;
}
static void
micro_dseq(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->u[0][0] = src[0].d[0] == src[1].d[0] ? ~0U : 0U;
dst->u[1][0] = src[0].d[1] == src[1].d[1] ? ~0U : 0U;
dst->u[2][0] = src[0].d[2] == src[1].d[2] ? ~0U : 0U;
dst->u[3][0] = src[0].d[3] == src[1].d[3] ? ~0U : 0U;
}
static void
micro_drcp(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->d[0] = 1.0 / src->d[0];
dst->d[1] = 1.0 / src->d[1];
dst->d[2] = 1.0 / src->d[2];
dst->d[3] = 1.0 / src->d[3];
}
static void
micro_dsqrt(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->d[0] = sqrt(src->d[0]);
dst->d[1] = sqrt(src->d[1]);
dst->d[2] = sqrt(src->d[2]);
dst->d[3] = sqrt(src->d[3]);
}
static void
micro_drsq(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->d[0] = 1.0 / sqrt(src->d[0]);
dst->d[1] = 1.0 / sqrt(src->d[1]);
dst->d[2] = 1.0 / sqrt(src->d[2]);
dst->d[3] = 1.0 / sqrt(src->d[3]);
}
static void
micro_dmad(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->d[0] = src[0].d[0] * src[1].d[0] + src[2].d[0];
dst->d[1] = src[0].d[1] * src[1].d[1] + src[2].d[1];
dst->d[2] = src[0].d[2] * src[1].d[2] + src[2].d[2];
dst->d[3] = src[0].d[3] * src[1].d[3] + src[2].d[3];
}
static void
micro_dfrac(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->d[0] = src->d[0] - floor(src->d[0]);
dst->d[1] = src->d[1] - floor(src->d[1]);
dst->d[2] = src->d[2] - floor(src->d[2]);
dst->d[3] = src->d[3] - floor(src->d[3]);
}
static void
micro_dflr(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->d[0] = floor(src->d[0]);
dst->d[1] = floor(src->d[1]);
dst->d[2] = floor(src->d[2]);
dst->d[3] = floor(src->d[3]);
}
static void
micro_dldexp(union tgsi_double_channel *dst,
const union tgsi_double_channel *src0,
union tgsi_exec_channel *src1)
{
dst->d[0] = ldexp(src0->d[0], src1->i[0]);
dst->d[1] = ldexp(src0->d[1], src1->i[1]);
dst->d[2] = ldexp(src0->d[2], src1->i[2]);
dst->d[3] = ldexp(src0->d[3], src1->i[3]);
}
static void
micro_dfracexp(union tgsi_double_channel *dst,
union tgsi_exec_channel *dst_exp,
const union tgsi_double_channel *src)
{
dst->d[0] = frexp(src->d[0], &dst_exp->i[0]);
dst->d[1] = frexp(src->d[1], &dst_exp->i[1]);
dst->d[2] = frexp(src->d[2], &dst_exp->i[2]);
dst->d[3] = frexp(src->d[3], &dst_exp->i[3]);
}
static void
micro_exp2(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
#if DEBUG
/* Inf is okay for this instruction, so clamp it to silence assertions. */
uint i;
union tgsi_exec_channel clamped;
for (i = 0; i < 4; i++) {
if (src->f[i] > 127.99999f) {
clamped.f[i] = 127.99999f;
} else if (src->f[i] < -126.99999f) {
clamped.f[i] = -126.99999f;
} else {
clamped.f[i] = src->f[i];
}
}
src = &clamped;
#endif /* DEBUG */
dst->f[0] = powf(2.0f, src->f[0]);
dst->f[1] = powf(2.0f, src->f[1]);
dst->f[2] = powf(2.0f, src->f[2]);
dst->f[3] = powf(2.0f, src->f[3]);
}
static void
micro_f2d(union tgsi_double_channel *dst,
const union tgsi_exec_channel *src)
{
dst->d[0] = (double)src->f[0];
dst->d[1] = (double)src->f[1];
dst->d[2] = (double)src->f[2];
dst->d[3] = (double)src->f[3];
}
static void
micro_flr(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
dst->f[0] = floorf(src->f[0]);
dst->f[1] = floorf(src->f[1]);
dst->f[2] = floorf(src->f[2]);
dst->f[3] = floorf(src->f[3]);
}
static void
micro_frc(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
dst->f[0] = src->f[0] - floorf(src->f[0]);
dst->f[1] = src->f[1] - floorf(src->f[1]);
dst->f[2] = src->f[2] - floorf(src->f[2]);
dst->f[3] = src->f[3] - floorf(src->f[3]);
}
static void
micro_i2d(union tgsi_double_channel *dst,
const union tgsi_exec_channel *src)
{
dst->d[0] = (double)src->i[0];
dst->d[1] = (double)src->i[1];
dst->d[2] = (double)src->i[2];
dst->d[3] = (double)src->i[3];
}
static void
micro_iabs(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
dst->i[0] = src->i[0] >= 0 ? src->i[0] : -src->i[0];
dst->i[1] = src->i[1] >= 0 ? src->i[1] : -src->i[1];
dst->i[2] = src->i[2] >= 0 ? src->i[2] : -src->i[2];
dst->i[3] = src->i[3] >= 0 ? src->i[3] : -src->i[3];
}
static void
micro_ineg(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
dst->i[0] = -src->i[0];
dst->i[1] = -src->i[1];
dst->i[2] = -src->i[2];
dst->i[3] = -src->i[3];
}
static void
micro_lg2(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
dst->f[0] = logf(src->f[0]) * 1.442695f;
dst->f[1] = logf(src->f[1]) * 1.442695f;
dst->f[2] = logf(src->f[2]) * 1.442695f;
dst->f[3] = logf(src->f[3]) * 1.442695f;
}
static void
micro_lrp(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src0,
const union tgsi_exec_channel *src1,
const union tgsi_exec_channel *src2)
{
dst->f[0] = src0->f[0] * (src1->f[0] - src2->f[0]) + src2->f[0];
dst->f[1] = src0->f[1] * (src1->f[1] - src2->f[1]) + src2->f[1];
dst->f[2] = src0->f[2] * (src1->f[2] - src2->f[2]) + src2->f[2];
dst->f[3] = src0->f[3] * (src1->f[3] - src2->f[3]) + src2->f[3];
}
static void
micro_mad(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src0,
const union tgsi_exec_channel *src1,
const union tgsi_exec_channel *src2)
{
dst->f[0] = src0->f[0] * src1->f[0] + src2->f[0];
dst->f[1] = src0->f[1] * src1->f[1] + src2->f[1];
dst->f[2] = src0->f[2] * src1->f[2] + src2->f[2];
dst->f[3] = src0->f[3] * src1->f[3] + src2->f[3];
}
static void
micro_mov(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
dst->u[0] = src->u[0];
dst->u[1] = src->u[1];
dst->u[2] = src->u[2];
dst->u[3] = src->u[3];
}
static void
micro_rcp(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
#if 0 /* for debugging */
assert(src->f[0] != 0.0f);
assert(src->f[1] != 0.0f);
assert(src->f[2] != 0.0f);
assert(src->f[3] != 0.0f);
#endif
dst->f[0] = 1.0f / src->f[0];
dst->f[1] = 1.0f / src->f[1];
dst->f[2] = 1.0f / src->f[2];
dst->f[3] = 1.0f / src->f[3];
}
static void
micro_rnd(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
dst->f[0] = _mesa_roundevenf(src->f[0]);
dst->f[1] = _mesa_roundevenf(src->f[1]);
dst->f[2] = _mesa_roundevenf(src->f[2]);
dst->f[3] = _mesa_roundevenf(src->f[3]);
}
static void
micro_rsq(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
#if 0 /* for debugging */
assert(src->f[0] != 0.0f);
assert(src->f[1] != 0.0f);
assert(src->f[2] != 0.0f);
assert(src->f[3] != 0.0f);
#endif
dst->f[0] = 1.0f / sqrtf(src->f[0]);
dst->f[1] = 1.0f / sqrtf(src->f[1]);
dst->f[2] = 1.0f / sqrtf(src->f[2]);
dst->f[3] = 1.0f / sqrtf(src->f[3]);
}
static void
