void
I810SubsequentScreenToScreenCopy(ScrnInfoPtr pScrn, int x1, int y1,
int x2, int y2, int w, int h)
{
I810Ptr pI810 = I810PTR(pScrn);
int src, dst;
int w_back = w;
if (I810_DEBUG & DEBUG_VERBOSE_ACCEL)
ErrorF( "I810SubsequentScreenToScreenCopy %d,%d - %d,%d %dx%d\n",
x1,y1,x2,y2,w,h);
/*
* This works around a bug in the i810 drawing engine.
* This was developed empirically so it may not catch all
* cases.
*/
#define I810_MWIDTH 8
if ( !(pI810->BR[13] & BR13_RIGHT_TO_LEFT) && (y2 - y1) < 3
&& (y2 - y1) >= 0 && (x2 - x1) <= (w + I810_MWIDTH)
&& (w > I810_MWIDTH))
w = I810_MWIDTH;
do {
if (pI810->BR[13] & BR13_PITCH_SIGN_BIT) {
src = (y1 + h - 1) * pScrn->displayWidth * pI810->cpp;
dst = (y2 + h - 1) * pScrn->displayWidth * pI810->cpp;
} else {
src = y1 * pScrn->displayWidth * pI810->cpp;
dst = y2 * pScrn->displayWidth * pI810->cpp;
}
if (pI810->BR[13] & BR13_RIGHT_TO_LEFT) {
src += (x1 + w - 1) * pI810->cpp + pI810->cpp - 1;
dst += (x2 + w - 1) * pI810->cpp + pI810->cpp - 1;
} else {
src += x1 * pI810->cpp;
dst += x2 * pI810->cpp;
}
/* SRC_COPY_BLT, p169 */
{
BEGIN_LP_RING(6);
OUT_RING( BR00_BITBLT_CLIENT | BR00_OP_SRC_COPY_BLT | 0x4 );
OUT_RING( pI810->BR[13]);
OUT_RING( (h << 16) | (w * pI810->cpp));
OUT_RING( pI810->bufferOffset + dst);
OUT_RING( pI810->BR[13] & 0xFFFF);
OUT_RING( pI810->bufferOffset + src);
ADVANCE_LP_RING();
}
w_back -= w;
if (w_back <= 0)
break;
x2 += w;
x1 += w;
if (w_back > I810_MWIDTH)
w = I810_MWIDTH;
else
w = w_back;
} while (1);
}
static inline void
out_dstpix(struct fd_ringbuffer *ring, PixmapPtr pix)
{
struct fd_bo *bo = pix->bo;
uint32_t w, h, p;
w = pix->width;
h = pix->height;
/* pitch specified in units of 32 bytes, it appears.. not quite sure
* max size yet, but I think 11 or 12 bits..
*/
p = (pix->pitch / 32) & 0xfff;
/* not quite sure if these first three dwords belong here, but all
* blits seem to start with these immediately before the dst surf
* parameters, so I'm putting them here for now
*
* Note that there are some similar dwords preceding src surf state
* although it varies slightly for composite (some extra bits set
* for src surface and no 0x11000000 like dword for mask surface..
* so this may need some shuffling around when I start playing with
* emitting dst/src/mask surf state in the corresponding Prepare
* fxns rather than for every blit..
*/
OUT_RING (ring, REG(G2D_ALPHABLEND) | 0x0);
OUT_RING (ring, REG(G2D_BLENDERCFG) | 0x0);
OUT_RING (ring, REG(G2D_GRADIENT) | 0x030000);
OUT_RING (ring, REG(GRADW_TEXSIZE) | ((h & 0xfff) << 13) | (w & 0xfff));
OUT_RING (ring, REG(G2D_CFG0) | p |
((pix->depth == 8) ? 0xe000 : 0x7000));
OUT_RING (ring, REGM(G2D_BASE0, 1));
OUT_RELOC(ring, bo);
OUT_RING (ring, REGM(GRADW_TEXBASE, 1));
OUT_RELOC(ring, bo);
OUT_RING (ring, REGM(GRADW_TEXCFG, 1));
OUT_RING (ring, 0x40000000 | p |
((pix->depth == 8) ? 0xe000 : 0x7000));
OUT_RING (ring, 0xd5000000);
OUT_RING (ring, REG(G2D_ALPHABLEND) | 0x0);
OUT_RING (ring, REG(G2D_SCISSORX) | (w & 0xfff) << 12);
OUT_RING (ring, REG(G2D_SCISSORY) | (h & 0xfff) << 12);
}
static void copy(PixmapPtr dest, PixmapPtr src, int srcX, int srcY,
int dstX, int dstY, int width, int height)
{
BEGIN_RING(45);
out_dstpix(ring, dest);
OUT_RING (ring, REGM(G2D_FOREGROUND, 2));
OUT_RING (ring, 0xff000000); /* G2D_FOREGROUND */
OUT_RING (ring, 0xff000000); /* G2D_BACKGROUND */
OUT_RING (ring, REG(G2D_BLENDERCFG) | 0x0);
OUT_RING (ring, 0xd0000000);
out_srcpix(ring, src);
OUT_RING (ring, 0xd5000000);
OUT_RING (ring, 0xd0000000);
OUT_RING (ring, REG(G2D_INPUT) | iena(G2D_INPUT_SCOORD1));
OUT_RING (ring, REG(G2D_INPUT) | iena(0));
OUT_RING (ring, REG(G2D_INPUT) | idis(G2D_INPUT_COLOR));
OUT_RING (ring, REG(G2D_INPUT) | iena(G2D_INPUT_COPYCOORD));
OUT_RING (ring, 0xd0000000);
OUT_RING (ring, REG(G2D_INPUT) | iena(0));
OUT_RING (ring, REG(G2D_INPUT) | iena(0));
OUT_RING (ring, REG(G2D_INPUT) | iena(0));
OUT_RING (ring, REG(G2D_CONFIG) | G2D_CONFIG_SRC1); /* we don't read from dst */
OUT_RING (ring, REGM(G2D_XY, 3));
OUT_RING (ring, (dstX & 0xffff) << 16 | (dstY & 0xffff)); /* G2D_XY */
OUT_RING (ring, (width & 0xfff) << 16 | (height & 0xffff)); /* G2D_WIDTHHEIGHT */
OUT_RING (ring, (srcX & 0xffff) << 16 | (srcY & 0xffff)); /* G2D_SXY */
OUT_RING (ring, 0xd0000000);
OUT_RING (ring, 0xd0000000);
OUT_RING (ring, 0xd0000000);
OUT_RING (ring, 0xd0000000);
OUT_RING (ring, 0xd0000000);
OUT_RING (ring, 0xd0000000);
END_RING ();
}
static void
nouveau_accel_init(struct nouveau_drm *drm)
{
struct nouveau_device *device = nv_device(drm->device);
struct nouveau_object *object;
u32 arg0, arg1;
int ret;
if (nouveau_noaccel || !nouveau_fifo(device) /*XXX*/)
return;
/* initialise synchronisation routines */
if (device->card_type < NV_10) ret = nv04_fence_create(drm);
else if (device->card_type < NV_11 ||
device->chipset < 0x17) ret = nv10_fence_create(drm);
else if (device->card_type < NV_50) ret = nv17_fence_create(drm);
else if (device->chipset < 0x84) ret = nv50_fence_create(drm);
else if (device->card_type < NV_C0) ret = nv84_fence_create(drm);
else ret = nvc0_fence_create(drm);
if (ret) {
NV_ERROR(drm, "failed to initialise sync subsystem, %d\n", ret);
nouveau_accel_fini(drm);
return;
}
if (device->card_type >= NV_E0) {
ret = nouveau_channel_new(drm, &drm->client, NVDRM_DEVICE,
NVDRM_CHAN + 1,
NVE0_CHANNEL_IND_ENGINE_CE0 |
NVE0_CHANNEL_IND_ENGINE_CE1, 0,
&drm->cechan);
if (ret)
NV_ERROR(drm, "failed to create ce channel, %d\n", ret);
arg0 = NVE0_CHANNEL_IND_ENGINE_GR;
arg1 = 1;
} else
if (device->chipset >= 0xa3 &&
device->chipset != 0xaa &&
device->chipset != 0xac) {
ret = nouveau_channel_new(drm, &drm->client, NVDRM_DEVICE,
NVDRM_CHAN + 1, NvDmaFB, NvDmaTT,
&drm->cechan);
if (ret)
NV_ERROR(drm, "failed to create ce channel, %d\n", ret);
arg0 = NvDmaFB;
arg1 = NvDmaTT;
} else {
arg0 = NvDmaFB;
arg1 = NvDmaTT;
}
ret = nouveau_channel_new(drm, &drm->client, NVDRM_DEVICE, NVDRM_CHAN,
arg0, arg1, &drm->channel);
if (ret) {
NV_ERROR(drm, "failed to create kernel channel, %d\n", ret);
nouveau_accel_fini(drm);
return;
}
ret = nouveau_object_new(nv_object(drm), NVDRM_CHAN, NVDRM_NVSW,
nouveau_abi16_swclass(drm), NULL, 0, &object);
if (ret == 0) {
struct nouveau_software_chan *swch = (void *)object->parent;
ret = RING_SPACE(drm->channel, 2);
if (ret == 0) {
if (device->card_type < NV_C0) {
BEGIN_NV04(drm->channel, NvSubSw, 0, 1);
OUT_RING (drm->channel, NVDRM_NVSW);
} else
if (device->card_type < NV_E0) {
BEGIN_NVC0(drm->channel, FermiSw, 0, 1);
OUT_RING (drm->channel, 0x001f0000);
}
}
swch = (void *)object->parent;
swch->flip = nouveau_flip_complete;
swch->flip_data = drm->channel;
}
if (ret) {
NV_ERROR(drm, "failed to allocate software object, %d\n", ret);
nouveau_accel_fini(drm);
return;
}
if (device->card_type < NV_C0) {
ret = nouveau_gpuobj_new(drm->device, NULL, 32, 0, 0,
&drm->notify);
if (ret) {
NV_ERROR(drm, "failed to allocate notifier, %d\n", ret);
nouveau_accel_fini(drm);
return;
}
ret = nouveau_object_new(nv_object(drm),
drm->channel->handle, NvNotify0,
0x003d, &(struct nv_dma_class) {
.flags = NV_DMA_TARGET_VRAM |
NV_DMA_ACCESS_RDWR,
.start = drm->notify->addr,
.limit = drm->notify->addr + 31
}, sizeof(struct nv_dma_class),
void fd_program_emit_state(struct fd_program *program, uint32_t first,
struct fd_parameters *uniforms, struct fd_parameters *attr,
struct fd_ringbuffer *ring)
{
struct fd_shader *vs = get_shader(program, FD_SHADER_VERTEX);
struct fd_shader *fs = get_shader(program, FD_SHADER_FRAGMENT);
struct ir3_shader_info *vsi = &vs->info;
struct ir3_shader_info *fsi = &fs->info;
uint32_t vsconstlen = constlen(vs);
uint32_t fsconstlen = constlen(fs);
uint32_t i, outloc;
uint32_t posregid = getpos(vs, "gl_Position", 0);
uint32_t psizeregid = getpos(vs, "gl_PointSize", (63 << 2));
uint32_t colorregid = getpos(fs, "gl_FragColor", 0);
uint32_t numvar = totalvar(fs);
assert (vs->ir->varyings_count == fs->ir->varyings_count);
OUT_PKT0(ring, REG_A3XX_HLSQ_CONTROL_0_REG, 6);
OUT_RING(ring, A3XX_HLSQ_CONTROL_0_REG_FSTHREADSIZE(FOUR_QUADS) |
A3XX_HLSQ_CONTROL_0_REG_SPSHADERRESTART |
A3XX_HLSQ_CONTROL_0_REG_SPCONSTFULLUPDATE);
OUT_RING(ring, A3XX_HLSQ_CONTROL_1_REG_VSTHREADSIZE(TWO_QUADS) |
A3XX_HLSQ_CONTROL_1_REG_VSSUPERTHREADENABLE);
OUT_RING(ring, A3XX_HLSQ_CONTROL_2_REG_PRIMALLOCTHRESHOLD(31));
OUT_RING(ring, 0x00000000); /* HLSQ_CONTROL_3_REG */
OUT_RING(ring, A3XX_HLSQ_VS_CONTROL_REG_CONSTLENGTH(vsconstlen) |
A3XX_HLSQ_VS_CONTROL_REG_CONSTSTARTOFFSET(0) |
A3XX_HLSQ_VS_CONTROL_REG_INSTRLENGTH(instrlen(vs)));
OUT_RING(ring, A3XX_HLSQ_FS_CONTROL_REG_CONSTLENGTH(fsconstlen) |
A3XX_HLSQ_FS_CONTROL_REG_CONSTSTARTOFFSET(128) |
A3XX_HLSQ_FS_CONTROL_REG_INSTRLENGTH(instrlen(fs)));
OUT_PKT0(ring, REG_A3XX_SP_SP_CTRL_REG, 1);
OUT_RING(ring, A3XX_SP_SP_CTRL_REG_CONSTMODE(0) |
A3XX_SP_SP_CTRL_REG_SLEEPMODE(1) |
// XXX "resolve" (?) bit set on gmem->mem pass..
COND(!uniforms, A3XX_SP_SP_CTRL_REG_RESOLVE) |
// XXX sometimes 0, sometimes 1:
A3XX_SP_SP_CTRL_REG_LOMODE(1));
/* emit unknown sequence of writes to 0x0ec4/0x0ec8 that the blob
* emits as part of the program state (it seems)..