micro_sqrt(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
dst->f[0] = sqrtf(src->f[0]);
dst->f[1] = sqrtf(src->f[1]);
dst->f[2] = sqrtf(src->f[2]);
dst->f[3] = sqrtf(src->f[3]);
}
static void
micro_seq(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src0,
const union tgsi_exec_channel *src1)
{
dst->f[0] = src0->f[0] == src1->f[0] ? 1.0f : 0.0f;
dst->f[1] = src0->f[1] == src1->f[1] ? 1.0f : 0.0f;
dst->f[2] = src0->f[2] == src1->f[2] ? 1.0f : 0.0f;
dst->f[3] = src0->f[3] == src1->f[3] ? 1.0f : 0.0f;
}
static void
micro_sge(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src0,
const union tgsi_exec_channel *src1)
{
dst->f[0] = src0->f[0] >= src1->f[0] ? 1.0f : 0.0f;
dst->f[1] = src0->f[1] >= src1->f[1] ? 1.0f : 0.0f;
dst->f[2] = src0->f[2] >= src1->f[2] ? 1.0f : 0.0f;
dst->f[3] = src0->f[3] >= src1->f[3] ? 1.0f : 0.0f;
}
static void
micro_sgn(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
dst->f[0] = src->f[0] < 0.0f ? -1.0f : src->f[0] > 0.0f ? 1.0f : 0.0f;
dst->f[1] = src->f[1] < 0.0f ? -1.0f : src->f[1] > 0.0f ? 1.0f : 0.0f;
dst->f[2] = src->f[2] < 0.0f ? -1.0f : src->f[2] > 0.0f ? 1.0f : 0.0f;
dst->f[3] = src->f[3] < 0.0f ? -1.0f : src->f[3] > 0.0f ? 1.0f : 0.0f;
}
static void
micro_isgn(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
dst->i[0] = src->i[0] < 0 ? -1 : src->i[0] > 0 ? 1 : 0;
dst->i[1] = src->i[1] < 0 ? -1 : src->i[1] > 0 ? 1 : 0;
dst->i[2] = src->i[2] < 0 ? -1 : src->i[2] > 0 ? 1 : 0;
dst->i[3] = src->i[3] < 0 ? -1 : src->i[3] > 0 ? 1 : 0;
}
static void
micro_sgt(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src0,
const union tgsi_exec_channel *src1)
{
dst->f[0] = src0->f[0] > src1->f[0] ? 1.0f : 0.0f;
dst->f[1] = src0->f[1] > src1->f[1] ? 1.0f : 0.0f;
dst->f[2] = src0->f[2] > src1->f[2] ? 1.0f : 0.0f;
dst->f[3] = src0->f[3] > src1->f[3] ? 1.0f : 0.0f;
}
static void
micro_sin(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
dst->f[0] = sinf(src->f[0]);
dst->f[1] = sinf(src->f[1]);
dst->f[2] = sinf(src->f[2]);
dst->f[3] = sinf(src->f[3]);
}
static void
micro_sle(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src0,
const union tgsi_exec_channel *src1)
{
dst->f[0] = src0->f[0] <= src1->f[0] ? 1.0f : 0.0f;
dst->f[1] = src0->f[1] <= src1->f[1] ? 1.0f : 0.0f;
dst->f[2] = src0->f[2] <= src1->f[2] ? 1.0f : 0.0f;
dst->f[3] = src0->f[3] <= src1->f[3] ? 1.0f : 0.0f;
}
static void
micro_slt(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src0,
const union tgsi_exec_channel *src1)
{
dst->f[0] = src0->f[0] < src1->f[0] ? 1.0f : 0.0f;
dst->f[1] = src0->f[1] < src1->f[1] ? 1.0f : 0.0f;
dst->f[2] = src0->f[2] < src1->f[2] ? 1.0f : 0.0f;
dst->f[3] = src0->f[3] < src1->f[3] ? 1.0f : 0.0f;
}
static void
micro_sne(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src0,
const union tgsi_exec_channel *src1)
{
dst->f[0] = src0->f[0] != src1->f[0] ? 1.0f : 0.0f;
dst->f[1] = src0->f[1] != src1->f[1] ? 1.0f : 0.0f;
dst->f[2] = src0->f[2] != src1->f[2] ? 1.0f : 0.0f;
dst->f[3] = src0->f[3] != src1->f[3] ? 1.0f : 0.0f;
}
static void
micro_trunc(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
dst->f[0] = truncf(src->f[0]);
dst->f[1] = truncf(src->f[1]);
dst->f[2] = truncf(src->f[2]);
dst->f[3] = truncf(src->f[3]);
}
static void
micro_u2d(union tgsi_double_channel *dst,
const union tgsi_exec_channel *src)
{
dst->d[0] = (double)src->u[0];
dst->d[1] = (double)src->u[1];
dst->d[2] = (double)src->u[2];
dst->d[3] = (double)src->u[3];
}
static void
micro_i64abs(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->i64[0] = src->i64[0] >= 0.0 ? src->i64[0] : -src->i64[0];
dst->i64[1] = src->i64[1] >= 0.0 ? src->i64[1] : -src->i64[1];
dst->i64[2] = src->i64[2] >= 0.0 ? src->i64[2] : -src->i64[2];
dst->i64[3] = src->i64[3] >= 0.0 ? src->i64[3] : -src->i64[3];
}
static void
micro_i64sgn(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->i64[0] = src->i64[0] < 0 ? -1 : src->i64[0] > 0 ? 1 : 0;
dst->i64[1] = src->i64[1] < 0 ? -1 : src->i64[1] > 0 ? 1 : 0;
dst->i64[2] = src->i64[2] < 0 ? -1 : src->i64[2] > 0 ? 1 : 0;
dst->i64[3] = src->i64[3] < 0 ? -1 : src->i64[3] > 0 ? 1 : 0;
}
static void
micro_i64neg(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->i64[0] = -src->i64[0];
dst->i64[1] = -src->i64[1];
dst->i64[2] = -src->i64[2];
dst->i64[3] = -src->i64[3];
}
static void
micro_u64seq(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->u[0][0] = src[0].u64[0] == src[1].u64[0] ? ~0U : 0U;
dst->u[1][0] = src[0].u64[1] == src[1].u64[1] ? ~0U : 0U;
dst->u[2][0] = src[0].u64[2] == src[1].u64[2] ? ~0U : 0U;
dst->u[3][0] = src[0].u64[3] == src[1].u64[3] ? ~0U : 0U;
}
static void
micro_u64sne(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->u[0][0] = src[0].u64[0] != src[1].u64[0] ? ~0U : 0U;
dst->u[1][0] = src[0].u64[1] != src[1].u64[1] ? ~0U : 0U;
dst->u[2][0] = src[0].u64[2] != src[1].u64[2] ? ~0U : 0U;
dst->u[3][0] = src[0].u64[3] != src[1].u64[3] ? ~0U : 0U;
}
static void
micro_i64slt(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->u[0][0] = src[0].i64[0] < src[1].i64[0] ? ~0U : 0U;
dst->u[1][0] = src[0].i64[1] < src[1].i64[1] ? ~0U : 0U;
dst->u[2][0] = src[0].i64[2] < src[1].i64[2] ? ~0U : 0U;
dst->u[3][0] = src[0].i64[3] < src[1].i64[3] ? ~0U : 0U;
}
static void
micro_u64slt(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->u[0][0] = src[0].u64[0] < src[1].u64[0] ? ~0U : 0U;
dst->u[1][0] = src[0].u64[1] < src[1].u64[1] ? ~0U : 0U;
dst->u[2][0] = src[0].u64[2] < src[1].u64[2] ? ~0U : 0U;
dst->u[3][0] = src[0].u64[3] < src[1].u64[3] ? ~0U : 0U;
}
static void
micro_i64sge(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->u[0][0] = src[0].i64[0] >= src[1].i64[0] ? ~0U : 0U;
dst->u[1][0] = src[0].i64[1] >= src[1].i64[1] ? ~0U : 0U;
dst->u[2][0] = src[0].i64[2] >= src[1].i64[2] ? ~0U : 0U;
dst->u[3][0] = src[0].i64[3] >= src[1].i64[3] ? ~0U : 0U;
}
static void
micro_u64sge(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->u[0][0] = src[0].u64[0] >= src[1].u64[0] ? ~0U : 0U;
dst->u[1][0] = src[0].u64[1] >= src[1].u64[1] ? ~0U : 0U;
dst->u[2][0] = src[0].u64[2] >= src[1].u64[2] ? ~0U : 0U;
dst->u[3][0] = src[0].u64[3] >= src[1].u64[3] ? ~0U : 0U;
}
static void
micro_u64max(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->u64[0] = src[0].u64[0] > src[1].u64[0] ? src[0].u64[0] : src[1].u64[0];
dst->u64[1] = src[0].u64[1] > src[1].u64[1] ? src[0].u64[1] : src[1].u64[1];
dst->u64[2] = src[0].u64[2] > src[1].u64[2] ? src[0].u64[2] : src[1].u64[2];
dst->u64[3] = src[0].u64[3] > src[1].u64[3] ? src[0].u64[3] : src[1].u64[3];
}
static void
micro_i64max(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->i64[0] = src[0].i64[0] > src[1].i64[0] ? src[0].i64[0] : src[1].i64[0];
dst->i64[1] = src[0].i64[1] > src[1].i64[1] ? src[0].i64[1] : src[1].i64[1];
dst->i64[2] = src[0].i64[2] > src[1].i64[2] ? src[0].i64[2] : src[1].i64[2];
dst->i64[3] = src[0].i64[3] > src[1].i64[3] ? src[0].i64[3] : src[1].i64[3];
}
static void
micro_u64min(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->u64[0] = src[0].u64[0] < src[1].u64[0] ? src[0].u64[0] : src[1].u64[0];
dst->u64[1] = src[0].u64[1] < src[1].u64[1] ? src[0].u64[1] : src[1].u64[1];
dst->u64[2] = src[0].u64[2] < src[1].u64[2] ? src[0].u64[2] : src[1].u64[2];
dst->u64[3] = src[0].u64[3] < src[1].u64[3] ? src[0].u64[3] : src[1].u64[3];
}
static void
micro_i64min(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->i64[0] = src[0].i64[0] < src[1].i64[0] ? src[0].i64[0] : src[1].i64[0];
dst->i64[1] = src[0].i64[1] < src[1].i64[1] ? src[0].i64[1] : src[1].i64[1];
dst->i64[2] = src[0].i64[2] < src[1].i64[2] ? src[0].i64[2] : src[1].i64[2];
dst->i64[3] = src[0].i64[3] < src[1].i64[3] ? src[0].i64[3] : src[1].i64[3];
}
static void
micro_u64add(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->u64[0] = src[0].u64[0] + src[1].u64[0];
dst->u64[1] = src[0].u64[1] + src[1].u64[1];
dst->u64[2] = src[0].u64[2] + src[1].u64[2];
dst->u64[3] = src[0].u64[3] + src[1].u64[3];
}
static void
micro_u64mul(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->u64[0] = src[0].u64[0] * src[1].u64[0];
dst->u64[1] = src[0].u64[1] * src[1].u64[1];
dst->u64[2] = src[0].u64[2] * src[1].u64[2];
dst->u64[3] = src[0].u64[3] * src[1].u64[3];
}
static void
micro_u64div(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->u64[0] = src[1].u64[0] ? src[0].u64[0] / src[1].u64[0] : ~0ull;
dst->u64[1] = src[1].u64[1] ? src[0].u64[1] / src[1].u64[1] : ~0ull;
dst->u64[2] = src[1].u64[2] ? src[0].u64[2] / src[1].u64[2] : ~0ull;
dst->u64[3] = src[1].u64[3] ? src[0].u64[3] / src[1].u64[3] : ~0ull;
}
static void
micro_i64div(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->i64[0] = src[1].i64[0] ? src[0].i64[0] / src[1].i64[0] : 0;
dst->i64[1] = src[1].i64[1] ? src[0].i64[1] / src[1].i64[1] : 0;
dst->i64[2] = src[1].i64[2] ? src[0].i64[2] / src[1].i64[2] : 0;
dst->i64[3] = src[1].i64[3] ? src[0].i64[3] / src[1].i64[3] : 0;
}
static void
micro_u64mod(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->u64[0] = src[1].u64[0] ? src[0].u64[0] % src[1].u64[0] : ~0ull;
dst->u64[1] = src[1].u64[1] ? src[0].u64[1] % src[1].u64[1] : ~0ull;
dst->u64[2] = src[1].u64[2] ? src[0].u64[2] % src[1].u64[2] : ~0ull;
dst->u64[3] = src[1].u64[3] ? src[0].u64[3] % src[1].u64[3] : ~0ull;
}
static void
micro_i64mod(union tgsi_double_channel *dst,
const union tgsi_double_channel *src)
{
dst->i64[0] = src[1].i64[0] ? src[0].i64[0] % src[1].i64[0] : ~0ll;
dst->i64[1] = src[1].i64[1] ? src[0].i64[1] % src[1].i64[1] : ~0ll;
dst->i64[2] = src[1].i64[2] ? src[0].i64[2] % src[1].i64[2] : ~0ll;
dst->i64[3] = src[1].i64[3] ? src[0].i64[3] % src[1].i64[3] : ~0ll;
}
static void
micro_u64shl(union tgsi_double_channel *dst,
const union tgsi_double_channel *src0,
union tgsi_exec_channel *src1)
{
unsigned masked_count;
masked_count = src1->u[0] & 0x3f;
dst->u64[0] = src0->u64[0] << masked_count;
masked_count = src1->u[1] & 0x3f;
dst->u64[1] = src0->u64[1] << masked_count;
masked_count = src1->u[2] & 0x3f;
dst->u64[2] = src0->u64[2] << masked_count;
masked_count = src1->u[3] & 0x3f;
dst->u64[3] = src0->u64[3] << masked_count;
}
static void
micro_i64shr(union tgsi_double_channel *dst,
const union tgsi_double_channel *src0,
union tgsi_exec_channel *src1)
{
unsigned masked_count;
masked_count = src1->u[0] & 0x3f;
dst->i64[0] = src0->i64[0] >> masked_count;
masked_count = src1->u[1] & 0x3f;
dst->i64[1] = src0->i64[1] >> masked_count;
masked_count = src1->u[2] & 0x3f;
dst->i64[2] = src0->i64[2] >> masked_count;
masked_count = src1->u[3] & 0x3f;
dst->i64[3] = src0->i64[3] >> masked_count;
}
static void
micro_u64shr(union tgsi_double_channel *dst,
const union tgsi_double_channel *src0,
union tgsi_exec_channel *src1)
{
unsigned masked_count;
masked_count = src1->u[0] & 0x3f;
dst->u64[0] = src0->u64[0] >> masked_count;
masked_count = src1->u[1] & 0x3f;
dst->u64[1] = src0->u64[1] >> masked_count;
masked_count = src1->u[2] & 0x3f;
dst->u64[2] = src0->u64[2] >> masked_count;
masked_count = src1->u[3] & 0x3f;
dst->u64[3] = src0->u64[3] >> masked_count;
}
enum tgsi_exec_datatype {
TGSI_EXEC_DATA_FLOAT,
TGSI_EXEC_DATA_INT,
TGSI_EXEC_DATA_UINT,
TGSI_EXEC_DATA_DOUBLE,
TGSI_EXEC_DATA_INT64,
TGSI_EXEC_DATA_UINT64,
};
/** The execution mask depends on the conditional mask and the loop mask */
#define UPDATE_EXEC_MASK(MACH) \
MACH->ExecMask = MACH->CondMask & MACH->LoopMask & MACH->ContMask & MACH->Switch.mask & MACH->FuncMask
static const union tgsi_exec_channel ZeroVec =
{ { 0.0, 0.0, 0.0, 0.0 } };
static const union tgsi_exec_channel OneVec = {
{1.0f, 1.0f, 1.0f, 1.0f}
};
static const union tgsi_exec_channel P128Vec = {
{128.0f, 128.0f, 128.0f, 128.0f}
};
static const union tgsi_exec_channel M128Vec = {
{-128.0f, -128.0f, -128.0f, -128.0f}
};
#ifdef DEBUG
static void
print_chan(const char *msg, const union tgsi_exec_channel *chan)
{
debug_printf("%s = {%f, %f, %f, %f}\n",
msg, chan->f[0], chan->f[1], chan->f[2], chan->f[3]);
}
#endif
#ifdef DEBUG
static void
print_temp(const struct tgsi_exec_machine *mach, uint index)
{
const struct tgsi_exec_vector *tmp = &mach->Temps[index];
int i;
debug_printf("Temp[%u] =\n", index);
for (i = 0; i < 4; i++) {
debug_printf(" %c: { %f, %f, %f, %f }\n",
"XYZW"[i],
tmp->xyzw[i].f[0],
tmp->xyzw[i].f[1],
tmp->xyzw[i].f[2],
tmp->xyzw[i].f[3]);
}
}
#endif
void
tgsi_exec_set_constant_buffers(struct tgsi_exec_machine *mach,
unsigned num_bufs,
const void **bufs,
const unsigned *buf_sizes)
{
unsigned i;
for (i = 0; i < num_bufs; i++) {
mach->Consts[i] = bufs[i];
mach->ConstsSize[i] = buf_sizes[i];
}
}
/**
* Initialize machine state by expanding tokens to full instructions,
* allocating temporary storage, setting up constants, etc.
* After this, we can call tgsi_exec_machine_run() many times.