*/
for (i = 0; i < 6; i++) {
OUT_PKT0(ring, REG_A3XX_SP_PERFCOUNTER0_SELECT, 1);
OUT_RING(ring, 0x00000000); /* SP_PERFCOUNTER0_SELECT */
OUT_PKT0(ring, REG_A3XX_SP_PERFCOUNTER3_SELECT, 1);
OUT_RING(ring, 0x00000000); /* SP_PERFCOUNTER3_SELECT */
}
OUT_PKT0(ring, REG_A3XX_SP_VS_LENGTH_REG, 1);
OUT_RING(ring, A3XX_SP_VS_LENGTH_REG_SHADERLENGTH(instrlen(vs)));
OUT_PKT0(ring, REG_A3XX_SP_VS_CTRL_REG0, 3);
OUT_RING(ring, A3XX_SP_VS_CTRL_REG0_THREADMODE(MULTI) |
A3XX_SP_VS_CTRL_REG0_INSTRBUFFERMODE(BUFFER) |
A3XX_SP_VS_CTRL_REG0_HALFREGFOOTPRINT(vsi->max_half_reg + 1) |
A3XX_SP_VS_CTRL_REG0_FULLREGFOOTPRINT(vsi->max_reg + 1) |
A3XX_SP_VS_CTRL_REG0_INOUTREGOVERLAP(0) |
A3XX_SP_VS_CTRL_REG0_THREADSIZE(TWO_QUADS) |
A3XX_SP_VS_CTRL_REG0_SUPERTHREADMODE |
A3XX_SP_VS_CTRL_REG0_LENGTH(instrlen(vs)));
OUT_RING(ring, A3XX_SP_VS_CTRL_REG1_CONSTLENGTH(vsconstlen) |
A3XX_SP_VS_CTRL_REG1_INITIALOUTSTANDING(totalattr(vs)) |
A3XX_SP_VS_CTRL_REG1_CONSTFOOTPRINT(max(vsi->max_const, 0)));
OUT_RING(ring, A3XX_SP_VS_PARAM_REG_POSREGID(posregid) |
A3XX_SP_VS_PARAM_REG_PSIZEREGID(psizeregid) |
A3XX_SP_VS_PARAM_REG_TOTALVSOUTVAR(fs->ir->varyings_count));
for (i = 0; i < vs->ir->varyings_count; ) {
struct ir3_varying *v;
uint32_t reg = 0;
OUT_PKT0(ring, REG_A3XX_SP_VS_OUT_REG(i/2), 1);
v = vs->ir->varyings[i++];
if (v) {
reg |= A3XX_SP_VS_OUT_REG_A_REGID(v->rstart->num);
reg |= A3XX_SP_VS_OUT_REG_A_COMPMASK(regmask(v->num));
}
v = vs->ir->varyings[i++];
if (v) {
reg |= A3XX_SP_VS_OUT_REG_B_REGID(v->rstart->num);
reg |= A3XX_SP_VS_OUT_REG_B_COMPMASK(regmask(v->num));
}
OUT_RING(ring, reg);
}
outloc = 8; /* I assume 0 and 4 are gl_Position/gl_PointSize? */
for (i = 0; i < vs->ir->varyings_count; ) {
struct ir3_varying *v;
uint32_t reg = 0;
OUT_PKT0(ring, REG_A3XX_SP_VS_VPC_DST_REG(i/4), 1);
/* note: if we supported anything other than vec4 varyings, we'd
* actually be incrementing outloc by the actual varying size in
* units of scalar registers (ie. vec3 -> 3)
*/
v = vs->ir->varyings[i++];
if (v) {
reg |= A3XX_SP_VS_VPC_DST_REG_OUTLOC0(outloc);
outloc += v->num;
}
v = vs->ir->varyings[i++];
if (v) {
reg |= A3XX_SP_VS_VPC_DST_REG_OUTLOC1(outloc);
outloc += v->num;
}
v = vs->ir->varyings[i++];
if (v) {
reg |= A3XX_SP_VS_VPC_DST_REG_OUTLOC2(outloc);
outloc += v->num;
}
v = vs->ir->varyings[i++];
if (v) {
reg |= A3XX_SP_VS_VPC_DST_REG_OUTLOC3(outloc);
outloc += v->num;
}
OUT_RING(ring, reg);
}
// TODO SP_VS_OBJ_OFFSET_REG / SP_VS_OBJ_START_REG
OUT_PKT0(ring, REG_A3XX_SP_FS_LENGTH_REG, 1);
OUT_RING(ring, A3XX_SP_FS_LENGTH_REG_SHADERLENGTH(instrlen(fs)));
OUT_PKT0(ring, REG_A3XX_SP_FS_CTRL_REG0, 2);
OUT_RING(ring, A3XX_SP_FS_CTRL_REG0_THREADMODE(MULTI) |
A3XX_SP_FS_CTRL_REG0_INSTRBUFFERMODE(BUFFER) |
A3XX_SP_FS_CTRL_REG0_HALFREGFOOTPRINT(fsi->max_half_reg + 1) |
A3XX_SP_FS_CTRL_REG0_FULLREGFOOTPRINT(fsi->max_reg + 1) |
A3XX_SP_FS_CTRL_REG0_INOUTREGOVERLAP(1) |
A3XX_SP_FS_CTRL_REG0_THREADSIZE(FOUR_QUADS) |
A3XX_SP_FS_CTRL_REG0_SUPERTHREADMODE |
COND(fs->ir->samplers_count > 0, A3XX_SP_FS_CTRL_REG0_PIXLODENABLE) |
A3XX_SP_FS_CTRL_REG0_LENGTH(instrlen(fs)));
OUT_RING(ring, A3XX_SP_FS_CTRL_REG1_CONSTLENGTH(fsconstlen) |
A3XX_SP_FS_CTRL_REG1_INITIALOUTSTANDING(0) |
A3XX_SP_FS_CTRL_REG1_CONSTFOOTPRINT(max(fsi->max_const, 0)) |
A3XX_SP_FS_CTRL_REG1_HALFPRECVAROFFSET(63));
// TODO SP_FS_OBJ_OFFSET_REG / SP_FS_OBJ_START_REG
OUT_PKT0(ring, REG_A3XX_SP_FS_FLAT_SHAD_MODE_REG_0, 2);
OUT_RING(ring, 0x00000000); /* SP_FS_FLAT_SHAD_MODE_REG_0 */
OUT_RING(ring, 0x00000000); /* SP_FS_FLAT_SHAD_MODE_REG_1 */
OUT_PKT0(ring, REG_A3XX_SP_FS_OUTPUT_REG, 1);
OUT_RING(ring, 0x00000000); /* SP_FS_OUTPUT_REG */
OUT_PKT0(ring, REG_A3XX_SP_FS_MRT_REG(0), 4);
OUT_RING(ring, A3XX_SP_FS_MRT_REG_REGID(colorregid) | /* SP_FS_MRT[0].REG */
A3XX_SP_FS_MRT_REG_HALF_PRECISION);
OUT_RING(ring, A3XX_SP_FS_MRT_REG_REGID(0)); /* SP_FS_MRT[1].REG */
OUT_RING(ring, A3XX_SP_FS_MRT_REG_REGID(0)); /* SP_FS_MRT[2].REG */
OUT_RING(ring, A3XX_SP_FS_MRT_REG_REGID(0)); /* SP_FS_MRT[3].REG */
OUT_PKT0(ring, REG_A3XX_VPC_ATTR, 2);
OUT_RING(ring, A3XX_VPC_ATTR_TOTALATTR(numvar) |
A3XX_VPC_ATTR_THRDASSIGN(1) |
A3XX_VPC_ATTR_LMSIZE(1));
OUT_RING(ring, A3XX_VPC_PACK_NUMFPNONPOSVAR(numvar) |
A3XX_VPC_PACK_NUMNONPOSVSVAR(numvar));
OUT_PKT0(ring, REG_A3XX_VPC_VARYING_INTERP_MODE(0), 4);
OUT_RING(ring, 0x00000000); /* VPC_VARYING_INTERP[0].MODE */
OUT_RING(ring, 0x00000000); /* VPC_VARYING_INTERP[1].MODE */
OUT_RING(ring, 0x00000000); /* VPC_VARYING_INTERP[2].MODE */
OUT_RING(ring, 0x00000000); /* VPC_VARYING_INTERP[3].MODE */
OUT_PKT0(ring, REG_A3XX_VPC_VARYING_PS_REPL_MODE(0), 4);
OUT_RING(ring, 0x00000000); /* VPC_VARYING_PS_REPL[0].MODE */
OUT_RING(ring, 0x00000000); /* VPC_VARYING_PS_REPL[1].MODE */
OUT_RING(ring, 0x00000000); /* VPC_VARYING_PS_REPL[2].MODE */
OUT_RING(ring, 0x00000000); /* VPC_VARYING_PS_REPL[3].MODE */
OUT_PKT0(ring, REG_A3XX_VFD_VS_THREADING_THRESHOLD, 1);
OUT_RING(ring, A3XX_VFD_VS_THREADING_THRESHOLD_REGID_THRESHOLD(15) |
A3XX_VFD_VS_THREADING_THRESHOLD_REGID_VTXCNT(252));
emit_shader(ring, vs, SB_VERT_SHADER);
OUT_PKT0(ring, REG_A3XX_VFD_PERFCOUNTER0_SELECT, 1);
OUT_RING(ring, 0x00000000); /* VFD_PERFCOUNTER0_SELECT */
emit_shader(ring, fs, SB_FRAG_SHADER);
OUT_PKT0(ring, REG_A3XX_VFD_PERFCOUNTER0_SELECT, 1);
OUT_RING(ring, 0x00000000); /* VFD_PERFCOUNTER0_SELECT */
OUT_PKT0(ring, REG_A3XX_VFD_CONTROL_0, 2);
OUT_RING(ring, A3XX_VFD_CONTROL_0_TOTALATTRTOVS(totalattr(vs)) |
A3XX_VFD_CONTROL_0_PACKETSIZE(2) |
A3XX_VFD_CONTROL_0_STRMDECINSTRCNT(vs->ir->attributes_count) |
A3XX_VFD_CONTROL_0_STRMFETCHINSTRCNT(vs->ir->attributes_count));
OUT_RING(ring, A3XX_VFD_CONTROL_1_MAXSTORAGE(1) | // XXX
A3XX_VFD_CONTROL_1_REGID4VTX(63 << 2) |
A3XX_VFD_CONTROL_1_REGID4INST(63 << 2));
emit_vtx_fetch(ring, vs, attr, first);
/* we have this sometimes, not others.. perhaps we could be clever
* and figure out actually when we need to invalidate cache:
*/
OUT_PKT0(ring, REG_A3XX_UCHE_CACHE_INVALIDATE0_REG, 2);
OUT_RING(ring, A3XX_UCHE_CACHE_INVALIDATE0_REG_ADDR(0));
OUT_RING(ring, A3XX_UCHE_CACHE_INVALIDATE1_REG_ADDR(0) |
A3XX_UCHE_CACHE_INVALIDATE1_REG_OPCODE(INVALIDATE) |
A3XX_UCHE_CACHE_INVALIDATE1_REG_ENTIRE_CACHE);
/* for RB_RESOLVE_PASS, I think the consts are not needed: */
if (uniforms) {
emit_uniconst(ring, vs, uniforms, SB_VERT_SHADER);
emit_uniconst(ring, fs, uniforms, SB_FRAG_SHADER);
}
}
static boolean
nv30_vertprog_validate(struct nv30_context *nv30)
{
struct pipe_screen *pscreen = nv30->pipe.screen;
struct nouveau_grobj *rankine = nv30->screen->rankine;
struct nv30_vertex_program *vp;
struct pipe_buffer *constbuf;
boolean upload_code = FALSE, upload_data = FALSE;
int i;
vp = nv30->vertprog;
constbuf = nv30->constbuf[PIPE_SHADER_VERTEX];
/* Translate TGSI shader into hw bytecode */
if (!vp->translated) {
nv30_vertprog_translate(nv30, vp);
if (!vp->translated)
return FALSE;
}
/* Allocate hw vtxprog exec slots */
if (!vp->exec) {
struct nouveau_resource *heap = nv30->screen->vp_exec_heap;
struct nouveau_stateobj *so;
uint vplen = vp->nr_insns;
if (nouveau_resource_alloc(heap, vplen, vp, &vp->exec)) {
while (heap->next && heap->size < vplen) {
struct nv30_vertex_program *evict;
evict = heap->next->priv;
nouveau_resource_free(&evict->exec);
}
if (nouveau_resource_alloc(heap, vplen, vp, &vp->exec))
assert(0);
}
so = so_new(2, 0);
so_method(so, rankine, NV34TCL_VP_START_FROM_ID, 1);
so_data (so, vp->exec->start);
so_ref(so, &vp->so);
so_ref(NULL, &so);
upload_code = TRUE;
}
/* Allocate hw vtxprog const slots */
if (vp->nr_consts && !vp->data) {
struct nouveau_resource *heap = nv30->screen->vp_data_heap;
if (nouveau_resource_alloc(heap, vp->nr_consts, vp, &vp->data)) {
while (heap->next && heap->size < vp->nr_consts) {
struct nv30_vertex_program *evict;
evict = heap->next->priv;
nouveau_resource_free(&evict->data);
}
if (nouveau_resource_alloc(heap, vp->nr_consts, vp,
&vp->data))
assert(0);
}
/*XXX: handle this some day */
assert(vp->data->start >= vp->data_start_min);
upload_data = TRUE;
if (vp->data_start != vp->data->start)
upload_code = TRUE;
}
/* If exec or data segments moved we need to patch the program to
* fixup offsets and register IDs.
*/
if (vp->exec_start != vp->exec->start) {
for (i = 0; i < vp->nr_insns; i++) {
struct nv30_vertex_program_exec *vpi = &vp->insns[i];
if (vpi->has_branch_offset) {
assert(0);
}
}
vp->exec_start = vp->exec->start;
}
if (vp->nr_consts && vp->data_start != vp->data->start) {
for (i = 0; i < vp->nr_insns; i++) {
struct nv30_vertex_program_exec *vpi = &vp->insns[i];
if (vpi->const_index >= 0) {
vpi->data[1] &= ~NV30_VP_INST_CONST_SRC_MASK;
vpi->data[1] |=
(vpi->const_index + vp->data->start) <<
NV30_VP_INST_CONST_SRC_SHIFT;
}
}
vp->data_start = vp->data->start;
}
/* Update + Upload constant values */
if (vp->nr_consts) {
float *map = NULL;
if (constbuf) {
map = pipe_buffer_map(pscreen, constbuf,
PIPE_BUFFER_USAGE_CPU_READ);
}
for (i = 0; i < vp->nr_consts; i++) {
struct nv30_vertex_program_data *vpd = &vp->consts[i];
if (vpd->index >= 0) {
if (!upload_data &&
!memcmp(vpd->value, &map[vpd->index * 4],
4 * sizeof(float)))
continue;
memcpy(vpd->value, &map[vpd->index * 4],
4 * sizeof(float));
}
BEGIN_RING(rankine, NV34TCL_VP_UPLOAD_CONST_ID, 5);
OUT_RING (i + vp->data->start);
OUT_RINGp ((uint32_t *)vpd->value, 4);
}
if (constbuf)
pipe_buffer_unmap(pscreen, constbuf);
}
/* Upload vtxprog */
if (upload_code) {
#if 0
for (i = 0; i < vp->nr_insns; i++) {
NOUVEAU_MSG("VP inst %d: 0x%08x 0x%08x 0x%08x 0x%08x\n",
i, vp->insns[i].data[0], vp->insns[i].data[1],
vp->insns[i].data[2], vp->insns[i].data[3]);
}
#endif
BEGIN_RING(rankine, NV34TCL_VP_UPLOAD_FROM_ID, 1);
OUT_RING (vp->exec->start);
for (i = 0; i < vp->nr_insns; i++) {
BEGIN_RING(rankine, NV34TCL_VP_UPLOAD_INST(0), 4);
OUT_RINGp (vp->insns[i].data, 4);
}
}
if (vp->so != nv30->state.hw[NV30_STATE_VERTPROG]) {
so_ref(vp->so, &nv30->state.hw[NV30_STATE_VERTPROG]);
return TRUE;
}
return FALSE;
}
void
nvfx_push_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
{
struct nvfx_context *nvfx = nvfx_context(pipe);
struct nouveau_channel *chan = nvfx->screen->base.channel;
struct nouveau_grobj *eng3d = nvfx->screen->eng3d;
struct push_context ctx;
struct util_split_prim s;
unsigned instances_left = info->instance_count;
int vtx_value;
unsigned hw_mode = nvgl_primitive(info->mode);
int i;
struct
{
uint8_t* map;
unsigned step;
} per_instance[16];
unsigned p_overhead = 64 /* magic fix */
+ 4 /* begin/end */
+ 4; /* potential edgeflag enable/disable */
ctx.chan = nvfx->screen->base.channel;
ctx.eng3d = nvfx->screen->eng3d;
ctx.translate = nvfx->vtxelt->translate;
ctx.idxbuf = NULL;
ctx.vertex_length = nvfx->vtxelt->vertex_length;
ctx.max_vertices_per_packet = nvfx->vtxelt->max_vertices_per_packet;
ctx.edgeflag = 0.5f;
// TODO: figure out if we really want to handle this, and do so in that case
ctx.edgeflag_attr = 0xff; // nvfx->vertprog->cfg.edgeflag_in;
if(!nvfx->use_vertex_buffers)
{
for(i = 0; i < nvfx->vtxelt->num_per_vertex_buffer_infos; ++i)
{
struct nvfx_per_vertex_buffer_info* vbi = &nvfx->vtxelt->per_vertex_buffer_info[i];
struct pipe_vertex_buffer *vb = &nvfx->vtxbuf[vbi->vertex_buffer_index];
uint8_t* data = nvfx_buffer(vb->buffer)->data + vb->buffer_offset;
if(info->indexed)
data += info->index_bias * vb->stride;
ctx.translate->set_buffer(ctx.translate, i, data, vb->stride, ~0);
}
if(ctx.edgeflag_attr < 16)
vtx_value = -(ctx.vertex_length + 3); /* vertex data and edgeflag header and value */
else
{
p_overhead += 1; /* initial vertex_data header */
vtx_value = -ctx.vertex_length; /* vertex data and edgeflag header and value */
}
if (info->indexed) {
// XXX: this case and is broken and probably need a new VTX_ATTR push path
if (nvfx->idxbuf.index_size == 1)
s.emit = emit_vertices_lookup8;
else if (nvfx->idxbuf.index_size == 2)
s.emit = emit_vertices_lookup16;
else
s.emit = emit_vertices_lookup32;
} else
s.emit = emit_vertices;
}
else
{
if(!info->indexed || nvfx->use_index_buffer)
{
s.emit = info->indexed ? emit_ib_ranges : emit_vb_ranges;
p_overhead += 3;
vtx_value = 0;
}
else if (nvfx->idxbuf.index_size == 4)
{
s.emit = emit_elt32;
p_overhead += 1;
vtx_value = 8;
}
else
{
s.emit = (nvfx->idxbuf.index_size == 2) ? emit_elt16 : emit_elt8;
p_overhead += 3;
vtx_value = 7;
}
}
ctx.idxbias = info->index_bias;
if(nvfx->use_vertex_buffers)
ctx.idxbias -= nvfx->base_vertex;
/* map index buffer, if present */
if (info->indexed && !nvfx->use_index_buffer)
ctx.idxbuf = nvfx_buffer(nvfx->idxbuf.buffer)->data + nvfx->idxbuf.offset;
s.priv = &ctx;
s.edge = emit_edgeflag;
for (i = 0; i < nvfx->vtxelt->num_per_instance; ++i)
{
struct nvfx_per_instance_element *ve = &nvfx->vtxelt->per_instance[i];
struct pipe_vertex_buffer *vb = &nvfx->vtxbuf[ve->base.vertex_buffer_index];
float v[4];
per_instance[i].step = info->start_instance % ve->instance_divisor;
per_instance[i].map = nvfx_buffer(vb->buffer)->data + vb->buffer_offset + ve->base.src_offset;
nvfx->vtxelt->per_instance[i].base.fetch_rgba_float(v, per_instance[i].map, 0, 0);
nvfx_emit_vtx_attr(chan, eng3d,
nvfx->vtxelt->per_instance[i].base.idx, v,
nvfx->vtxelt->per_instance[i].base.ncomp);
}
/* per-instance loop */
while (instances_left--) {
int max_verts;
boolean done;
util_split_prim_init(&s, info->mode, info->start, info->count);
nvfx_state_emit(nvfx);
for(;;) {
max_verts = AVAIL_RING(chan);
max_verts -= p_overhead;
/* if vtx_value < 0, each vertex is -vtx_value words long
* otherwise, each vertex is 2^(vtx_value) / 255 words long (this is an approximation)
*/
if(vtx_value < 0)
{
max_verts /= -vtx_value;
max_verts -= (max_verts >> 10); /* vertex data headers */
}
else
{
if(max_verts >= (1 << 23)) /* avoid overflow here */
max_verts = (1 << 23);
max_verts = (max_verts * 255) >> vtx_value;
}
//printf("avail %u max_verts %u\n", AVAIL_RING(chan), max_verts);
if(max_verts >= 16)
{
/* XXX: any command a lot of times seems to (mostly) fix corruption that would otherwise happen */
/* this seems to cause issues on nv3x, and also be unneeded there */
if(nvfx->is_nv4x)
{
int i;
for(i = 0; i < 32; ++i)
{
BEGIN_RING(chan, eng3d,
0x1dac, 1);
OUT_RING(chan, 0);
}
}
BEGIN_RING(chan, eng3d,
NV30_3D_VERTEX_BEGIN_END, 1);
OUT_RING(chan, hw_mode);
done = util_split_prim_next(&s, max_verts);
BEGIN_RING(chan, eng3d,
NV30_3D_VERTEX_BEGIN_END, 1);
OUT_RING(chan, 0);
if(done)
break;
}
FIRE_RING(chan);
nvfx_state_emit(nvfx);
}
void
NV30EXAComposite(PixmapPtr pdPix, int srcX , int srcY,
int maskX, int maskY,
int dstX , int dstY,
int width, int height)
{
ScrnInfoPtr pScrn = xf86Screens[pdPix->drawable.pScreen->myNum];
NVPtr pNv = NVPTR(pScrn);
struct nouveau_channel *chan = pNv->chan;
struct nouveau_grobj *rankine = pNv->Nv3D;
float sX0, sX1, sX2, sY0, sY1, sY2;
float mX0, mX1, mX2, mY0, mY1, mY2;
NV30EXA_STATE;
WAIT_RING(chan, 64);
/* We're drawing a triangle, we need to scissor it to a quad. */
/* The scissors are here for a good reason, we don't get the full image, but just a part. */
/* Handling the cliprects is done for us already. */
BEGIN_RING(chan, rankine, NV34TCL_SCISSOR_HORIZ, 2);
OUT_RING (chan, (width << 16) | dstX);
OUT_RING (chan, (height << 16) | dstY);
BEGIN_RING(chan, rankine, NV34TCL_VERTEX_BEGIN_END, 1);
OUT_RING (chan, NV34TCL_VERTEX_BEGIN_END_TRIANGLES);
#if 0
ErrorF("Composite [%dx%d] (%d,%d)IN(%d,%d)OP(%d,%d)\n",width,height,srcX,srcY,maskX,maskY,dstX,dstY);
#endif
NV30EXATransformCoord(state->unit[0].transform,
srcX, srcY - height,
state->unit[0].width,
state->unit[0].height, &sX0, &sY0);
NV30EXATransformCoord(state->unit[0].transform,
srcX, srcY + height,
state->unit[0].width,
state->unit[0].height, &sX1, &sY1);
NV30EXATransformCoord(state->unit[0].transform,
srcX + 2*width, srcY + height,
state->unit[0].width,
state->unit[0].height, &sX2, &sY2);
if (state->have_mask) {
NV30EXATransformCoord(state->unit[1].transform,
maskX, maskY - height,
state->unit[1].width,
state->unit[1].height, &mX0, &mY0);
NV30EXATransformCoord(state->unit[1].transform,
maskX, maskY + height,
state->unit[1].width,
state->unit[1].height, &mX1, &mY1);
NV30EXATransformCoord(state->unit[1].transform,
maskX + 2*width, maskY + height,
state->unit[1].width,
state->unit[1].height, &mX2, &mY2);
CV_OUTm(sX0 , sY0 , mX0, mY0, dstX , dstY - height);
CV_OUTm(sX1 , sY1 , mX1, mY1, dstX , dstY + height);
CV_OUTm(sX2 , sY2 , mX2, mY2, dstX + 2*width , dstY + height);
} else {
CV_OUT(sX0 , sY0 , dstX , dstY - height);
CV_OUT(sX1 , sY1 , dstX , dstY + height);
CV_OUT(sX2 , sY2 , dstX + 2*width , dstY + height);
}
BEGIN_RING(chan, rankine, NV34TCL_VERTEX_BEGIN_END, 1);
OUT_RING (chan, 0);
}
void
nvc0_tfb_validate(struct nvc0_context *nvc0)
{
struct nouveau_channel *chan = nvc0->screen->base.channel;
struct nvc0_transform_feedback_state *tfb;
unsigned b, n, i;
if (nvc0->gmtyprog) tfb = nvc0->gmtyprog->tfb;
else
if (nvc0->tevlprog) tfb = nvc0->tevlprog->tfb;
else
tfb = nvc0->vertprog->tfb;
IMMED_RING(chan, RING_3D(TFB_ENABLE), (tfb && nvc0->num_tfbbufs) ? 1 : 0);
if (tfb && tfb != nvc0->state.tfb) {
uint8_t var[128];
for (n = 0, b = 0; b < 4; n += tfb->varying_count[b++]) {
if (tfb->varying_count[b]) {
BEGIN_RING(chan, RING_3D(TFB_STREAM(b)), 3);
OUT_RING (chan, 0);
OUT_RING (chan, tfb->varying_count[b]);
OUT_RING (chan, tfb->stride[b]);
for (i = 0; i < tfb->varying_count[b]; ++i)
var[i] = tfb->varying_index[n + i];
for (; i & 3; ++i)
var[i] = 0; /* zero rest of method word bits */
BEGIN_RING(chan, RING_3D(TFB_VARYING_LOCS(b, 0)), i / 4);
OUT_RINGp (chan, var, i / 4);
if (nvc0->tfbbuf[b])
nvc0_so_target(nvc0->tfbbuf[b])->stride = tfb->stride[b];
} else {
IMMED_RING(chan, RING_3D(TFB_VARYING_COUNT(b)), 0);
}
}
}
nvc0->state.tfb = tfb;
if (!(nvc0->dirty & NVC0_NEW_TFB_TARGETS))
return;
nvc0_bufctx_reset(nvc0, NVC0_BUFCTX_TFB);
for (b = 0; b < nvc0->num_tfbbufs; ++b) {
struct nvc0_so_target *targ = nvc0_so_target(nvc0->tfbbuf[b]);
struct nv04_resource *buf = nv04_resource(targ->pipe.buffer);
if (tfb)
targ->stride = tfb->stride[b];
if (!(nvc0->tfbbuf_dirty & (1 << b)))
continue;
if (!targ->clean)
nvc0_query_fifo_wait(chan, targ->pq);
BEGIN_RING(chan, RING_3D(TFB_BUFFER_ENABLE(b)), 5);
OUT_RING (chan, 1);
OUT_RESRCh(chan, buf, targ->pipe.buffer_offset, NOUVEAU_BO_WR);
OUT_RESRCl(chan, buf, targ->pipe.buffer_offset, NOUVEAU_BO_WR);
OUT_RING (chan, targ->pipe.buffer_size);
if (!targ->clean) {
nvc0_query_pushbuf_submit(chan, targ->pq, 0x4);
} else {
OUT_RING(chan, 0); /* TFB_BUFFER_OFFSET */
targ->clean = FALSE;
}
nvc0_bufctx_add_resident(nvc0, NVC0_BUFCTX_TFB, buf, NOUVEAU_BO_WR);
}
for (; b < 4; ++b)
IMMED_RING(chan, RING_3D(TFB_BUFFER_ENABLE(b)), 0);
}
void
nvc0_m2mf_transfer_rect(struct pipe_screen *pscreen,
const struct nv50_m2mf_rect *dst,
const struct nv50_m2mf_rect *src,
uint32_t nblocksx, uint32_t nblocksy)
{
struct nouveau_channel *chan = nouveau_screen(pscreen)->channel;
const int cpp = dst->cpp;
uint32_t src_ofst = src->base;
uint32_t dst_ofst = dst->base;
uint32_t height = nblocksy;
uint32_t sy = src->y;
uint32_t dy = dst->y;
uint32_t exec = (1 << 20);
assert(dst->cpp == src->cpp);
if (nouveau_bo_tile_layout(src->bo)) {
BEGIN_RING(chan, RING_MF(TILING_MODE_IN), 5);
OUT_RING (chan, src->tile_mode);
OUT_RING (chan, src->width * cpp);
OUT_RING (chan, src->height);
OUT_RING (chan, src->depth);
OUT_RING (chan, src->z);
} else {
src_ofst += src->y * src->pitch + src->x * cpp;
BEGIN_RING(chan, RING_MF(PITCH_IN), 1);
OUT_RING (chan, src->width * cpp);
exec |= NVC0_M2MF_EXEC_LINEAR_IN;
}
if (nouveau_bo_tile_layout(dst->bo)) {
BEGIN_RING(chan, RING_MF(TILING_MODE_OUT), 5);
OUT_RING (chan, dst->tile_mode);
OUT_RING (chan, dst->width * cpp);
OUT_RING (chan, dst->height);
OUT_RING (chan, dst->depth);
OUT_RING (chan, dst->z);
} else {
dst_ofst += dst->y * dst->pitch + dst->x * cpp;
BEGIN_RING(chan, RING_MF(PITCH_OUT), 1);
OUT_RING (chan, dst->width * cpp);
exec |= NVC0_M2MF_EXEC_LINEAR_OUT;
}
while (height) {
int line_count = height > 2047 ? 2047 : height;
MARK_RING (chan, 17, 4);
BEGIN_RING(chan, RING_MF(OFFSET_IN_HIGH), 2);
OUT_RELOCh(chan, src->bo, src_ofst, src->domain | NOUVEAU_BO_RD);
OUT_RELOCl(chan, src->bo, src_ofst, src->domain | NOUVEAU_BO_RD);
BEGIN_RING(chan, RING_MF(OFFSET_OUT_HIGH), 2);
OUT_RELOCh(chan, dst->bo, dst_ofst, dst->domain | NOUVEAU_BO_WR);
OUT_RELOCl(chan, dst->bo, dst_ofst, dst->domain | NOUVEAU_BO_WR);
if (!(exec & NVC0_M2MF_EXEC_LINEAR_IN)) {
BEGIN_RING(chan, RING_MF(TILING_POSITION_IN_X), 2);
OUT_RING (chan, src->x * cpp);
OUT_RING (chan, sy);
} else {
src_ofst += line_count * src->pitch;
}
if (!(exec & NVC0_M2MF_EXEC_LINEAR_OUT)) {
BEGIN_RING(chan, RING_MF(TILING_POSITION_OUT_X), 2);
OUT_RING (chan, dst->x * cpp);
OUT_RING (chan, dy);
} else {
dst_ofst += line_count * dst->pitch;
}
BEGIN_RING(chan, RING_MF(LINE_LENGTH_IN), 2);
OUT_RING (chan, nblocksx * cpp);
OUT_RING (chan, line_count);
BEGIN_RING(chan, RING_MF(EXEC), 1);
OUT_RING (chan, exec);
height -= line_count;
sy += line_count;
dy += line_count;
}
}
Bool
NV30EXAPrepareComposite(int op, PicturePtr psPict,
PicturePtr pmPict,
PicturePtr pdPict,
PixmapPtr psPix,
PixmapPtr pmPix,
PixmapPtr pdPix)
{
ScrnInfoPtr pScrn = xf86Screens[psPix->drawable.pScreen->myNum];
NVPtr pNv = NVPTR(pScrn);
struct nouveau_channel *chan = pNv->chan;
struct nouveau_grobj *rankine = pNv->Nv3D;
nv_pict_op_t *blend;
int fpid = NV30EXA_FPID_PASS_COL0;
NV30EXA_STATE;
if (MARK_RING(chan, 128, 1 + 1 + 4))
return FALSE;
blend = NV30_GetPictOpRec(op);
NV30_SetupBlend(pScrn, blend, pdPict->format,
(pmPict && pmPict->componentAlpha &&
PICT_FORMAT_RGB(pmPict->format)));
if (!NV30_SetupSurface(pScrn, pdPix, pdPict) ||
!NV30EXATexture(pScrn, psPix, psPict, 0)) {
MARK_UNDO(chan);
return FALSE;
}
#if 0
#define printformat(f) ErrorF("(%xh %s %dbpp A%dR%dG%dB%d)",f,(f>>16)&0xf==2?"ARGB":"ABGR",(f>>24),(f&0xf000)>>12,(f&0xf00)>>8,(f&0xf0)>>4,f&0xf)
ErrorF("Preparecomposite src(%dx%d)",psPict->pDrawable->width,psPict->pDrawable->height);
printformat((psPict->format));
ErrorF(" dst(%dx%d)",pdPict->pDrawable->width,pdPict->pDrawable->height);
printformat((pdPict->format));
if (pmPict)
{
ErrorF(" mask(%dx%d)",pmPict->pDrawable->width,pmPict->pDrawable->height);
printformat((pmPict->format));
}
ErrorF("\n");
#endif
if (pmPict) {
if (!NV30EXATexture(pScrn, pmPix, pmPict, 1)) {
MARK_UNDO(chan);
return FALSE;
}
if (pmPict->componentAlpha && PICT_FORMAT_RGB(pmPict->format)) {
if (blend->src_alpha)
fpid = NV30EXA_FPID_COMPOSITE_MASK_SA_CA;
else
fpid = NV30EXA_FPID_COMPOSITE_MASK_CA;
} else {
fpid = NV30EXA_FPID_COMPOSITE_MASK;
}
state->have_mask = TRUE;
} else {
fpid = NV30EXA_FPID_PASS_TEX0;
state->have_mask = FALSE;
}
if (!NV30_LoadFragProg(pScrn, (pdPict->format == PICT_a8) ?