*/
void
tgsi_exec_machine_bind_shader(
struct tgsi_exec_machine *mach,
const struct tgsi_token *tokens,
struct tgsi_sampler *sampler,
struct tgsi_image *image,
struct tgsi_buffer *buffer)
{
uint k;
struct tgsi_parse_context parse;
struct tgsi_full_instruction *instructions;
struct tgsi_full_declaration *declarations;
uint maxInstructions = 10, numInstructions = 0;
uint maxDeclarations = 10, numDeclarations = 0;
#if 0
tgsi_dump(tokens, 0);
#endif
mach->Tokens = tokens;
mach->Sampler = sampler;
mach->Image = image;
mach->Buffer = buffer;
if (!tokens) {
/* unbind and free all */
FREE(mach->Declarations);
mach->Declarations = NULL;
mach->NumDeclarations = 0;
FREE(mach->Instructions);
mach->Instructions = NULL;
mach->NumInstructions = 0;
return;
}
k = tgsi_parse_init (&parse, mach->Tokens);
if (k != TGSI_PARSE_OK) {
debug_printf( "Problem parsing!\n" );
return;
}
mach->ImmLimit = 0;
mach->NumOutputs = 0;
for (k = 0; k < TGSI_SEMANTIC_COUNT; k++)
mach->SysSemanticToIndex[k] = -1;
if (mach->ShaderType == PIPE_SHADER_GEOMETRY &&
!mach->UsedGeometryShader) {
struct tgsi_exec_vector *inputs;
struct tgsi_exec_vector *outputs;
inputs = align_malloc(sizeof(struct tgsi_exec_vector) *
TGSI_MAX_PRIM_VERTICES * PIPE_MAX_SHADER_INPUTS,
16);
if (!inputs)
return;
outputs = align_malloc(sizeof(struct tgsi_exec_vector) *
TGSI_MAX_TOTAL_VERTICES, 16);
if (!outputs) {
align_free(inputs);
return;
}
align_free(mach->Inputs);
align_free(mach->Outputs);
mach->Inputs = inputs;
mach->Outputs = outputs;
mach->UsedGeometryShader = TRUE;
}
declarations = (struct tgsi_full_declaration *)
MALLOC( maxDeclarations * sizeof(struct tgsi_full_declaration) );
if (!declarations) {
return;
}
instructions = (struct tgsi_full_instruction *)
MALLOC( maxInstructions * sizeof(struct tgsi_full_instruction) );
if (!instructions) {
FREE( declarations );
return;
}
while( !tgsi_parse_end_of_tokens( &parse ) ) {
uint i;
tgsi_parse_token( &parse );
switch( parse.FullToken.Token.Type ) {
case TGSI_TOKEN_TYPE_DECLARATION:
/* save expanded declaration */
if (numDeclarations == maxDeclarations) {
declarations = REALLOC(declarations,
maxDeclarations
* sizeof(struct tgsi_full_declaration),
(maxDeclarations + 10)
* sizeof(struct tgsi_full_declaration));
maxDeclarations += 10;
}
if (parse.FullToken.FullDeclaration.Declaration.File == TGSI_FILE_OUTPUT)
mach->NumOutputs = MAX2(mach->NumOutputs, parse.FullToken.FullDeclaration.Range.Last + 1);
else if (parse.FullToken.FullDeclaration.Declaration.File == TGSI_FILE_SYSTEM_VALUE) {
const struct tgsi_full_declaration *decl = &parse.FullToken.FullDeclaration;
mach->SysSemanticToIndex[decl->Semantic.Name] = decl->Range.First;
}
memcpy(declarations + numDeclarations,
&parse.FullToken.FullDeclaration,
sizeof(declarations[0]));
numDeclarations++;
break;
case TGSI_TOKEN_TYPE_IMMEDIATE:
{
uint size = parse.FullToken.FullImmediate.Immediate.NrTokens - 1;
assert( size <= 4 );
if (mach->ImmLimit >= mach->ImmsReserved) {
unsigned newReserved = mach->ImmsReserved ? 2 * mach->ImmsReserved : 128;
float4 *imms = REALLOC(mach->Imms, mach->ImmsReserved, newReserved * sizeof(float4));
if (imms) {
mach->ImmsReserved = newReserved;
mach->Imms = imms;
} else {
debug_printf("Unable to (re)allocate space for immidiate constants\n");
break;
}
}
for( i = 0; i < size; i++ ) {
mach->Imms[mach->ImmLimit][i] =
parse.FullToken.FullImmediate.u[i].Float;
}
mach->ImmLimit += 1;
}
break;
case TGSI_TOKEN_TYPE_INSTRUCTION:
/* save expanded instruction */
if (numInstructions == maxInstructions) {
instructions = REALLOC(instructions,
maxInstructions
* sizeof(struct tgsi_full_instruction),
(maxInstructions + 10)
* sizeof(struct tgsi_full_instruction));
maxInstructions += 10;
}
memcpy(instructions + numInstructions,
&parse.FullToken.FullInstruction,
sizeof(instructions[0]));
numInstructions++;
break;
case TGSI_TOKEN_TYPE_PROPERTY:
if (mach->ShaderType == PIPE_SHADER_GEOMETRY) {
if (parse.FullToken.FullProperty.Property.PropertyName == TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES) {
mach->MaxOutputVertices = parse.FullToken.FullProperty.u[0].Data;
}
}
break;
default:
assert( 0 );
}
}
tgsi_parse_free (&parse);
FREE(mach->Declarations);
mach->Declarations = declarations;
mach->NumDeclarations = numDeclarations;
FREE(mach->Instructions);
mach->Instructions = instructions;
mach->NumInstructions = numInstructions;
}
struct tgsi_exec_machine *
tgsi_exec_machine_create(enum pipe_shader_type shader_type)
{
struct tgsi_exec_machine *mach;
mach = align_malloc( sizeof *mach, 16 );
if (!mach)
goto fail;
memset(mach, 0, sizeof(*mach));
mach->ShaderType = shader_type;
if (shader_type != PIPE_SHADER_COMPUTE) {
mach->Inputs = align_malloc(sizeof(struct tgsi_exec_vector) * PIPE_MAX_SHADER_INPUTS, 16);
mach->Outputs = align_malloc(sizeof(struct tgsi_exec_vector) * PIPE_MAX_SHADER_OUTPUTS, 16);
if (!mach->Inputs || !mach->Outputs)
goto fail;
}
if (shader_type == PIPE_SHADER_FRAGMENT) {
mach->InputSampleOffsetApply = align_malloc(sizeof(apply_sample_offset_func) * PIPE_MAX_SHADER_INPUTS, 16);
if (!mach->InputSampleOffsetApply)
goto fail;
}
#ifdef DEBUG
/* silence warnings */
(void) print_chan;
(void) print_temp;
#endif
return mach;
fail:
if (mach) {
align_free(mach->InputSampleOffsetApply);
align_free(mach->Inputs);
align_free(mach->Outputs);
align_free(mach);
}
return NULL;
}
void
tgsi_exec_machine_destroy(struct tgsi_exec_machine *mach)
{
if (mach) {
FREE(mach->Instructions);
FREE(mach->Declarations);
FREE(mach->Imms);
align_free(mach->InputSampleOffsetApply);
align_free(mach->Inputs);
align_free(mach->Outputs);
align_free(mach);
}
}
static void
micro_add(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src0,
const union tgsi_exec_channel *src1)
{
dst->f[0] = src0->f[0] + src1->f[0];
dst->f[1] = src0->f[1] + src1->f[1];
dst->f[2] = src0->f[2] + src1->f[2];
dst->f[3] = src0->f[3] + src1->f[3];
}
static void
micro_div(
union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src0,
const union tgsi_exec_channel *src1 )
{
dst->f[0] = src0->f[0] / src1->f[0];
dst->f[1] = src0->f[1] / src1->f[1];
dst->f[2] = src0->f[2] / src1->f[2];
dst->f[3] = src0->f[3] / src1->f[3];
}
static void
micro_lt(
union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src0,
const union tgsi_exec_channel *src1,
const union tgsi_exec_channel *src2,
const union tgsi_exec_channel *src3 )
{
dst->f[0] = src0->f[0] < src1->f[0] ? src2->f[0] : src3->f[0];
dst->f[1] = src0->f[1] < src1->f[1] ? src2->f[1] : src3->f[1];
dst->f[2] = src0->f[2] < src1->f[2] ? src2->f[2] : src3->f[2];
dst->f[3] = src0->f[3] < src1->f[3] ? src2->f[3] : src3->f[3];
}
static void
micro_max(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src0,
const union tgsi_exec_channel *src1)
{
dst->f[0] = fmaxf(src0->f[0], src1->f[0]);
dst->f[1] = fmaxf(src0->f[1], src1->f[1]);
dst->f[2] = fmaxf(src0->f[2], src1->f[2]);
dst->f[3] = fmaxf(src0->f[3], src1->f[3]);
}
static void
micro_min(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src0,
const union tgsi_exec_channel *src1)
{
dst->f[0] = fminf(src0->f[0], src1->f[0]);
dst->f[1] = fminf(src0->f[1], src1->f[1]);
dst->f[2] = fminf(src0->f[2], src1->f[2]);
dst->f[3] = fminf(src0->f[3], src1->f[3]);
}
static void
micro_mul(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src0,
const union tgsi_exec_channel *src1)
{
dst->f[0] = src0->f[0] * src1->f[0];
dst->f[1] = src0->f[1] * src1->f[1];
dst->f[2] = src0->f[2] * src1->f[2];
dst->f[3] = src0->f[3] * src1->f[3];
}
static void
micro_neg(
union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src )
{
dst->f[0] = -src->f[0];
dst->f[1] = -src->f[1];
dst->f[2] = -src->f[2];
dst->f[3] = -src->f[3];
}
static void
micro_pow(
union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src0,
const union tgsi_exec_channel *src1 )
{
dst->f[0] = powf( src0->f[0], src1->f[0] );
dst->f[1] = powf( src0->f[1], src1->f[1] );
dst->f[2] = powf( src0->f[2], src1->f[2] );
dst->f[3] = powf( src0->f[3], src1->f[3] );
}
static void
micro_ldexp(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src0,
const union tgsi_exec_channel *src1)
{
dst->f[0] = ldexpf(src0->f[0], src1->i[0]);
dst->f[1] = ldexpf(src0->f[1], src1->i[1]);
dst->f[2] = ldexpf(src0->f[2], src1->i[2]);
dst->f[3] = ldexpf(src0->f[3], src1->i[3]);
}
static void
micro_sub(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src0,
const union tgsi_exec_channel *src1)
{
dst->f[0] = src0->f[0] - src1->f[0];
dst->f[1] = src0->f[1] - src1->f[1];
dst->f[2] = src0->f[2] - src1->f[2];
dst->f[3] = src0->f[3] - src1->f[3];
}
static void
fetch_src_file_channel(const struct tgsi_exec_machine *mach,
const uint file,
const uint swizzle,
const union tgsi_exec_channel *index,
const union tgsi_exec_channel *index2D,
union tgsi_exec_channel *chan)
{
uint i;
assert(swizzle < 4);
switch (file) {
case TGSI_FILE_CONSTANT:
for (i = 0; i < TGSI_QUAD_SIZE; i++) {
/* NOTE: copying the const value as a uint instead of float */
const uint constbuf = index2D->i[i];
const unsigned pos = index->i[i] * 4 + swizzle;
/* const buffer bounds check */
if (pos >= mach->ConstsSize[constbuf] / 4) {
if (0) {
/* Debug: print warning */
static int count = 0;
if (count++ < 100)
debug_printf("TGSI Exec: const buffer index %d"
" out of bounds\n", pos);
}
chan->u[i] = 0;
} else {
const uint *buf = (const uint *)mach->Consts[constbuf];
chan->u[i] = buf[pos];
}
}
break;
case TGSI_FILE_INPUT:
for (i = 0; i < TGSI_QUAD_SIZE; i++) {
/*
if (PIPE_SHADER_GEOMETRY == mach->ShaderType) {
debug_printf("Fetching Input[%d] (2d=%d, 1d=%d)\n",
index2D->i[i] * TGSI_EXEC_MAX_INPUT_ATTRIBS + index->i[i],
index2D->i[i], index->i[i]);
}*/
int pos = index2D->i[i] * TGSI_EXEC_MAX_INPUT_ATTRIBS + index->i[i];
assert(pos >= 0);
assert(pos < TGSI_MAX_PRIM_VERTICES * PIPE_MAX_ATTRIBS);
chan->u[i] = mach->Inputs[pos].xyzw[swizzle].u[i];
}
break;
case TGSI_FILE_SYSTEM_VALUE:
for (i = 0; i < TGSI_QUAD_SIZE; i++) {
chan->u[i] = mach->SystemValue[index->i[i]].xyzw[swizzle].u[i];
}
break;
case TGSI_FILE_TEMPORARY:
for (i = 0; i < TGSI_QUAD_SIZE; i++) {
assert(index->i[i] < TGSI_EXEC_NUM_TEMPS);
assert(index2D->i[i] == 0);
chan->u[i] = mach->Temps[index->i[i]].xyzw[swizzle].u[i];
}
break;
case TGSI_FILE_IMMEDIATE:
for (i = 0; i < TGSI_QUAD_SIZE; i++) {
assert(index->i[i] >= 0 && index->i[i] < (int)mach->ImmLimit);
assert(index2D->i[i] == 0);
chan->f[i] = mach->Imms[index->i[i]][swizzle];
}
break;
case TGSI_FILE_ADDRESS:
for (i = 0; i < TGSI_QUAD_SIZE; i++) {
assert(index->i[i] >= 0 && index->i[i] < ARRAY_SIZE(mach->Addrs));
assert(index2D->i[i] == 0);
chan->u[i] = mach->Addrs[index->i[i]].xyzw[swizzle].u[i];
}
break;
case TGSI_FILE_OUTPUT:
/* vertex/fragment output vars can be read too */
for (i = 0; i < TGSI_QUAD_SIZE; i++) {
assert(index->i[i] >= 0);
assert(index2D->i[i] == 0);
chan->u[i] = mach->Outputs[index->i[i]].xyzw[swizzle].u[i];
}
break;
default:
assert(0);
for (i = 0; i < TGSI_QUAD_SIZE; i++) {
chan->u[i] = 0;
}
}
}
static void
get_index_registers(const struct tgsi_exec_machine *mach,
const struct tgsi_full_src_register *reg,
union tgsi_exec_channel *index,
union tgsi_exec_channel *index2D)
{
/* We start with a direct index into a register file.
*
* file[1],
* where:
* file = Register.File
* [1] = Register.Index
*/
index->i[0] =
index->i[1] =
index->i[2] =
index->i[3] = reg->Register.Index;
/* There is an extra source register that indirectly subscripts
* a register file. The direct index now becomes an offset
* that is being added to the indirect register.
*
* file[ind[2].x+1],
* where:
* ind = Indirect.File
* [2] = Indirect.Index
* .x = Indirect.SwizzleX
*/
if (reg->Register.Indirect) {
const uint execmask = mach->ExecMask;
assert(reg->Indirect.File == TGSI_FILE_ADDRESS);
const union tgsi_exec_channel *addr = &mach->Addrs[reg->Indirect.Index].xyzw[reg->Indirect.Swizzle];
for (int i = 0; i < TGSI_QUAD_SIZE; i++)
index->i[i] += addr->u[i];
/* for disabled execution channels, zero-out the index to
* avoid using a potential garbage value.
*/
for (int i = 0; i < TGSI_QUAD_SIZE; i++) {
if ((execmask & (1 << i)) == 0)
index->i[i] = 0;
}
}
/* There is an extra source register that is a second
* subscript to a register file. Effectively it means that
* the register file is actually a 2D array of registers.
*
* file[3][1],
* where:
* [3] = Dimension.Index
*/
if (reg->Register.Dimension) {
index2D->i[0] =
index2D->i[1] =
index2D->i[2] =
index2D->i[3] = reg->Dimension.Index;
/* Again, the second subscript index can be addressed indirectly
* identically to the first one.
* Nothing stops us from indirectly addressing the indirect register,
* but there is no need for that, so we won't exercise it.
*
* file[ind[4].y+3][1],
* where:
* ind = DimIndirect.File
* [4] = DimIndirect.Index
* .y = DimIndirect.SwizzleX
*/
if (reg->Dimension.Indirect) {
const uint execmask = mach->ExecMask;
assert(reg->DimIndirect.File == TGSI_FILE_ADDRESS);
const union tgsi_exec_channel *addr = &mach->Addrs[reg->DimIndirect.Index].xyzw[reg->DimIndirect.Swizzle];
for (int i = 0; i < TGSI_QUAD_SIZE; i++)
index2D->i[i] += addr->u[i];
/* for disabled execution channels, zero-out the index to
* avoid using a potential garbage value.
*/
for (int i = 0; i < TGSI_QUAD_SIZE; i++) {
if ((execmask & (1 << i)) == 0) {
index2D->i[i] = 0;
}
}
}
/* If by any chance there was a need for a 3D array of register
* files, we would have to check whether Dimension is followed
* by a dimension register and continue the saga.
*/
} else {
index2D->i[0] =
index2D->i[1] =
index2D->i[2] =
index2D->i[3] = 0;
}
}
static void
fetch_source_d(const struct tgsi_exec_machine *mach,
union tgsi_exec_channel *chan,
const struct tgsi_full_src_register *reg,
const uint chan_index)
{
union tgsi_exec_channel index;
union tgsi_exec_channel index2D;
uint swizzle;
get_index_registers(mach, reg, &index, &index2D);
swizzle = tgsi_util_get_full_src_register_swizzle( reg, chan_index );
fetch_src_file_channel(mach,
reg->Register.File,
swizzle,
&index,
&index2D,
chan);
}
static void
fetch_source(const struct tgsi_exec_machine *mach,
union tgsi_exec_channel *chan,
const struct tgsi_full_src_register *reg,
const uint chan_index,
enum tgsi_exec_datatype src_datatype)
{
fetch_source_d(mach, chan, reg, chan_index);
if (reg->Register.Absolute) {
assert(src_datatype == TGSI_EXEC_DATA_FLOAT);
micro_abs(chan, chan);
}
if (reg->Register.Negate) {
if (src_datatype == TGSI_EXEC_DATA_FLOAT) {
micro_neg(chan, chan);
} else {
micro_ineg(chan, chan);
}
}
}
static union tgsi_exec_channel *
store_dest_dstret(struct tgsi_exec_machine *mach,
const union tgsi_exec_channel *chan,
const struct tgsi_full_dst_register *reg,
uint chan_index)
{
static union tgsi_exec_channel null;
union tgsi_exec_channel *dst;
int offset = 0; /* indirection offset */
int index;
/* There is an extra source register that indirectly subscripts
* a register file. The direct index now becomes an offset
* that is being added to the indirect register.