nv40_fp_map_a8[fpid] : nv40_fp_map[fpid])) {
MARK_UNDO(chan);
return FALSE;
}
BEGIN_RING(chan, rankine, 0x23c, 1);
OUT_RING (chan, pmPict?3:1);
pNv->alu = op;
pNv->pspict = psPict;
pNv->pmpict = pmPict;
pNv->pdpict = pdPict;
pNv->pspix = psPix;
pNv->pmpix = pmPix;
pNv->pdpix = pdPix;
chan->flush_notify = NV30EXAStateCompositeReemit;
return TRUE;
}
void
fd5_program_emit(struct fd_ringbuffer *ring, struct fd5_emit *emit,
int nr, struct pipe_surface **bufs)
{
struct stage s[MAX_STAGES];
uint32_t pos_regid, posz_regid, psize_regid, color_regid[8];
uint32_t face_regid, coord_regid, zwcoord_regid;
uint32_t vcoord_regid, vertex_regid, instance_regid;
int i, j;
debug_assert(nr <= ARRAY_SIZE(color_regid));
if (emit->key.binning_pass)
nr = 0;
setup_stages(emit, s);
pos_regid = ir3_find_output_regid(s[VS].v, VARYING_SLOT_POS);
posz_regid = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DEPTH);
psize_regid = ir3_find_output_regid(s[VS].v, VARYING_SLOT_PSIZ);
vertex_regid = ir3_find_output_regid(s[VS].v, SYSTEM_VALUE_VERTEX_ID_ZERO_BASE);
instance_regid = ir3_find_output_regid(s[VS].v, SYSTEM_VALUE_INSTANCE_ID);
if (s[FS].v->color0_mrt) {
color_regid[0] = color_regid[1] = color_regid[2] = color_regid[3] =
color_regid[4] = color_regid[5] = color_regid[6] = color_regid[7] =
ir3_find_output_regid(s[FS].v, FRAG_RESULT_COLOR);
} else {
color_regid[0] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA0);
color_regid[1] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA1);
color_regid[2] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA2);
color_regid[3] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA3);
color_regid[4] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA4);
color_regid[5] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA5);
color_regid[6] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA6);
color_regid[7] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA7);
}
/* TODO get these dynamically: */
face_regid = s[FS].v->frag_face ? regid(0,0) : regid(63,0);
coord_regid = s[FS].v->frag_coord ? regid(0,0) : regid(63,0);
zwcoord_regid = s[FS].v->frag_coord ? regid(0,2) : regid(63,0);
vcoord_regid = (s[FS].v->total_in > 0) ? regid(0,0) : regid(63,0);
/* we could probably divide this up into things that need to be
* emitted if frag-prog is dirty vs if vert-prog is dirty..
*/
OUT_PKT4(ring, REG_A5XX_HLSQ_VS_CONTROL_REG, 5);
OUT_RING(ring, A5XX_HLSQ_VS_CONTROL_REG_CONSTOBJECTOFFSET(s[VS].constoff) |
A5XX_HLSQ_VS_CONTROL_REG_SHADEROBJOFFSET(s[VS].instroff) |
COND(s[VS].v, A5XX_HLSQ_VS_CONTROL_REG_ENABLED));
OUT_RING(ring, A5XX_HLSQ_FS_CONTROL_REG_CONSTOBJECTOFFSET(s[FS].constoff) |
A5XX_HLSQ_FS_CONTROL_REG_SHADEROBJOFFSET(s[FS].instroff) |
COND(s[FS].v, A5XX_HLSQ_FS_CONTROL_REG_ENABLED));
OUT_RING(ring, A5XX_HLSQ_HS_CONTROL_REG_CONSTOBJECTOFFSET(s[HS].constoff) |
A5XX_HLSQ_HS_CONTROL_REG_SHADEROBJOFFSET(s[HS].instroff) |
COND(s[HS].v, A5XX_HLSQ_HS_CONTROL_REG_ENABLED));
OUT_RING(ring, A5XX_HLSQ_DS_CONTROL_REG_CONSTOBJECTOFFSET(s[DS].constoff) |
A5XX_HLSQ_DS_CONTROL_REG_SHADEROBJOFFSET(s[DS].instroff) |
COND(s[DS].v, A5XX_HLSQ_DS_CONTROL_REG_ENABLED));
OUT_RING(ring, A5XX_HLSQ_GS_CONTROL_REG_CONSTOBJECTOFFSET(s[GS].constoff) |
A5XX_HLSQ_GS_CONTROL_REG_SHADEROBJOFFSET(s[GS].instroff) |
COND(s[GS].v, A5XX_HLSQ_GS_CONTROL_REG_ENABLED));
OUT_PKT4(ring, REG_A5XX_HLSQ_CS_CONFIG, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT4(ring, REG_A5XX_HLSQ_VS_CNTL, 5);
OUT_RING(ring, A5XX_HLSQ_VS_CNTL_INSTRLEN(s[VS].instrlen));
OUT_RING(ring, A5XX_HLSQ_FS_CNTL_INSTRLEN(s[FS].instrlen));
OUT_RING(ring, A5XX_HLSQ_HS_CNTL_INSTRLEN(s[HS].instrlen));
OUT_RING(ring, A5XX_HLSQ_DS_CNTL_INSTRLEN(s[DS].instrlen));
OUT_RING(ring, A5XX_HLSQ_GS_CNTL_INSTRLEN(s[GS].instrlen));
OUT_PKT4(ring, REG_A5XX_SP_VS_CONTROL_REG, 5);
OUT_RING(ring, A5XX_SP_VS_CONTROL_REG_CONSTOBJECTOFFSET(s[VS].constoff) |
A5XX_SP_VS_CONTROL_REG_SHADEROBJOFFSET(s[VS].instroff) |
COND(s[VS].v, A5XX_SP_VS_CONTROL_REG_ENABLED));
OUT_RING(ring, A5XX_SP_FS_CONTROL_REG_CONSTOBJECTOFFSET(s[FS].constoff) |
A5XX_SP_FS_CONTROL_REG_SHADEROBJOFFSET(s[FS].instroff) |
COND(s[FS].v, A5XX_SP_FS_CONTROL_REG_ENABLED));
OUT_RING(ring, A5XX_SP_HS_CONTROL_REG_CONSTOBJECTOFFSET(s[HS].constoff) |
A5XX_SP_HS_CONTROL_REG_SHADEROBJOFFSET(s[HS].instroff) |
COND(s[HS].v, A5XX_SP_HS_CONTROL_REG_ENABLED));
OUT_RING(ring, A5XX_SP_DS_CONTROL_REG_CONSTOBJECTOFFSET(s[DS].constoff) |
A5XX_SP_DS_CONTROL_REG_SHADEROBJOFFSET(s[DS].instroff) |
COND(s[DS].v, A5XX_SP_DS_CONTROL_REG_ENABLED));
OUT_RING(ring, A5XX_SP_GS_CONTROL_REG_CONSTOBJECTOFFSET(s[GS].constoff) |
A5XX_SP_GS_CONTROL_REG_SHADEROBJOFFSET(s[GS].instroff) |
COND(s[GS].v, A5XX_SP_GS_CONTROL_REG_ENABLED));
OUT_PKT4(ring, REG_A5XX_SP_CS_CONFIG, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT4(ring, REG_A5XX_HLSQ_VS_CONSTLEN, 2);
OUT_RING(ring, s[VS].constlen); /* HLSQ_VS_CONSTLEN */
OUT_RING(ring, s[VS].instrlen); /* HLSQ_VS_INSTRLEN */
OUT_PKT4(ring, REG_A5XX_HLSQ_FS_CONSTLEN, 2);
OUT_RING(ring, s[FS].constlen); /* HLSQ_FS_CONSTLEN */
OUT_RING(ring, s[FS].instrlen); /* HLSQ_FS_INSTRLEN */
OUT_PKT4(ring, REG_A5XX_HLSQ_HS_CONSTLEN, 2);
OUT_RING(ring, s[HS].constlen); /* HLSQ_HS_CONSTLEN */
OUT_RING(ring, s[HS].instrlen); /* HLSQ_HS_INSTRLEN */
OUT_PKT4(ring, REG_A5XX_HLSQ_DS_CONSTLEN, 2);
OUT_RING(ring, s[DS].constlen); /* HLSQ_DS_CONSTLEN */
OUT_RING(ring, s[DS].instrlen); /* HLSQ_DS_INSTRLEN */
OUT_PKT4(ring, REG_A5XX_HLSQ_GS_CONSTLEN, 2);
OUT_RING(ring, s[GS].constlen); /* HLSQ_GS_CONSTLEN */
OUT_RING(ring, s[GS].instrlen); /* HLSQ_GS_INSTRLEN */
OUT_PKT4(ring, REG_A5XX_HLSQ_CONTEXT_SWITCH_CS_SW_3, 2);
OUT_RING(ring, 0x00000000); /* HLSQ_CONTEXT_SWITCH_CS_SW_3 */
OUT_RING(ring, 0x00000000); /* HLSQ_CONTEXT_SWITCH_CS_SW_4 */
OUT_PKT4(ring, REG_A5XX_SP_VS_CTRL_REG0, 1);
OUT_RING(ring, A5XX_SP_VS_CTRL_REG0_HALFREGFOOTPRINT(s[VS].i->max_half_reg + 1) |
A5XX_SP_VS_CTRL_REG0_FULLREGFOOTPRINT(s[VS].i->max_reg + 1) |
0x6 | /* XXX seems to be always set? */
A5XX_SP_VS_CTRL_REG0_BRANCHSTACK(0x3) | // XXX need to figure this out somehow..
COND(s[VS].v->has_samp, A5XX_SP_VS_CTRL_REG0_PIXLODENABLE));
struct ir3_shader_linkage l = {0};
ir3_link_shaders(&l, s[VS].v, s[FS].v);
BITSET_DECLARE(varbs, 128) = {0};
uint32_t *varmask = (uint32_t *)varbs;
for (i = 0; i < l.cnt; i++)
for (j = 0; j < util_last_bit(l.var[i].compmask); j++)
BITSET_SET(varbs, l.var[i].loc + j);
OUT_PKT4(ring, REG_A5XX_VPC_VAR_DISABLE(0), 4);
OUT_RING(ring, ~varmask[0]); /* VPC_VAR[0].DISABLE */
OUT_RING(ring, ~varmask[1]); /* VPC_VAR[1].DISABLE */
OUT_RING(ring, ~varmask[2]); /* VPC_VAR[2].DISABLE */
OUT_RING(ring, ~varmask[3]); /* VPC_VAR[3].DISABLE */
/* a5xx appends pos/psize to end of the linkage map: */
if (pos_regid != regid(63,0))
ir3_link_add(&l, pos_regid, 0xf, l.max_loc);
if (psize_regid != regid(63,0))
ir3_link_add(&l, psize_regid, 0x1, l.max_loc);
for (i = 0, j = 0; (i < 16) && (j < l.cnt); i++) {
uint32_t reg = 0;
OUT_PKT4(ring, REG_A5XX_SP_VS_OUT_REG(i), 1);
reg |= A5XX_SP_VS_OUT_REG_A_REGID(l.var[j].regid);
reg |= A5XX_SP_VS_OUT_REG_A_COMPMASK(l.var[j].compmask);
j++;
reg |= A5XX_SP_VS_OUT_REG_B_REGID(l.var[j].regid);
reg |= A5XX_SP_VS_OUT_REG_B_COMPMASK(l.var[j].compmask);
j++;
OUT_RING(ring, reg);
}
for (i = 0, j = 0; (i < 8) && (j < l.cnt); i++) {
uint32_t reg = 0;
OUT_PKT4(ring, REG_A5XX_SP_VS_VPC_DST_REG(i), 1);
reg |= A5XX_SP_VS_VPC_DST_REG_OUTLOC0(l.var[j++].loc);
reg |= A5XX_SP_VS_VPC_DST_REG_OUTLOC1(l.var[j++].loc);
reg |= A5XX_SP_VS_VPC_DST_REG_OUTLOC2(l.var[j++].loc);
reg |= A5XX_SP_VS_VPC_DST_REG_OUTLOC3(l.var[j++].loc);
OUT_RING(ring, reg);
}
OUT_PKT4(ring, REG_A5XX_SP_VS_OBJ_START_LO, 2);
OUT_RELOC(ring, s[VS].v->bo, 0, 0, 0); /* SP_VS_OBJ_START_LO/HI */
if (s[VS].instrlen)
emit_shader(ring, s[VS].v);
// TODO depending on other bits in this reg (if any) set somewhere else?