*
* file[ind[2].x+1],
* where:
* ind = Indirect.File
* [2] = Indirect.Index
* .x = Indirect.SwizzleX
*/
if (reg->Register.Indirect) {
union tgsi_exec_channel index;
union tgsi_exec_channel indir_index;
uint swizzle;
/* which address register (always zero for now) */
index.i[0] =
index.i[1] =
index.i[2] =
index.i[3] = reg->Indirect.Index;
/* get current value of address register[swizzle] */
swizzle = reg->Indirect.Swizzle;
/* fetch values from the address/indirection register */
fetch_src_file_channel(mach,
reg->Indirect.File,
swizzle,
&index,
&ZeroVec,
&indir_index);
/* save indirection offset */
offset = indir_index.i[0];
}
switch (reg->Register.File) {
case TGSI_FILE_NULL:
dst = &null;
break;
case TGSI_FILE_OUTPUT:
index = mach->OutputVertexOffset + reg->Register.Index;
dst = &mach->Outputs[offset + index].xyzw[chan_index];
#if 0
debug_printf("NumOutputs = %d, TEMP_O_C/I = %d, redindex = %d\n",
mach->NumOutputs, mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0],
reg->Register.Index);
if (PIPE_SHADER_GEOMETRY == mach->ShaderType) {
debug_printf("STORING OUT[%d] mask(%d), = (", offset + index, execmask);
for (i = 0; i < TGSI_QUAD_SIZE; i++)
if (execmask & (1 << i))
debug_printf("%f, ", chan->f[i]);
debug_printf(")\n");
}
#endif
break;
case TGSI_FILE_TEMPORARY:
index = reg->Register.Index;
assert( index < TGSI_EXEC_NUM_TEMPS );
dst = &mach->Temps[offset + index].xyzw[chan_index];
break;
case TGSI_FILE_ADDRESS:
index = reg->Register.Index;
assert(index >= 0 && index < ARRAY_SIZE(mach->Addrs));
dst = &mach->Addrs[index].xyzw[chan_index];
break;
default:
unreachable("Bad destination file");
}
return dst;
}
static void
store_dest_double(struct tgsi_exec_machine *mach,
const union tgsi_exec_channel *chan,
const struct tgsi_full_dst_register *reg,
uint chan_index)
{
union tgsi_exec_channel *dst;
const uint execmask = mach->ExecMask;
int i;
dst = store_dest_dstret(mach, chan, reg, chan_index);
if (!dst)
return;
/* doubles path */
for (i = 0; i < TGSI_QUAD_SIZE; i++)
if (execmask & (1 << i))
dst->i[i] = chan->i[i];
}
static void
store_dest(struct tgsi_exec_machine *mach,
const union tgsi_exec_channel *chan,
const struct tgsi_full_dst_register *reg,
const struct tgsi_full_instruction *inst,
uint chan_index)
{
union tgsi_exec_channel *dst;
const uint execmask = mach->ExecMask;
int i;
dst = store_dest_dstret(mach, chan, reg, chan_index);
if (!dst)
return;
if (!inst->Instruction.Saturate) {
for (i = 0; i < TGSI_QUAD_SIZE; i++)
if (execmask & (1 << i))
dst->i[i] = chan->i[i];
}
else {
for (i = 0; i < TGSI_QUAD_SIZE; i++)
if (execmask & (1 << i))
dst->f[i] = fminf(fmaxf(chan->f[i], 0.0f), 1.0f);
}
}
#define FETCH(VAL,INDEX,CHAN)\
fetch_source(mach, VAL, &inst->Src[INDEX], CHAN, TGSI_EXEC_DATA_FLOAT)
#define IFETCH(VAL,INDEX,CHAN)\
fetch_source(mach, VAL, &inst->Src[INDEX], CHAN, TGSI_EXEC_DATA_INT)
/**
* Execute ARB-style KIL which is predicated by a src register.
* Kill fragment if any of the four values is less than zero.
*/
static void
exec_kill_if(struct tgsi_exec_machine *mach,
const struct tgsi_full_instruction *inst)
{
uint uniquemask;
uint chan_index;
uint kilmask = 0; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */
union tgsi_exec_channel r[1];
/* This mask stores component bits that were already tested. */
uniquemask = 0;
for (chan_index = 0; chan_index < 4; chan_index++)
{
uint swizzle;
uint i;
/* unswizzle channel */
swizzle = tgsi_util_get_full_src_register_swizzle (
&inst->Src[0],
chan_index);
/* check if the component has not been already tested */
if (uniquemask & (1 << swizzle))
continue;
uniquemask |= 1 << swizzle;
FETCH(&r[0], 0, chan_index);
for (i = 0; i < 4; i++)
if (r[0].f[i] < 0.0f)
kilmask |= 1 << i;
}
/* restrict to fragments currently executing */
kilmask &= mach->ExecMask;
mach->KillMask |= kilmask;
}
/**
* Unconditional fragment kill/discard.
*/
static void
exec_kill(struct tgsi_exec_machine *mach)
{
/* kill fragment for all fragments currently executing.
* bit 0 = pixel 0, bit 1 = pixel 1, etc.
*/
mach->KillMask |= mach->ExecMask;
}
static void
emit_vertex(struct tgsi_exec_machine *mach,
const struct tgsi_full_instruction *inst)
{
union tgsi_exec_channel r[1];
unsigned stream_id;
unsigned prim_count;
/* FIXME: check for exec mask correctly
unsigned i;
for (i = 0; i < TGSI_QUAD_SIZE; ++i) {
if ((mach->ExecMask & (1 << i)))
*/
IFETCH(&r[0], 0, TGSI_CHAN_X);
stream_id = r[0].u[0];
prim_count = mach->OutputPrimCount[stream_id];
if (mach->ExecMask) {
if (mach->Primitives[stream_id][prim_count] >= mach->MaxOutputVertices)
return;
if (mach->Primitives[stream_id][prim_count] == 0)
mach->PrimitiveOffsets[stream_id][prim_count] = mach->OutputVertexOffset;
mach->OutputVertexOffset += mach->NumOutputs;
mach->Primitives[stream_id][prim_count]++;
}
}
static void
emit_primitive(struct tgsi_exec_machine *mach,
const struct tgsi_full_instruction *inst)
{
unsigned *prim_count;
union tgsi_exec_channel r[1];
unsigned stream_id = 0;
/* FIXME: check for exec mask correctly
unsigned i;
for (i = 0; i < TGSI_QUAD_SIZE; ++i) {
if ((mach->ExecMask & (1 << i)))
*/
if (inst) {
IFETCH(&r[0], 0, TGSI_CHAN_X);
stream_id = r[0].u[0];
}
prim_count = &mach->OutputPrimCount[stream_id];
if (mach->ExecMask) {
++(*prim_count);
assert((*prim_count * mach->NumOutputs) < TGSI_MAX_TOTAL_VERTICES);
mach->Primitives[stream_id][*prim_count] = 0;
}
}
static void
conditional_emit_primitive(struct tgsi_exec_machine *mach)
{
if (PIPE_SHADER_GEOMETRY == mach->ShaderType) {
int emitted_verts = mach->Primitives[0][mach->OutputPrimCount[0]];
if (emitted_verts) {
emit_primitive(mach, NULL);
}
}
}
/*
* Fetch four texture samples using STR texture coordinates.
*/
static void
fetch_texel( struct tgsi_sampler *sampler,
const unsigned sview_idx,
const unsigned sampler_idx,
const union tgsi_exec_channel *s,
const union tgsi_exec_channel *t,
const union tgsi_exec_channel *p,
const union tgsi_exec_channel *c0,
const union tgsi_exec_channel *c1,
float derivs[3][2][TGSI_QUAD_SIZE],
const int8_t offset[3],
enum tgsi_sampler_control control,
union tgsi_exec_channel *r,
union tgsi_exec_channel *g,
union tgsi_exec_channel *b,
union tgsi_exec_channel *a )
{
uint j;
float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
/* FIXME: handle explicit derivs, offsets */
sampler->get_samples(sampler, sview_idx, sampler_idx,
s->f, t->f, p->f, c0->f, c1->f, derivs, offset, control, rgba);
for (j = 0; j < 4; j++) {
r->f[j] = rgba[0][j];
g->f[j] = rgba[1][j];
b->f[j] = rgba[2][j];
a->f[j] = rgba[3][j];
}
}
#define TEX_MODIFIER_NONE 0
#define TEX_MODIFIER_PROJECTED 1
#define TEX_MODIFIER_LOD_BIAS 2
#define TEX_MODIFIER_EXPLICIT_LOD 3
#define TEX_MODIFIER_LEVEL_ZERO 4
#define TEX_MODIFIER_GATHER 5
/*
* Fetch all 3 (for s,t,r coords) texel offsets, put them into int array.
*/
static void
fetch_texel_offsets(struct tgsi_exec_machine *mach,
const struct tgsi_full_instruction *inst,
int8_t offsets[3])
{
if (inst->Texture.NumOffsets == 1) {
union tgsi_exec_channel index;
union tgsi_exec_channel offset[3];
index.i[0] = index.i[1] = index.i[2] = index.i[3] = inst->TexOffsets[0].Index;
fetch_src_file_channel(mach, inst->TexOffsets[0].File,
inst->TexOffsets[0].SwizzleX, &index, &ZeroVec, &offset[0]);
fetch_src_file_channel(mach, inst->TexOffsets[0].File,
inst->TexOffsets[0].SwizzleY, &index, &ZeroVec, &offset[1]);
fetch_src_file_channel(mach, inst->TexOffsets[0].File,
inst->TexOffsets[0].SwizzleZ, &index, &ZeroVec, &offset[2]);
offsets[0] = offset[0].i[0];
offsets[1] = offset[1].i[0];
offsets[2] = offset[2].i[0];
} else {
assert(inst->Texture.NumOffsets == 0);
offsets[0] = offsets[1] = offsets[2] = 0;
}
}
/*
* Fetch dx and dy values for one channel (s, t or r).
* Put dx values into one float array, dy values into another.
*/
static void
fetch_assign_deriv_channel(struct tgsi_exec_machine *mach,
const struct tgsi_full_instruction *inst,
unsigned regdsrcx,
unsigned chan,
float derivs[2][TGSI_QUAD_SIZE])
{
union tgsi_exec_channel d;
FETCH(&d, regdsrcx, chan);
derivs[0][0] = d.f[0];
derivs[0][1] = d.f[1];
derivs[0][2] = d.f[2];
derivs[0][3] = d.f[3];
FETCH(&d, regdsrcx + 1, chan);
derivs[1][0] = d.f[0];
derivs[1][1] = d.f[1];
derivs[1][2] = d.f[2];
derivs[1][3] = d.f[3];
}
static uint
fetch_sampler_unit(struct tgsi_exec_machine *mach,
const struct tgsi_full_instruction *inst,
uint sampler)
{
uint unit = 0;
int i;
if (inst->Src[sampler].Register.Indirect) {
const struct tgsi_full_src_register *reg = &inst->Src[sampler];
union tgsi_exec_channel indir_index, index2;
const uint execmask = mach->ExecMask;
index2.i[0] =
index2.i[1] =
index2.i[2] =
index2.i[3] = reg->Indirect.Index;
fetch_src_file_channel(mach,
reg->Indirect.File,
reg->Indirect.Swizzle,
&index2,
&ZeroVec,
&indir_index);
for (i = 0; i < TGSI_QUAD_SIZE; i++) {
if (execmask & (1 << i)) {
unit = inst->Src[sampler].Register.Index + indir_index.i[i];
break;
}
}
} else {
unit = inst->Src[sampler].Register.Index;
}
return unit;
}
/*
* execute a texture instruction.
*
* modifier is used to control the channel routing for the
* instruction variants like proj, lod, and texture with lod bias.
* sampler indicates which src register the sampler is contained in.
*/
static void
exec_tex(struct tgsi_exec_machine *mach,
const struct tgsi_full_instruction *inst,
uint modifier, uint sampler)
{
const union tgsi_exec_channel *args[5], *proj = NULL;
union tgsi_exec_channel r[5];
enum tgsi_sampler_control control = TGSI_SAMPLER_LOD_NONE;
uint chan;
uint unit;
int8_t offsets[3];
int dim, shadow_ref, i;
unit = fetch_sampler_unit(mach, inst, sampler);
/* always fetch all 3 offsets, overkill but keeps code simple */
fetch_texel_offsets(mach, inst, offsets);
assert(modifier != TEX_MODIFIER_LEVEL_ZERO);
assert(inst->Texture.Texture != TGSI_TEXTURE_BUFFER);
dim = tgsi_util_get_texture_coord_dim(inst->Texture.Texture);
shadow_ref = tgsi_util_get_shadow_ref_src_index(inst->Texture.Texture);
assert(dim <= 4);
if (shadow_ref >= 0)
assert(shadow_ref >= dim && shadow_ref < (int)ARRAY_SIZE(args));
/* fetch modifier to the last argument */
if (modifier != TEX_MODIFIER_NONE) {
const int last = ARRAY_SIZE(args) - 1;
/* fetch modifier from src0.w or src1.x */
if (sampler == 1) {
assert(dim <= TGSI_CHAN_W && shadow_ref != TGSI_CHAN_W);
FETCH(&r[last], 0, TGSI_CHAN_W);
}
else {
FETCH(&r[last], 1, TGSI_CHAN_X);
}
if (modifier != TEX_MODIFIER_PROJECTED) {
args[last] = &r[last];
}
else {
proj = &r[last];
args[last] = &ZeroVec;
}
/* point unused arguments to zero vector */
for (i = dim; i < last; i++)
args[i] = &ZeroVec;
if (modifier == TEX_MODIFIER_EXPLICIT_LOD)
control = TGSI_SAMPLER_LOD_EXPLICIT;
else if (modifier == TEX_MODIFIER_LOD_BIAS)
control = TGSI_SAMPLER_LOD_BIAS;
else if (modifier == TEX_MODIFIER_GATHER)
control = TGSI_SAMPLER_GATHER;
}
else {
for (i = dim; i < (int)ARRAY_SIZE(args); i++)
args[i] = &ZeroVec;
}
/* fetch coordinates */
for (i = 0; i < dim; i++) {
FETCH(&r[i], 0, TGSI_CHAN_X + i);
if (proj)
micro_div(&r[i], &r[i], proj);
args[i] = &r[i];
}
/* fetch reference value */
if (shadow_ref >= 0) {
FETCH(&r[shadow_ref], shadow_ref / 4, TGSI_CHAN_X + (shadow_ref % 4));
if (proj)
micro_div(&r[shadow_ref], &r[shadow_ref], proj);
args[shadow_ref] = &r[shadow_ref];
}
fetch_texel(mach->Sampler, unit, unit,
args[0], args[1], args[2], args[3], args[4],
NULL, offsets, control,
&r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */
#if 0
debug_printf("fetch r: %g %g %g %g\n",
r[0].f[0], r[0].f[1], r[0].f[2], r[0].f[3]);
debug_printf("fetch g: %g %g %g %g\n",
r[1].f[0], r[1].f[1], r[1].f[2], r[1].f[3]);
debug_printf("fetch b: %g %g %g %g\n",
r[2].f[0], r[2].f[1], r[2].f[2], r[2].f[3]);
debug_printf("fetch a: %g %g %g %g\n",
r[3].f[0], r[3].f[1], r[3].f[2], r[3].f[3]);
#endif
for (chan = 0; chan < TGSI_NUM_CHANNELS; chan++) {
if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
store_dest(mach, &r[chan], &inst->Dst[0], inst, chan);
}
}
}
static void
exec_lodq(struct tgsi_exec_machine *mach,
const struct tgsi_full_instruction *inst)
{
uint resource_unit, sampler_unit;
unsigned dim;
unsigned i;
union tgsi_exec_channel coords[4];
const union tgsi_exec_channel *args[ARRAY_SIZE(coords)];
union tgsi_exec_channel r[2];
resource_unit = fetch_sampler_unit(mach, inst, 1);
if (inst->Instruction.Opcode == TGSI_OPCODE_LOD) {
uint target = mach->SamplerViews[resource_unit].Resource;
dim = tgsi_util_get_texture_coord_dim(target);
sampler_unit = fetch_sampler_unit(mach, inst, 2);
} else {
dim = tgsi_util_get_texture_coord_dim(inst->Texture.Texture);
sampler_unit = resource_unit;
}
assert(dim <= ARRAY_SIZE(coords));
/* fetch coordinates */
for (i = 0; i < dim; i++) {
FETCH(&coords[i], 0, TGSI_CHAN_X + i);
args[i] = &coords[i];
}
for (i = dim; i < ARRAY_SIZE(coords); i++) {
args[i] = &ZeroVec;
}
mach->Sampler->query_lod(mach->Sampler, resource_unit, sampler_unit,
args[0]->f,
args[1]->f,
args[2]->f,
args[3]->f,
TGSI_SAMPLER_LOD_NONE,
r[0].f,
r[1].f);
if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
store_dest(mach, &r[0], &inst->Dst[0], inst, TGSI_CHAN_X);
}
if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