OUT_PKT4(ring, REG_A5XX_PC_PRIM_VTX_CNTL, 1);
OUT_RING(ring, COND(s[VS].v->writes_psize, A5XX_PC_PRIM_VTX_CNTL_PSIZE));
if (emit->key.binning_pass) {
OUT_PKT4(ring, REG_A5XX_SP_FS_OBJ_START_LO, 2);
OUT_RING(ring, 0x00000000); /* SP_FS_OBJ_START_LO */
OUT_RING(ring, 0x00000000); /* SP_FS_OBJ_START_HI */
} else {
uint32_t stride_in_vpc = align(s[FS].v->total_in, 4) + 4;
if (s[VS].v->writes_psize)
stride_in_vpc++;
// TODO if some of these other bits depend on something other than
// program state we should probably move these next three regs:
OUT_PKT4(ring, REG_A5XX_SP_PRIMITIVE_CNTL, 1);
OUT_RING(ring, A5XX_SP_PRIMITIVE_CNTL_VSOUT(l.cnt));
OUT_PKT4(ring, REG_A5XX_VPC_CNTL_0, 1);
OUT_RING(ring, A5XX_VPC_CNTL_0_STRIDE_IN_VPC(stride_in_vpc) |
COND(s[FS].v->total_in > 0, A5XX_VPC_CNTL_0_VARYING) |
0x10000); // XXX
OUT_PKT4(ring, REG_A5XX_PC_PRIMITIVE_CNTL, 1);
OUT_RING(ring, A5XX_PC_PRIMITIVE_CNTL_STRIDE_IN_VPC(stride_in_vpc) |
0x400); // XXX
OUT_PKT4(ring, REG_A5XX_SP_FS_OBJ_START_LO, 2);
OUT_RELOC(ring, s[FS].v->bo, 0, 0, 0); /* SP_FS_OBJ_START_LO/HI */
}
OUT_PKT4(ring, REG_A5XX_HLSQ_CONTROL_0_REG, 5);
OUT_RING(ring, 0x00000881); /* XXX HLSQ_CONTROL_0 */
OUT_RING(ring, A5XX_HLSQ_CONTROL_1_REG_PRIMALLOCTHRESHOLD(63));
OUT_RING(ring, A5XX_HLSQ_CONTROL_2_REG_FACEREGID(face_regid) |
0xfcfcfc00); /* XXX */
OUT_RING(ring, A5XX_HLSQ_CONTROL_3_REG_FRAGCOORDXYREGID(vcoord_regid) |
0xfcfcfc00); /* XXX */
OUT_RING(ring, A5XX_HLSQ_CONTROL_4_REG_XYCOORDREGID(coord_regid) |
A5XX_HLSQ_CONTROL_4_REG_ZWCOORDREGID(zwcoord_regid) |
0x0000fcfc); /* XXX */
OUT_PKT4(ring, REG_A5XX_SP_FS_CTRL_REG0, 1);
OUT_RING(ring, COND(s[FS].v->total_in > 0, A5XX_SP_FS_CTRL_REG0_VARYING) |
0x4000e | /* XXX set pretty much everywhere */
A5XX_SP_FS_CTRL_REG0_HALFREGFOOTPRINT(s[FS].i->max_half_reg + 1) |
A5XX_SP_FS_CTRL_REG0_FULLREGFOOTPRINT(s[FS].i->max_reg + 1) |
A5XX_SP_FS_CTRL_REG0_BRANCHSTACK(0x3) | // XXX need to figure this out somehow..
COND(s[FS].v->has_samp, A5XX_SP_FS_CTRL_REG0_PIXLODENABLE));
OUT_PKT4(ring, REG_A5XX_HLSQ_UPDATE_CNTL, 1);
OUT_RING(ring, 0x020fffff); /* XXX */
OUT_PKT4(ring, REG_A5XX_VPC_GS_SIV_CNTL, 1);
OUT_RING(ring, 0x0000ffff); /* XXX */
OUT_PKT4(ring, REG_A5XX_SP_SP_CNTL, 1);
OUT_RING(ring, 0x00000010); /* XXX */
OUT_PKT4(ring, REG_A5XX_GRAS_CNTL, 1);
OUT_RING(ring, COND(s[FS].v->total_in > 0, A5XX_GRAS_CNTL_VARYING) |
COND(s[FS].v->frag_coord, A5XX_GRAS_CNTL_XCOORD |
A5XX_GRAS_CNTL_YCOORD |
A5XX_GRAS_CNTL_ZCOORD |
A5XX_GRAS_CNTL_WCOORD |
A5XX_GRAS_CNTL_UNK3) |
COND(s[FS].v->frag_face, A5XX_GRAS_CNTL_UNK3));
OUT_PKT4(ring, REG_A5XX_RB_RENDER_CONTROL0, 3);
OUT_RING(ring,
COND(s[FS].v->total_in > 0, A5XX_RB_RENDER_CONTROL0_VARYING) |
COND(s[FS].v->frag_coord, A5XX_RB_RENDER_CONTROL0_XCOORD |
A5XX_RB_RENDER_CONTROL0_YCOORD |
A5XX_RB_RENDER_CONTROL0_ZCOORD |
A5XX_RB_RENDER_CONTROL0_WCOORD |
A5XX_RB_RENDER_CONTROL0_UNK3) |
COND(s[FS].v->frag_face, A5XX_RB_RENDER_CONTROL0_UNK3));
OUT_RING(ring,
COND(s[FS].v->frag_face, A5XX_RB_RENDER_CONTROL1_FACENESS));
OUT_RING(ring, A5XX_RB_FS_OUTPUT_CNTL_MRT(nr) |
COND(s[FS].v->writes_pos, A5XX_RB_FS_OUTPUT_CNTL_FRAG_WRITES_Z));
OUT_PKT4(ring, REG_A5XX_SP_FS_OUTPUT_CNTL, 9);
OUT_RING(ring, A5XX_SP_FS_OUTPUT_CNTL_MRT(nr) |
A5XX_SP_FS_OUTPUT_CNTL_DEPTH_REGID(posz_regid) |
A5XX_SP_FS_OUTPUT_CNTL_SAMPLEMASK_REGID(regid(63, 0)));
for (i = 0; i < 8; i++) {
OUT_RING(ring, A5XX_SP_FS_OUTPUT_REG_REGID(color_regid[i]) |
COND(emit->key.half_precision,
A5XX_SP_FS_OUTPUT_REG_HALF_PRECISION));
}
if (emit->key.binning_pass) {
OUT_PKT4(ring, REG_A5XX_VPC_PACK, 1);
OUT_RING(ring, A5XX_VPC_PACK_NUMNONPOSVAR(0));
} else {
uint32_t vinterp[8], vpsrepl[8];
memset(vinterp, 0, sizeof(vinterp));
memset(vpsrepl, 0, sizeof(vpsrepl));
/* looks like we need to do int varyings in the frag
* shader on a5xx (no flatshad reg? or a420.0 bug?):
*
* (sy)(ss)nop
* (sy)ldlv.u32 r0.x,l[r0.x], 1
* ldlv.u32 r0.y,l[r0.x+1], 1
* (ss)bary.f (ei)r63.x, 0, r0.x
* (ss)(rpt1)cov.s32f16 hr0.x, (r)r0.x
* (rpt5)nop
* sam (f16)(xyzw)hr0.x, hr0.x, s#0, t#0
*
* Possibly on later a5xx variants we'll be able to use
* something like the code below instead of workaround
* in the shader:
*/
/* figure out VARYING_INTERP / VARYING_PS_REPL register values: */
for (j = -1; (j = ir3_next_varying(s[FS].v, j)) < (int)s[FS].v->inputs_count; ) {
/* NOTE: varyings are packed, so if compmask is 0xb
* then first, third, and fourth component occupy
* three consecutive varying slots:
*/
unsigned compmask = s[FS].v->inputs[j].compmask;
uint32_t inloc = s[FS].v->inputs[j].inloc;
if ((s[FS].v->inputs[j].interpolate == INTERP_MODE_FLAT) ||
(s[FS].v->inputs[j].rasterflat && emit->rasterflat)) {
uint32_t loc = inloc;
for (i = 0; i < 4; i++) {
if (compmask & (1 << i)) {
vinterp[loc / 16] |= 1 << ((loc % 16) * 2);
//flatshade[loc / 32] |= 1 << (loc % 32);
loc++;
}
}
}
gl_varying_slot slot = s[FS].v->inputs[j].slot;
/* since we don't enable PIPE_CAP_TGSI_TEXCOORD: */
if (slot >= VARYING_SLOT_VAR0) {
unsigned texmask = 1 << (slot - VARYING_SLOT_VAR0);
/* Replace the .xy coordinates with S/T from the point sprite. Set
* interpolation bits for .zw such that they become .01
*/
if (emit->sprite_coord_enable & texmask) {
/* mask is two 2-bit fields, where:
* '01' -> S
* '10' -> T
* '11' -> 1 - T (flip mode)
*/
unsigned mask = emit->sprite_coord_mode ? 0b1101 : 0b1001;
uint32_t loc = inloc;
if (compmask & 0x1) {
vpsrepl[loc / 16] |= ((mask >> 0) & 0x3) << ((loc % 16) * 2);
loc++;
}
if (compmask & 0x2) {
vpsrepl[loc / 16] |= ((mask >> 2) & 0x3) << ((loc % 16) * 2);
loc++;
}
if (compmask & 0x4) {
/* .z <- 0.0f */
vinterp[loc / 16] |= 0b10 << ((loc % 16) * 2);
loc++;
}
if (compmask & 0x8) {
/* .w <- 1.0f */
vinterp[loc / 16] |= 0b11 << ((loc % 16) * 2);
loc++;
}
}
void
fd3_program_emit(struct fd_ringbuffer *ring, struct fd3_emit *emit)
{
const struct ir3_shader_variant *vp, *fp;
const struct ir3_info *vsi, *fsi;
enum a3xx_instrbuffermode fpbuffer, vpbuffer;
uint32_t fpbuffersz, vpbuffersz, fsoff;
uint32_t pos_regid, posz_regid, psize_regid, color_regid;
int constmode;
int i, j, k;
vp = fd3_emit_get_vp(emit);
if (emit->key.binning_pass) {
/* use dummy stateobj to simplify binning vs non-binning: */
static const struct ir3_shader_variant binning_fp = {};
fp = &binning_fp;
} else {
fp = fd3_emit_get_fp(emit);
}
vsi = &vp->info;
fsi = &fp->info;
fpbuffer = BUFFER;
vpbuffer = BUFFER;
fpbuffersz = fp->instrlen;
vpbuffersz = vp->instrlen;
/*
* Decide whether to use BUFFER or CACHE mode for VS and FS. It
* appears like 256 is the hard limit, but when the combined size
* exceeds 128 then blob will try to keep FS in BUFFER mode and
* switch to CACHE for VS until VS is too large. The blob seems
* to switch FS out of BUFFER mode at slightly under 128. But
* a bit fuzzy on the decision tree, so use slightly conservative
* limits.
*
* TODO check if these thresholds for BUFFER vs CACHE mode are the
* same for all a3xx or whether we need to consider the gpuid
*/
if ((fpbuffersz + vpbuffersz) > 128) {
if (fpbuffersz < 112) {
/* FP:BUFFER VP:CACHE */
vpbuffer = CACHE;
vpbuffersz = 256 - fpbuffersz;
} else if (vpbuffersz < 112) {
/* FP:CACHE VP:BUFFER */
fpbuffer = CACHE;
fpbuffersz = 256 - vpbuffersz;
} else {
/* FP:CACHE VP:CACHE */
vpbuffer = fpbuffer = CACHE;
vpbuffersz = fpbuffersz = 192;
}
}
if (fpbuffer == BUFFER) {
fsoff = 128 - fpbuffersz;
} else {
fsoff = 256 - fpbuffersz;
}
/* seems like vs->constlen + fs->constlen > 256, then CONSTMODE=1 */
constmode = ((vp->constlen + fp->constlen) > 256) ? 1 : 0;
pos_regid = find_output_regid(vp,
ir3_semantic_name(TGSI_SEMANTIC_POSITION, 0));
posz_regid = find_output_regid(fp,
ir3_semantic_name(TGSI_SEMANTIC_POSITION, 0));
psize_regid = find_output_regid(vp,
ir3_semantic_name(TGSI_SEMANTIC_PSIZE, 0));
color_regid = find_output_regid(fp,
ir3_semantic_name(TGSI_SEMANTIC_COLOR, 0));
/* we could probably divide this up into things that need to be
* emitted if frag-prog is dirty vs if vert-prog is dirty..
*/
OUT_PKT0(ring, REG_A3XX_HLSQ_CONTROL_0_REG, 6);
OUT_RING(ring, A3XX_HLSQ_CONTROL_0_REG_FSTHREADSIZE(FOUR_QUADS) |
A3XX_HLSQ_CONTROL_0_REG_CONSTMODE(constmode) |
/* NOTE: I guess SHADERRESTART and CONSTFULLUPDATE maybe
* flush some caches? I think we only need to set those
* bits if we have updated const or shader..