store_dest(mach, &r[1], &inst->Dst[0], inst, TGSI_CHAN_Y);
}
if (inst->Instruction.Opcode == TGSI_OPCODE_LOD) {
unsigned char swizzles[4];
unsigned chan;
swizzles[0] = inst->Src[1].Register.SwizzleX;
swizzles[1] = inst->Src[1].Register.SwizzleY;
swizzles[2] = inst->Src[1].Register.SwizzleZ;
swizzles[3] = inst->Src[1].Register.SwizzleW;
for (chan = 0; chan < TGSI_NUM_CHANNELS; chan++) {
if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
if (swizzles[chan] >= 2) {
store_dest(mach, &ZeroVec,
&inst->Dst[0], inst, chan);
} else {
store_dest(mach, &r[swizzles[chan]],
&inst->Dst[0], inst, chan);
}
}
}
} else {
if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
store_dest(mach, &r[0], &inst->Dst[0], inst, TGSI_CHAN_X);
}
if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
store_dest(mach, &r[1], &inst->Dst[0], inst, TGSI_CHAN_Y);
}
}
}
static void
exec_txd(struct tgsi_exec_machine *mach,
const struct tgsi_full_instruction *inst)
{
union tgsi_exec_channel r[4];
float derivs[3][2][TGSI_QUAD_SIZE];
uint chan;
uint unit;
int8_t offsets[3];
unit = fetch_sampler_unit(mach, inst, 3);
/* always fetch all 3 offsets, overkill but keeps code simple */
fetch_texel_offsets(mach, inst, offsets);
switch (inst->Texture.Texture) {
case TGSI_TEXTURE_1D:
FETCH(&r[0], 0, TGSI_CHAN_X);
fetch_assign_deriv_channel(mach, inst, 1, TGSI_CHAN_X, derivs[0]);
fetch_texel(mach->Sampler, unit, unit,
&r[0], &ZeroVec, &ZeroVec, &ZeroVec, &ZeroVec, /* S, T, P, C, LOD */
derivs, offsets, TGSI_SAMPLER_DERIVS_EXPLICIT,
&r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */
break;
case TGSI_TEXTURE_SHADOW1D:
case TGSI_TEXTURE_1D_ARRAY:
case TGSI_TEXTURE_SHADOW1D_ARRAY:
/* SHADOW1D/1D_ARRAY would not need Y/Z respectively, but don't bother */
FETCH(&r[0], 0, TGSI_CHAN_X);
FETCH(&r[1], 0, TGSI_CHAN_Y);
FETCH(&r[2], 0, TGSI_CHAN_Z);
fetch_assign_deriv_channel(mach, inst, 1, TGSI_CHAN_X, derivs[0]);
fetch_texel(mach->Sampler, unit, unit,
&r[0], &r[1], &r[2], &ZeroVec, &ZeroVec, /* S, T, P, C, LOD */
derivs, offsets, TGSI_SAMPLER_DERIVS_EXPLICIT,
&r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */
break;
case TGSI_TEXTURE_2D:
case TGSI_TEXTURE_RECT:
FETCH(&r[0], 0, TGSI_CHAN_X);
FETCH(&r[1], 0, TGSI_CHAN_Y);
fetch_assign_deriv_channel(mach, inst, 1, TGSI_CHAN_X, derivs[0]);
fetch_assign_deriv_channel(mach, inst, 1, TGSI_CHAN_Y, derivs[1]);
fetch_texel(mach->Sampler, unit, unit,
&r[0], &r[1], &ZeroVec, &ZeroVec, &ZeroVec, /* S, T, P, C, LOD */
derivs, offsets, TGSI_SAMPLER_DERIVS_EXPLICIT,
&r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */
break;
case TGSI_TEXTURE_SHADOW2D:
case TGSI_TEXTURE_SHADOWRECT:
case TGSI_TEXTURE_2D_ARRAY:
case TGSI_TEXTURE_SHADOW2D_ARRAY:
/* only SHADOW2D_ARRAY actually needs W */
FETCH(&r[0], 0, TGSI_CHAN_X);
FETCH(&r[1], 0, TGSI_CHAN_Y);
FETCH(&r[2], 0, TGSI_CHAN_Z);
FETCH(&r[3], 0, TGSI_CHAN_W);
fetch_assign_deriv_channel(mach, inst, 1, TGSI_CHAN_X, derivs[0]);
fetch_assign_deriv_channel(mach, inst, 1, TGSI_CHAN_Y, derivs[1]);
fetch_texel(mach->Sampler, unit, unit,
&r[0], &r[1], &r[2], &r[3], &ZeroVec, /* inputs */
derivs, offsets, TGSI_SAMPLER_DERIVS_EXPLICIT,
&r[0], &r[1], &r[2], &r[3]); /* outputs */
break;
case TGSI_TEXTURE_3D:
case TGSI_TEXTURE_CUBE:
case TGSI_TEXTURE_CUBE_ARRAY:
case TGSI_TEXTURE_SHADOWCUBE:
/* only TEXTURE_CUBE_ARRAY and TEXTURE_SHADOWCUBE actually need W */
FETCH(&r[0], 0, TGSI_CHAN_X);
FETCH(&r[1], 0, TGSI_CHAN_Y);
FETCH(&r[2], 0, TGSI_CHAN_Z);
FETCH(&r[3], 0, TGSI_CHAN_W);
fetch_assign_deriv_channel(mach, inst, 1, TGSI_CHAN_X, derivs[0]);
fetch_assign_deriv_channel(mach, inst, 1, TGSI_CHAN_Y, derivs[1]);
fetch_assign_deriv_channel(mach, inst, 1, TGSI_CHAN_Z, derivs[2]);
fetch_texel(mach->Sampler, unit, unit,
&r[0], &r[1], &r[2], &r[3], &ZeroVec, /* inputs */
derivs, offsets, TGSI_SAMPLER_DERIVS_EXPLICIT,
&r[0], &r[1], &r[2], &r[3]); /* outputs */
break;
default:
assert(0);
}
for (chan = 0; chan < TGSI_NUM_CHANNELS; chan++) {
if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
store_dest(mach, &r[chan], &inst->Dst[0], inst, chan);
}
}
}
static void
exec_txf(struct tgsi_exec_machine *mach,
const struct tgsi_full_instruction *inst)
{
union tgsi_exec_channel r[4];
uint chan;
uint unit;
float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
int j;
int8_t offsets[3];
unsigned target;
unit = fetch_sampler_unit(mach, inst, 1);
/* always fetch all 3 offsets, overkill but keeps code simple */
fetch_texel_offsets(mach, inst, offsets);
IFETCH(&r[3], 0, TGSI_CHAN_W);
if (inst->Instruction.Opcode == TGSI_OPCODE_SAMPLE_I ||
inst->Instruction.Opcode == TGSI_OPCODE_SAMPLE_I_MS) {
target = mach->SamplerViews[unit].Resource;
}
else {
target = inst->Texture.Texture;
}
switch(target) {
case TGSI_TEXTURE_3D:
case TGSI_TEXTURE_2D_ARRAY:
case TGSI_TEXTURE_SHADOW2D_ARRAY:
case TGSI_TEXTURE_2D_ARRAY_MSAA:
IFETCH(&r[2], 0, TGSI_CHAN_Z);
FALLTHROUGH;
case TGSI_TEXTURE_2D:
case TGSI_TEXTURE_RECT:
case TGSI_TEXTURE_SHADOW1D_ARRAY:
case TGSI_TEXTURE_SHADOW2D:
case TGSI_TEXTURE_SHADOWRECT:
case TGSI_TEXTURE_1D_ARRAY:
case TGSI_TEXTURE_2D_MSAA:
IFETCH(&r[1], 0, TGSI_CHAN_Y);
FALLTHROUGH;
case TGSI_TEXTURE_BUFFER:
case TGSI_TEXTURE_1D:
case TGSI_TEXTURE_SHADOW1D:
IFETCH(&r[0], 0, TGSI_CHAN_X);
break;
default:
assert(0);
break;
}
mach->Sampler->get_texel(mach->Sampler, unit, r[0].i, r[1].i, r[2].i, r[3].i,
offsets, rgba);
for (j = 0; j < TGSI_QUAD_SIZE; j++) {
r[0].f[j] = rgba[0][j];
r[1].f[j] = rgba[1][j];
r[2].f[j] = rgba[2][j];
r[3].f[j] = rgba[3][j];
}
if (inst->Instruction.Opcode == TGSI_OPCODE_SAMPLE_I ||
inst->Instruction.Opcode == TGSI_OPCODE_SAMPLE_I_MS) {
unsigned char swizzles[4];
swizzles[0] = inst->Src[1].Register.SwizzleX;
swizzles[1] = inst->Src[1].Register.SwizzleY;
swizzles[2] = inst->Src[1].Register.SwizzleZ;
swizzles[3] = inst->Src[1].Register.SwizzleW;
for (chan = 0; chan < TGSI_NUM_CHANNELS; chan++) {
if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
store_dest(mach, &r[swizzles[chan]],
&inst->Dst[0], inst, chan);
}
}
}
else {
for (chan = 0; chan < TGSI_NUM_CHANNELS; chan++) {
if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
store_dest(mach, &r[chan], &inst->Dst[0], inst, chan);
}
}
}
}
static void
exec_txq(struct tgsi_exec_machine *mach,
const struct tgsi_full_instruction *inst)
{
int result[4];