*/
A3XX_HLSQ_CONTROL_0_REG_SPSHADERRESTART |
A3XX_HLSQ_CONTROL_0_REG_SPCONSTFULLUPDATE);
OUT_RING(ring, A3XX_HLSQ_CONTROL_1_REG_VSTHREADSIZE(TWO_QUADS) |
A3XX_HLSQ_CONTROL_1_REG_VSSUPERTHREADENABLE |
COND(fp->frag_coord, A3XX_HLSQ_CONTROL_1_REG_ZWCOORD));
OUT_RING(ring, A3XX_HLSQ_CONTROL_2_REG_PRIMALLOCTHRESHOLD(31));
OUT_RING(ring, A3XX_HLSQ_CONTROL_3_REG_REGID(fp->pos_regid));
OUT_RING(ring, A3XX_HLSQ_VS_CONTROL_REG_CONSTLENGTH(vp->constlen) |
A3XX_HLSQ_VS_CONTROL_REG_CONSTSTARTOFFSET(0) |
A3XX_HLSQ_VS_CONTROL_REG_INSTRLENGTH(vpbuffersz));
OUT_RING(ring, A3XX_HLSQ_FS_CONTROL_REG_CONSTLENGTH(fp->constlen) |
A3XX_HLSQ_FS_CONTROL_REG_CONSTSTARTOFFSET(128) |
A3XX_HLSQ_FS_CONTROL_REG_INSTRLENGTH(fpbuffersz));
OUT_PKT0(ring, REG_A3XX_SP_SP_CTRL_REG, 1);
OUT_RING(ring, A3XX_SP_SP_CTRL_REG_CONSTMODE(constmode) |
COND(emit->key.binning_pass, A3XX_SP_SP_CTRL_REG_BINNING) |
A3XX_SP_SP_CTRL_REG_SLEEPMODE(1) |
A3XX_SP_SP_CTRL_REG_L0MODE(0));
OUT_PKT0(ring, REG_A3XX_SP_VS_LENGTH_REG, 1);
OUT_RING(ring, A3XX_SP_VS_LENGTH_REG_SHADERLENGTH(vp->instrlen));
OUT_PKT0(ring, REG_A3XX_SP_VS_CTRL_REG0, 3);
OUT_RING(ring, A3XX_SP_VS_CTRL_REG0_THREADMODE(MULTI) |
A3XX_SP_VS_CTRL_REG0_INSTRBUFFERMODE(vpbuffer) |
COND(vpbuffer == CACHE, A3XX_SP_VS_CTRL_REG0_CACHEINVALID) |
A3XX_SP_VS_CTRL_REG0_HALFREGFOOTPRINT(vsi->max_half_reg + 1) |
A3XX_SP_VS_CTRL_REG0_FULLREGFOOTPRINT(vsi->max_reg + 1) |
A3XX_SP_VS_CTRL_REG0_INOUTREGOVERLAP(0) |
A3XX_SP_VS_CTRL_REG0_THREADSIZE(TWO_QUADS) |
A3XX_SP_VS_CTRL_REG0_SUPERTHREADMODE |
COND(vp->has_samp, A3XX_SP_VS_CTRL_REG0_PIXLODENABLE) |
A3XX_SP_VS_CTRL_REG0_LENGTH(vpbuffersz));
OUT_RING(ring, A3XX_SP_VS_CTRL_REG1_CONSTLENGTH(vp->constlen) |
A3XX_SP_VS_CTRL_REG1_INITIALOUTSTANDING(vp->total_in) |
A3XX_SP_VS_CTRL_REG1_CONSTFOOTPRINT(MAX2(vp->constlen + 1, 0)));
OUT_RING(ring, A3XX_SP_VS_PARAM_REG_POSREGID(pos_regid) |
A3XX_SP_VS_PARAM_REG_PSIZEREGID(psize_regid) |
A3XX_SP_VS_PARAM_REG_TOTALVSOUTVAR(align(fp->total_in, 4) / 4));
for (i = 0, j = -1; (i < 8) && (j < (int)fp->inputs_count); i++) {
uint32_t reg = 0;
OUT_PKT0(ring, REG_A3XX_SP_VS_OUT_REG(i), 1);
j = next_varying(fp, j);
if (j < fp->inputs_count) {
k = find_output(vp, fp->inputs[j].semantic);
reg |= A3XX_SP_VS_OUT_REG_A_REGID(vp->outputs[k].regid);
reg |= A3XX_SP_VS_OUT_REG_A_COMPMASK(fp->inputs[j].compmask);
}
j = next_varying(fp, j);
if (j < fp->inputs_count) {
k = find_output(vp, fp->inputs[j].semantic);
reg |= A3XX_SP_VS_OUT_REG_B_REGID(vp->outputs[k].regid);
reg |= A3XX_SP_VS_OUT_REG_B_COMPMASK(fp->inputs[j].compmask);
}
OUT_RING(ring, reg);
}
for (i = 0, j = -1; (i < 4) && (j < (int)fp->inputs_count); i++) {
uint32_t reg = 0;
OUT_PKT0(ring, REG_A3XX_SP_VS_VPC_DST_REG(i), 1);
j = next_varying(fp, j);
if (j < fp->inputs_count)
reg |= A3XX_SP_VS_VPC_DST_REG_OUTLOC0(fp->inputs[j].inloc);
j = next_varying(fp, j);
if (j < fp->inputs_count)
reg |= A3XX_SP_VS_VPC_DST_REG_OUTLOC1(fp->inputs[j].inloc);
j = next_varying(fp, j);
if (j < fp->inputs_count)
reg |= A3XX_SP_VS_VPC_DST_REG_OUTLOC2(fp->inputs[j].inloc);
j = next_varying(fp, j);
if (j < fp->inputs_count)
reg |= A3XX_SP_VS_VPC_DST_REG_OUTLOC3(fp->inputs[j].inloc);
OUT_RING(ring, reg);
}
OUT_PKT0(ring, REG_A3XX_SP_VS_OBJ_OFFSET_REG, 2);
OUT_RING(ring, A3XX_SP_VS_OBJ_OFFSET_REG_CONSTOBJECTOFFSET(0) |
A3XX_SP_VS_OBJ_OFFSET_REG_SHADEROBJOFFSET(0));
OUT_RELOC(ring, vp->bo, 0, 0, 0); /* SP_VS_OBJ_START_REG */
if (emit->key.binning_pass) {
OUT_PKT0(ring, REG_A3XX_SP_FS_LENGTH_REG, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A3XX_SP_FS_CTRL_REG0, 2);
OUT_RING(ring, A3XX_SP_FS_CTRL_REG0_THREADMODE(MULTI) |
A3XX_SP_FS_CTRL_REG0_INSTRBUFFERMODE(BUFFER));
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A3XX_SP_FS_OBJ_OFFSET_REG, 1);
OUT_RING(ring, A3XX_SP_FS_OBJ_OFFSET_REG_CONSTOBJECTOFFSET(128) |
A3XX_SP_FS_OBJ_OFFSET_REG_SHADEROBJOFFSET(0));
} else {
OUT_PKT0(ring, REG_A3XX_SP_FS_LENGTH_REG, 1);
OUT_RING(ring, A3XX_SP_FS_LENGTH_REG_SHADERLENGTH(fp->instrlen));
OUT_PKT0(ring, REG_A3XX_SP_FS_CTRL_REG0, 2);
OUT_RING(ring, A3XX_SP_FS_CTRL_REG0_THREADMODE(MULTI) |
A3XX_SP_FS_CTRL_REG0_INSTRBUFFERMODE(fpbuffer) |
COND(fpbuffer == CACHE, A3XX_SP_FS_CTRL_REG0_CACHEINVALID) |
A3XX_SP_FS_CTRL_REG0_HALFREGFOOTPRINT(fsi->max_half_reg + 1) |
A3XX_SP_FS_CTRL_REG0_FULLREGFOOTPRINT(fsi->max_reg + 1) |
A3XX_SP_FS_CTRL_REG0_INOUTREGOVERLAP(1) |
A3XX_SP_FS_CTRL_REG0_THREADSIZE(FOUR_QUADS) |
A3XX_SP_FS_CTRL_REG0_SUPERTHREADMODE |
COND(fp->has_samp > 0, A3XX_SP_FS_CTRL_REG0_PIXLODENABLE) |
A3XX_SP_FS_CTRL_REG0_LENGTH(fpbuffersz));
OUT_RING(ring, A3XX_SP_FS_CTRL_REG1_CONSTLENGTH(fp->constlen) |
A3XX_SP_FS_CTRL_REG1_INITIALOUTSTANDING(fp->total_in) |
A3XX_SP_FS_CTRL_REG1_CONSTFOOTPRINT(MAX2(fp->constlen + 1, 0)) |
A3XX_SP_FS_CTRL_REG1_HALFPRECVAROFFSET(63));
OUT_PKT0(ring, REG_A3XX_SP_FS_OBJ_OFFSET_REG, 2);
OUT_RING(ring, A3XX_SP_FS_OBJ_OFFSET_REG_CONSTOBJECTOFFSET(
MAX2(128, vp->constlen)) |
A3XX_SP_FS_OBJ_OFFSET_REG_SHADEROBJOFFSET(fsoff));
OUT_RELOC(ring, fp->bo, 0, 0, 0); /* SP_FS_OBJ_START_REG */
}
OUT_PKT0(ring, REG_A3XX_SP_FS_OUTPUT_REG, 1);
if (fp->writes_pos) {
OUT_RING(ring, A3XX_SP_FS_OUTPUT_REG_DEPTH_ENABLE |
A3XX_SP_FS_OUTPUT_REG_DEPTH_REGID(posz_regid));
} else {
OUT_RING(ring, 0x00000000);
}
OUT_PKT0(ring, REG_A3XX_SP_FS_MRT_REG(0), 4);
OUT_RING(ring, A3XX_SP_FS_MRT_REG_REGID(color_regid) |
COND(fp->key.half_precision, A3XX_SP_FS_MRT_REG_HALF_PRECISION));
OUT_RING(ring, A3XX_SP_FS_MRT_REG_REGID(0));
OUT_RING(ring, A3XX_SP_FS_MRT_REG_REGID(0));
OUT_RING(ring, A3XX_SP_FS_MRT_REG_REGID(0));
if (emit->key.binning_pass) {
OUT_PKT0(ring, REG_A3XX_VPC_ATTR, 2);
OUT_RING(ring, A3XX_VPC_ATTR_THRDASSIGN(1) |
A3XX_VPC_ATTR_LMSIZE(1) |
COND(vp->writes_psize, A3XX_VPC_ATTR_PSIZE));
OUT_RING(ring, 0x00000000);
} else {
uint32_t vinterp[4] = {0}, flatshade[2] = {0};
/* figure out VARYING_INTERP / FLAT_SHAD register values: */
for (j = -1; (j = next_varying(fp, j)) < (int)fp->inputs_count; ) {
uint32_t interp = fp->inputs[j].interpolate;
if ((interp == TGSI_INTERPOLATE_CONSTANT) ||
((interp == TGSI_INTERPOLATE_COLOR) && emit->rasterflat)) {
/* TODO might be cleaner to just +8 in SP_VS_VPC_DST_REG
* instead.. rather than -8 everywhere else..
*/
uint32_t loc = fp->inputs[j].inloc - 8;
/* currently assuming varyings aligned to 4 (not
* packed):
*/
debug_assert((loc % 4) == 0);
for (i = 0; i < 4; i++, loc++) {
vinterp[loc / 16] |= FLAT << ((loc % 16) * 2);
flatshade[loc / 32] |= 1 << (loc % 32);
}
}
}
OUT_PKT0(ring, REG_A3XX_VPC_ATTR, 2);
OUT_RING(ring, A3XX_VPC_ATTR_TOTALATTR(fp->total_in) |
A3XX_VPC_ATTR_THRDASSIGN(1) |
A3XX_VPC_ATTR_LMSIZE(1) |
COND(vp->writes_psize, A3XX_VPC_ATTR_PSIZE));
OUT_RING(ring, A3XX_VPC_PACK_NUMFPNONPOSVAR(fp->total_in) |
A3XX_VPC_PACK_NUMNONPOSVSVAR(fp->total_in));
OUT_PKT0(ring, REG_A3XX_VPC_VARYING_INTERP_MODE(0), 4);
OUT_RING(ring, vinterp[0]); /* VPC_VARYING_INTERP[0].MODE */
OUT_RING(ring, vinterp[1]); /* VPC_VARYING_INTERP[1].MODE */
OUT_RING(ring, vinterp[2]); /* VPC_VARYING_INTERP[2].MODE */
OUT_RING(ring, vinterp[3]); /* VPC_VARYING_INTERP[3].MODE */
OUT_PKT0(ring, REG_A3XX_VPC_VARYING_PS_REPL_MODE(0), 4);
OUT_RING(ring, fp->shader->vpsrepl[0]); /* VPC_VARYING_PS_REPL[0].MODE */
OUT_RING(ring, fp->shader->vpsrepl[1]); /* VPC_VARYING_PS_REPL[1].MODE */
OUT_RING(ring, fp->shader->vpsrepl[2]); /* VPC_VARYING_PS_REPL[2].MODE */
OUT_RING(ring, fp->shader->vpsrepl[3]); /* VPC_VARYING_PS_REPL[3].MODE */
OUT_PKT0(ring, REG_A3XX_SP_FS_FLAT_SHAD_MODE_REG_0, 2);
OUT_RING(ring, flatshade[0]); /* SP_FS_FLAT_SHAD_MODE_REG_0 */
OUT_RING(ring, flatshade[1]); /* SP_FS_FLAT_SHAD_MODE_REG_1 */
}
OUT_PKT0(ring, REG_A3XX_VFD_VS_THREADING_THRESHOLD, 1);
OUT_RING(ring, A3XX_VFD_VS_THREADING_THRESHOLD_REGID_THRESHOLD(15) |
A3XX_VFD_VS_THREADING_THRESHOLD_REGID_VTXCNT(252));
if (vpbuffer == BUFFER)
emit_shader(ring, vp);
OUT_PKT0(ring, REG_A3XX_VFD_PERFCOUNTER0_SELECT, 1);
OUT_RING(ring, 0x00000000); /* VFD_PERFCOUNTER0_SELECT */
if (!emit->key.binning_pass) {
if (fpbuffer == BUFFER)
emit_shader(ring, fp);
OUT_PKT0(ring, REG_A3XX_VFD_PERFCOUNTER0_SELECT, 1);
OUT_RING(ring, 0x00000000); /* VFD_PERFCOUNTER0_SELECT */
}
}
static void
nouveau_accel_init(struct nouveau_drm *drm)
{
struct nvif_device *device = &drm->device;
u32 arg0, arg1;
u32 sclass[16];
int ret, i;
if (nouveau_noaccel)
return;
/* initialise synchronisation routines */
/*XXX: this is crap, but the fence/channel stuff is a little
* backwards in some places. this will be fixed.
*/
ret = nvif_object_sclass(&device->base, sclass, ARRAY_SIZE(sclass));
if (ret < 0)
return;
for (ret = -ENOSYS, i = 0; ret && i < ARRAY_SIZE(sclass); i++) {
switch (sclass[i]) {
case NV03_CHANNEL_DMA:
ret = nv04_fence_create(drm);
break;
case NV10_CHANNEL_DMA:
ret = nv10_fence_create(drm);
break;
case NV17_CHANNEL_DMA:
case NV40_CHANNEL_DMA:
ret = nv17_fence_create(drm);
break;
case NV50_CHANNEL_GPFIFO:
ret = nv50_fence_create(drm);
break;
case G82_CHANNEL_GPFIFO:
ret = nv84_fence_create(drm);
break;
case FERMI_CHANNEL_GPFIFO:
case KEPLER_CHANNEL_GPFIFO_A:
ret = nvc0_fence_create(drm);
break;
default:
break;
}
}
if (ret) {
NV_ERROR(drm, "failed to initialise sync subsystem, %d\n", ret);
nouveau_accel_fini(drm);
return;
}
if (device->info.family >= NV_DEVICE_INFO_V0_KEPLER) {
ret = nouveau_channel_new(drm, &drm->device, NVDRM_CHAN + 1,
KEPLER_CHANNEL_GPFIFO_A_V0_ENGINE_CE0|
KEPLER_CHANNEL_GPFIFO_A_V0_ENGINE_CE1,
0, &drm->cechan);
if (ret)
NV_ERROR(drm, "failed to create ce channel, %d\n", ret);
arg0 = KEPLER_CHANNEL_GPFIFO_A_V0_ENGINE_GR;
arg1 = 1;
} else
if (device->info.chipset >= 0xa3 &&
device->info.chipset != 0xaa &&
device->info.chipset != 0xac) {
ret = nouveau_channel_new(drm, &drm->device, NVDRM_CHAN + 1,
NvDmaFB, NvDmaTT, &drm->cechan);
if (ret)
NV_ERROR(drm, "failed to create ce channel, %d\n", ret);
arg0 = NvDmaFB;
arg1 = NvDmaTT;
} else {
arg0 = NvDmaFB;
arg1 = NvDmaTT;
}
ret = nouveau_channel_new(drm, &drm->device, NVDRM_CHAN, arg0, arg1,
&drm->channel);
if (ret) {
NV_ERROR(drm, "failed to create kernel channel, %d\n", ret);
nouveau_accel_fini(drm);
return;
}
ret = nvif_object_init(drm->channel->object, NULL, NVDRM_NVSW,
nouveau_abi16_swclass(drm), NULL, 0, &drm->nvsw);
if (ret == 0) {
struct nouveau_software_chan *swch;
ret = RING_SPACE(drm->channel, 2);
if (ret == 0) {
if (device->info.family < NV_DEVICE_INFO_V0_FERMI) {
BEGIN_NV04(drm->channel, NvSubSw, 0, 1);
OUT_RING (drm->channel, NVDRM_NVSW);
} else
if (device->info.family < NV_DEVICE_INFO_V0_KEPLER) {
BEGIN_NVC0(drm->channel, FermiSw, 0, 1);
OUT_RING (drm->channel, 0x001f0000);
}
}
swch = (void *)nvkm_object(&drm->nvsw)->parent;
swch->flip = nouveau_flip_complete;
swch->flip_data = drm->channel;
}
if (ret) {
NV_ERROR(drm, "failed to allocate software object, %d\n", ret);
nouveau_accel_fini(drm);
return;
}
if (device->info.family < NV_DEVICE_INFO_V0_FERMI) {
ret = nouveau_gpuobj_new(nvkm_object(&drm->device), NULL, 32,
0, 0, &drm->notify);
if (ret) {
NV_ERROR(drm, "failed to allocate notifier, %d\n", ret);
nouveau_accel_fini(drm);
return;
}
ret = nvif_object_init(drm->channel->object, NULL, NvNotify0,
NV_DMA_IN_MEMORY,
&(struct nv_dma_v0) {
.target = NV_DMA_V0_TARGET_VRAM,
.access = NV_DMA_V0_ACCESS_RDWR,
.start = drm->notify->addr,
.limit = drm->notify->addr + 31
}, sizeof(struct nv_dma_v0),
static void
fd4_emit_tile_gmem2mem(struct fd_context *ctx, struct fd_tile *tile)
{
struct fd4_context *fd4_ctx = fd4_context(ctx);
struct fd_ringbuffer *ring = ctx->ring;
struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
struct fd4_emit emit = {
.vtx = &fd4_ctx->solid_vbuf_state,
.prog = &ctx->solid_prog,
.key = key,
.format = fd4_emit_format(pfb->cbufs[0]),
};
OUT_PKT0(ring, REG_A4XX_RB_DEPTH_CONTROL, 1);
OUT_RING(ring, A4XX_RB_DEPTH_CONTROL_ZFUNC(FUNC_NEVER));
OUT_PKT0(ring, REG_A4XX_RB_STENCIL_CONTROL, 2);
OUT_RING(ring, A4XX_RB_STENCIL_CONTROL_FUNC(FUNC_NEVER) |
A4XX_RB_STENCIL_CONTROL_FAIL(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZPASS(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZFAIL(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_FUNC_BF(FUNC_NEVER) |
A4XX_RB_STENCIL_CONTROL_FAIL_BF(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZPASS_BF(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZFAIL_BF(STENCIL_KEEP));
OUT_RING(ring, 0x00000000); /* RB_STENCIL_CONTROL2 */
OUT_PKT0(ring, REG_A4XX_RB_STENCILREFMASK, 2);
OUT_RING(ring, 0xff000000 |
A4XX_RB_STENCILREFMASK_STENCILREF(0) |
A4XX_RB_STENCILREFMASK_STENCILMASK(0) |
A4XX_RB_STENCILREFMASK_STENCILWRITEMASK(0xff));
OUT_RING(ring, 0xff000000 |
A4XX_RB_STENCILREFMASK_BF_STENCILREF(0) |
A4XX_RB_STENCILREFMASK_BF_STENCILMASK(0) |
A4XX_RB_STENCILREFMASK_BF_STENCILWRITEMASK(0xff));
OUT_PKT0(ring, REG_A4XX_GRAS_SU_MODE_CONTROL, 1);
OUT_RING(ring, A4XX_GRAS_SU_MODE_CONTROL_LINEHALFWIDTH(0));
fd_wfi(ctx, ring);
OUT_PKT0(ring, REG_A4XX_GRAS_CL_CLIP_CNTL, 1);
OUT_RING(ring, 0x80000); /* GRAS_CL_CLIP_CNTL */
OUT_PKT0(ring, REG_A4XX_GRAS_CL_VPORT_XOFFSET_0, 6);
OUT_RING(ring, A4XX_GRAS_CL_VPORT_XOFFSET_0((float)pfb->width/2.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_XSCALE_0((float)pfb->width/2.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_YOFFSET_0((float)pfb->height/2.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_YSCALE_0(-(float)pfb->height/2.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_ZOFFSET_0(0.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_ZSCALE_0(1.0));
OUT_PKT0(ring, REG_A4XX_RB_RENDER_CONTROL, 1);
OUT_RING(ring, A4XX_RB_RENDER_CONTROL_DISABLE_COLOR_PIPE |
0xa); /* XXX */
OUT_PKT0(ring, REG_A4XX_GRAS_SC_CONTROL, 1);
OUT_RING(ring, A4XX_GRAS_SC_CONTROL_RENDER_MODE(RB_RESOLVE_PASS) |
A4XX_GRAS_SC_CONTROL_MSAA_DISABLE |
A4XX_GRAS_SC_CONTROL_MSAA_SAMPLES(MSAA_ONE) |
A4XX_GRAS_SC_CONTROL_RASTER_MODE(1));
OUT_PKT0(ring, REG_A4XX_PC_PRIM_VTX_CNTL, 1);
OUT_RING(ring, A4XX_PC_PRIM_VTX_CNTL_PROVOKING_VTX_LAST);
OUT_PKT0(ring, REG_A4XX_GRAS_ALPHA_CONTROL, 1);
OUT_RING(ring, 0x00000002);
OUT_PKT0(ring, REG_A4XX_GRAS_SC_WINDOW_SCISSOR_BR, 2);
OUT_RING(ring, A4XX_GRAS_SC_WINDOW_SCISSOR_BR_X(pfb->width - 1) |
A4XX_GRAS_SC_WINDOW_SCISSOR_BR_Y(pfb->height - 1));
OUT_RING(ring, A4XX_GRAS_SC_WINDOW_SCISSOR_TL_X(0) |
A4XX_GRAS_SC_WINDOW_SCISSOR_TL_Y(0));
OUT_PKT0(ring, REG_A4XX_VFD_INDEX_OFFSET, 2);
OUT_RING(ring, 0); /* VFD_INDEX_OFFSET */
OUT_RING(ring, 0); /* ??? UNKNOWN_2209 */
fd4_program_emit(ring, &emit);
fd4_emit_vertex_bufs(ring, &emit);
if (ctx->resolve & (FD_BUFFER_DEPTH | FD_BUFFER_STENCIL)) {
uint32_t base = depth_base(ctx);
emit_gmem2mem_surf(ctx, base, pfb->zsbuf);
}
if (ctx->resolve & FD_BUFFER_COLOR) {
emit_gmem2mem_surf(ctx, 0, pfb->cbufs[0]);
}
OUT_PKT0(ring, REG_A4XX_GRAS_SC_CONTROL, 1);
OUT_RING(ring, A4XX_GRAS_SC_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
A4XX_GRAS_SC_CONTROL_MSAA_DISABLE |
A4XX_GRAS_SC_CONTROL_MSAA_SAMPLES(MSAA_ONE) |
A4XX_GRAS_SC_CONTROL_RASTER_MODE(0));
}
/* transfer from system memory to gmem */
static void
emit_mem2gmem_surf(struct fd_context *ctx, uint32_t base,
struct pipe_surface *psurf, uint32_t bin_w)
{
struct fd_ringbuffer *ring = ctx->ring;
emit_mrt(ring, 1, &psurf, &base, bin_w);
fd4_emit_gmem_restore_tex(ring, psurf);
fd4_draw(ctx, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
DI_SRC_SEL_AUTO_INDEX, 2, 1, INDEX_SIZE_IGN, 0, 0, NULL);
}
static void
fd4_emit_tile_mem2gmem(struct fd_context *ctx, struct fd_tile *tile)
{
struct fd4_context *fd4_ctx = fd4_context(ctx);
struct fd_gmem_stateobj *gmem = &ctx->gmem;
struct fd_ringbuffer *ring = ctx->ring;
struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
struct fd4_emit emit = {
.vtx = &fd4_ctx->blit_vbuf_state,
.prog = &ctx->blit_prog[0],
.key = key,
.format = fd4_emit_format(pfb->cbufs[0]),
};
float x0, y0, x1, y1;
unsigned bin_w = tile->bin_w;
unsigned bin_h = tile->bin_h;
unsigned i;
/* write texture coordinates to vertexbuf: */
x0 = ((float)tile->xoff) / ((float)pfb->width);
x1 = ((float)tile->xoff + bin_w) / ((float)pfb->width);
y0 = ((float)tile->yoff) / ((float)pfb->height);
y1 = ((float)tile->yoff + bin_h) / ((float)pfb->height);
OUT_PKT3(ring, CP_MEM_WRITE, 5);
OUT_RELOCW(ring, fd_resource(fd4_ctx->blit_texcoord_vbuf)->bo, 0, 0, 0);
OUT_RING(ring, fui(x0));
OUT_RING(ring, fui(y0));
OUT_RING(ring, fui(x1));
OUT_RING(ring, fui(y1));
for (i = 0; i < 8; i++) {
OUT_PKT0(ring, REG_A4XX_RB_MRT_CONTROL(i), 1);
OUT_RING(ring, A4XX_RB_MRT_CONTROL_FASTCLEAR |
A4XX_RB_MRT_CONTROL_B11 |
A4XX_RB_MRT_CONTROL_COMPONENT_ENABLE(0xf));
OUT_PKT0(ring, REG_A4XX_RB_MRT_BLEND_CONTROL(i), 1);
OUT_RING(ring, A4XX_RB_MRT_BLEND_CONTROL_RGB_SRC_FACTOR(FACTOR_ONE) |
A4XX_RB_MRT_BLEND_CONTROL_RGB_BLEND_OPCODE(BLEND_DST_PLUS_SRC) |
A4XX_RB_MRT_BLEND_CONTROL_RGB_DEST_FACTOR(FACTOR_ZERO) |
A4XX_RB_MRT_BLEND_CONTROL_ALPHA_SRC_FACTOR(FACTOR_ONE) |
A4XX_RB_MRT_BLEND_CONTROL_ALPHA_BLEND_OPCODE(BLEND_DST_PLUS_SRC) |
A4XX_RB_MRT_BLEND_CONTROL_ALPHA_DEST_FACTOR(FACTOR_ZERO));
}
OUT_PKT0(ring, REG_A4XX_RB_RENDER_CONTROL, 1);
OUT_RING(ring, 0x8); /* XXX RB_RENDER_CONTROL */
OUT_PKT0(ring, REG_A4XX_RB_DEPTH_CONTROL, 1);
OUT_RING(ring, A4XX_RB_DEPTH_CONTROL_ZFUNC(FUNC_LESS));
OUT_PKT0(ring, REG_A4XX_GRAS_CL_CLIP_CNTL, 1);
OUT_RING(ring, 0x280000); /* XXX GRAS_CL_CLIP_CNTL */
OUT_PKT0(ring, REG_A4XX_GRAS_SU_MODE_CONTROL, 1);
OUT_RING(ring, A4XX_GRAS_SU_MODE_CONTROL_LINEHALFWIDTH(0) |
A4XX_GRAS_SU_MODE_CONTROL_RENDERING_PASS);
OUT_PKT0(ring, REG_A4XX_GRAS_CL_VPORT_XOFFSET_0, 6);
OUT_RING(ring, A4XX_GRAS_CL_VPORT_XOFFSET_0((float)bin_w/2.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_XSCALE_0((float)bin_w/2.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_YOFFSET_0((float)bin_h/2.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_YSCALE_0(-(float)bin_h/2.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_ZOFFSET_0(0.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_ZSCALE_0(1.0));
OUT_PKT0(ring, REG_A4XX_GRAS_SC_WINDOW_SCISSOR_BR, 2);
OUT_RING(ring, A4XX_GRAS_SC_WINDOW_SCISSOR_BR_X(bin_w - 1) |
A4XX_GRAS_SC_WINDOW_SCISSOR_BR_Y(bin_h - 1));
OUT_RING(ring, A4XX_GRAS_SC_WINDOW_SCISSOR_TL_X(0) |
A4XX_GRAS_SC_WINDOW_SCISSOR_TL_Y(0));
OUT_PKT0(ring, REG_A4XX_GRAS_SC_SCREEN_SCISSOR_TL, 2);
OUT_RING(ring, A4XX_GRAS_SC_SCREEN_SCISSOR_TL_X(0) |
A4XX_GRAS_SC_SCREEN_SCISSOR_TL_Y(0));
OUT_RING(ring, A4XX_GRAS_SC_SCREEN_SCISSOR_BR_X(bin_w - 1) |
A4XX_GRAS_SC_SCREEN_SCISSOR_BR_Y(bin_h - 1));
OUT_PKT0(ring, REG_A4XX_RB_MODE_CONTROL, 1);
OUT_RING(ring, A4XX_RB_MODE_CONTROL_WIDTH(gmem->bin_w) |
A4XX_RB_MODE_CONTROL_HEIGHT(gmem->bin_h));
OUT_PKT0(ring, REG_A4XX_RB_STENCIL_CONTROL, 2);
OUT_RING(ring, A4XX_RB_STENCIL_CONTROL_FUNC(FUNC_ALWAYS) |
A4XX_RB_STENCIL_CONTROL_FAIL(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZPASS(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZFAIL(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_FUNC_BF(FUNC_ALWAYS) |
A4XX_RB_STENCIL_CONTROL_FAIL_BF(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZPASS_BF(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZFAIL_BF(STENCIL_KEEP));
OUT_RING(ring, 0x00000000); /* RB_STENCIL_CONTROL2 */
OUT_PKT0(ring, REG_A4XX_GRAS_SC_CONTROL, 1);
OUT_RING(ring, A4XX_GRAS_SC_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
A4XX_GRAS_SC_CONTROL_MSAA_DISABLE |
A4XX_GRAS_SC_CONTROL_MSAA_SAMPLES(MSAA_ONE) |
A4XX_GRAS_SC_CONTROL_RASTER_MODE(1));
OUT_PKT0(ring, REG_A4XX_PC_PRIM_VTX_CNTL, 1);
OUT_RING(ring, A4XX_PC_PRIM_VTX_CNTL_PROVOKING_VTX_LAST |
A4XX_PC_PRIM_VTX_CNTL_VAROUT(1));
OUT_PKT0(ring, REG_A4XX_VFD_INDEX_OFFSET, 2);
OUT_RING(ring, 0); /* VFD_INDEX_OFFSET */
OUT_RING(ring, 0); /* ??? UNKNOWN_2209 */
fd4_program_emit(ring, &emit);
fd4_emit_vertex_bufs(ring, &emit);
/* for gmem pitch/base calculations, we need to use the non-
* truncated tile sizes:
*/
bin_w = gmem->bin_w;
bin_h = gmem->bin_h;
if (fd_gmem_needs_restore(ctx, tile, FD_BUFFER_DEPTH | FD_BUFFER_STENCIL))
emit_mem2gmem_surf(ctx, depth_base(ctx), pfb->zsbuf, bin_w);
if (fd_gmem_needs_restore(ctx, tile, FD_BUFFER_COLOR))
emit_mem2gmem_surf(ctx, 0, pfb->cbufs[0], bin_w);
OUT_PKT0(ring, REG_A4XX_GRAS_SC_CONTROL, 1);
OUT_RING(ring, A4XX_GRAS_SC_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
A4XX_GRAS_SC_CONTROL_MSAA_SAMPLES(MSAA_ONE) |
A4XX_GRAS_SC_CONTROL_RASTER_MODE(0));
OUT_PKT0(ring, REG_A4XX_RB_MODE_CONTROL, 1);
OUT_RING(ring, A4XX_RB_MODE_CONTROL_WIDTH(gmem->bin_w) |
A4XX_RB_MODE_CONTROL_HEIGHT(gmem->bin_h) |
0x00010000); /* XXX */
}
static void
patch_draws(struct fd_context *ctx, enum pc_di_vis_cull_mode vismode)
{
unsigned i;
for (i = 0; i < fd_patch_num_elements(&ctx->draw_patches); i++) {
struct fd_cs_patch *patch = fd_patch_element(&ctx->draw_patches, i);
*patch->cs = patch->val | DRAW4(0, 0, 0, vismode);
}
util_dynarray_resize(&ctx->draw_patches, 0);
}
static void
patch_rbrc(struct fd_context *ctx, uint32_t val)
{
struct fd4_context *fd4_ctx = fd4_context(ctx);
unsigned i;
for (i = 0; i < fd_patch_num_elements(&fd4_ctx->rbrc_patches); i++) {
struct fd_cs_patch *patch = fd_patch_element(&fd4_ctx->rbrc_patches, i);
*patch->cs = patch->val | val;
}
util_dynarray_resize(&fd4_ctx->rbrc_patches, 0);
}
/* for rendering directly to system memory: */
static void
fd4_emit_sysmem_prep(struct fd_context *ctx)
{
struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
struct fd_ringbuffer *ring = ctx->ring;
fd4_emit_restore(ctx);
OUT_PKT0(ring, REG_A4XX_RB_FRAME_BUFFER_DIMENSION, 1);
OUT_RING(ring, A4XX_RB_FRAME_BUFFER_DIMENSION_WIDTH(pfb->width) |
A4XX_RB_FRAME_BUFFER_DIMENSION_HEIGHT(pfb->height));
emit_mrt(ring, pfb->nr_cbufs, pfb->cbufs, NULL, 0);
/* setup scissor/offset for current tile: */
OUT_PKT0(ring, REG_A4XX_RB_BIN_OFFSET, 1);
OUT_RING(ring, A4XX_RB_BIN_OFFSET_X(0) |
A4XX_RB_BIN_OFFSET_Y(0));
OUT_PKT0(ring, REG_A4XX_GRAS_SC_SCREEN_SCISSOR_TL, 2);
OUT_RING(ring, A4XX_GRAS_SC_SCREEN_SCISSOR_TL_X(0) |
A4XX_GRAS_SC_SCREEN_SCISSOR_TL_Y(0));
OUT_RING(ring, A4XX_GRAS_SC_SCREEN_SCISSOR_BR_X(pfb->width - 1) |
A4XX_GRAS_SC_SCREEN_SCISSOR_BR_Y(pfb->height - 1));
OUT_PKT0(ring, REG_A4XX_RB_MODE_CONTROL, 1);
OUT_RING(ring, A4XX_RB_MODE_CONTROL_WIDTH(0) |
A4XX_RB_MODE_CONTROL_HEIGHT(0) |
0x00c00000); /* XXX */
patch_draws(ctx, IGNORE_VISIBILITY);
patch_rbrc(ctx, 0); // XXX
}
static void
update_vsc_pipe(struct fd_context *ctx)
{
struct fd4_context *fd4_ctx = fd4_context(ctx);
struct fd_ringbuffer *ring = ctx->ring;
int i;
OUT_PKT0(ring, REG_A4XX_VSC_SIZE_ADDRESS, 1);
OUT_RELOCW(ring, fd4_ctx->vsc_size_mem, 0, 0, 0); /* VSC_SIZE_ADDRESS */
OUT_PKT0(ring, REG_A4XX_VSC_PIPE_CONFIG_REG(0), 8);
for (i = 0; i < 8; i++) {
struct fd_vsc_pipe *pipe = &ctx->pipe[i];
OUT_RING(ring, A4XX_VSC_PIPE_CONFIG_REG_X(pipe->x) |
A4XX_VSC_PIPE_CONFIG_REG_Y(pipe->y) |
A4XX_VSC_PIPE_CONFIG_REG_W(pipe->w) |
A4XX_VSC_PIPE_CONFIG_REG_H(pipe->h));
}
OUT_PKT0(ring, REG_A4XX_VSC_PIPE_DATA_ADDRESS_REG(0), 8);
for (i = 0; i < 8; i++) {
struct fd_vsc_pipe *pipe = &ctx->pipe[i];
if (!pipe->bo) {
pipe->bo = fd_bo_new(ctx->dev, 0x40000,
DRM_FREEDRENO_GEM_TYPE_KMEM);
}
OUT_RELOCW(ring, pipe->bo, 0, 0, 0); /* VSC_PIPE_DATA_ADDRESS[i] */
}
OUT_PKT0(ring, REG_A4XX_VSC_PIPE_DATA_LENGTH_REG(0), 8);
for (i = 0; i < 8; i++) {
struct fd_vsc_pipe *pipe = &ctx->pipe[i];
OUT_RING(ring, fd_bo_size(pipe->bo) - 32); /* VSC_PIPE_DATA_LENGTH[i] */
}
}
/* before first tile */
static void
fd4_emit_tile_init(struct fd_context *ctx)
{
struct fd_ringbuffer *ring = ctx->ring;
struct fd_gmem_stateobj *gmem = &ctx->gmem;
uint32_t rb_render_control;
fd4_emit_restore(ctx);
OUT_PKT0(ring, REG_A4XX_VSC_BIN_SIZE, 1);
OUT_RING(ring, A4XX_VSC_BIN_SIZE_WIDTH(gmem->bin_w) |
A4XX_VSC_BIN_SIZE_HEIGHT(gmem->bin_h));
OUT_PKT0(ring, REG_A4XX_RB_MODE_CONTROL, 1);
OUT_RING(ring, A4XX_RB_MODE_CONTROL_WIDTH(gmem->bin_w) |
A4XX_RB_MODE_CONTROL_HEIGHT(gmem->bin_h) |
0x00010000); /* XXX */
update_vsc_pipe(ctx);
patch_draws(ctx, IGNORE_VISIBILITY);
rb_render_control = 0; // XXX or BINNING_PASS.. but maybe we can emit only from gmem
patch_rbrc(ctx, rb_render_control);
}
/* before mem2gmem */
static void
fd4_emit_tile_prep(struct fd_context *ctx, struct fd_tile *tile)
{
struct fd_ringbuffer *ring = ctx->ring;
struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
struct fd_gmem_stateobj *gmem = &ctx->gmem;
uint32_t reg;
OUT_PKT0(ring, REG_A4XX_RB_DEPTH_INFO, 3);
reg = A4XX_RB_DEPTH_INFO_DEPTH_BASE(depth_base(ctx));
if (pfb->zsbuf) {
reg |= A4XX_RB_DEPTH_INFO_DEPTH_FORMAT(fd4_pipe2depth(pfb->zsbuf->format));
}
OUT_RING(ring, reg);
if (pfb->zsbuf) {
uint32_t cpp = util_format_get_blocksize(pfb->zsbuf->format);
OUT_RING(ring, A4XX_RB_DEPTH_PITCH(cpp * gmem->bin_w));
OUT_RING(ring, A4XX_RB_DEPTH_PITCH2(cpp * gmem->bin_w));
} else {
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000000);
}
OUT_PKT0(ring, REG_A4XX_GRAS_DEPTH_CONTROL, 1);
if (pfb->zsbuf) {
OUT_RING(ring, A4XX_GRAS_DEPTH_CONTROL_FORMAT(
fd4_pipe2depth(pfb->zsbuf->format)));
} else {
OUT_RING(ring, A4XX_GRAS_DEPTH_CONTROL_FORMAT(DEPTH4_NONE));
}
if (ctx->needs_rb_fbd) {
fd_wfi(ctx, ring);
OUT_PKT0(ring, REG_A4XX_RB_FRAME_BUFFER_DIMENSION, 1);
OUT_RING(ring, A4XX_RB_FRAME_BUFFER_DIMENSION_WIDTH(pfb->width) |
A4XX_RB_FRAME_BUFFER_DIMENSION_HEIGHT(pfb->height));
ctx->needs_rb_fbd = false;
}
}
/* before IB to rendering cmds: */
static void
fd4_emit_tile_renderprep(struct fd_context *ctx, struct fd_tile *tile)
{
struct fd_ringbuffer *ring = ctx->ring;
struct fd_gmem_stateobj *gmem = &ctx->gmem;
struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
uint32_t x1 = tile->xoff;
uint32_t y1 = tile->yoff;
uint32_t x2 = tile->xoff + tile->bin_w - 1;
uint32_t y2 = tile->yoff + tile->bin_h - 1;
OUT_PKT3(ring, CP_SET_BIN, 3);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, CP_SET_BIN_1_X1(x1) | CP_SET_BIN_1_Y1(y1));
OUT_RING(ring, CP_SET_BIN_2_X2(x2) | CP_SET_BIN_2_Y2(y2));
emit_mrt(ring, pfb->nr_cbufs, pfb->cbufs, NULL, gmem->bin_w);
/* setup scissor/offset for current tile: */
OUT_PKT0(ring, REG_A4XX_RB_BIN_OFFSET, 1);
OUT_RING(ring, A4XX_RB_BIN_OFFSET_X(tile->xoff) |
A4XX_RB_BIN_OFFSET_Y(tile->yoff));
OUT_PKT0(ring, REG_A4XX_GRAS_SC_SCREEN_SCISSOR_TL, 2);
OUT_RING(ring, A4XX_GRAS_SC_SCREEN_SCISSOR_TL_X(x1) |
A4XX_GRAS_SC_SCREEN_SCISSOR_TL_Y(y1));
OUT_RING(ring, A4XX_GRAS_SC_SCREEN_SCISSOR_BR_X(x2) |
A4XX_GRAS_SC_SCREEN_SCISSOR_BR_Y(y2));
}
void
fd4_gmem_init(struct pipe_context *pctx)
{
struct fd_context *ctx = fd_context(pctx);
ctx->emit_sysmem_prep = fd4_emit_sysmem_prep;
ctx->emit_tile_init = fd4_emit_tile_init;
ctx->emit_tile_prep = fd4_emit_tile_prep;
ctx->emit_tile_mem2gmem = fd4_emit_tile_mem2gmem;
ctx->emit_tile_renderprep = fd4_emit_tile_renderprep;
ctx->emit_tile_gmem2mem = fd4_emit_tile_gmem2mem;
}
int
nv50_surface_do_copy(struct nv50_screen *screen, struct pipe_surface *dst,
int dx, int dy, struct pipe_surface *src, int sx, int sy,
int w, int h)
{
struct nouveau_channel *chan = screen->eng2d->channel;
struct nouveau_grobj *eng2d = screen->eng2d;
int ret;
WAIT_RING (chan, 32);
ret = nv50_surface_set(screen, dst, 1);
if (ret)
return ret;
ret = nv50_surface_set(screen, src, 0);
if (ret)
return ret;
BEGIN_RING(chan, eng2d, 0x088c, 1);
OUT_RING (chan, 0);
BEGIN_RING(chan, eng2d, NV50_2D_BLIT_DST_X, 4);
OUT_RING (chan, dx);
OUT_RING (chan, dy);
OUT_RING (chan, w);
OUT_RING (chan, h);
BEGIN_RING(chan, eng2d, 0x08c0, 4);
OUT_RING (chan, 0);
OUT_RING (chan, 1);
OUT_RING (chan, 0);
OUT_RING (chan, 1);
BEGIN_RING(chan, eng2d, 0x08d0, 4);
OUT_RING (chan, 0);
OUT_RING (chan, sx);
OUT_RING (chan, 0);
OUT_RING (chan, sy);
return 0;
}
static void
emit_mrt(struct fd_ringbuffer *ring, unsigned nr_bufs,
struct pipe_surface **bufs, uint32_t *bases, uint32_t bin_w)
{
enum a4xx_tile_mode tile_mode;
unsigned i;
if (bin_w) {
tile_mode = 2;
} else {
tile_mode = TILE4_LINEAR;
}
for (i = 0; i < 8; i++) {
enum a4xx_color_fmt format = 0;
enum a3xx_color_swap swap = WZYX;
struct fd_resource *rsc = NULL;
struct fd_resource_slice *slice = NULL;
uint32_t stride = 0;
uint32_t base = 0;
uint32_t offset = 0;
if ((i < nr_bufs) && bufs[i]) {
struct pipe_surface *psurf = bufs[i];
rsc = fd_resource(psurf->texture);
slice = fd_resource_slice(rsc, psurf->u.tex.level);
format = fd4_pipe2color(psurf->format);
swap = fd4_pipe2swap(psurf->format);
debug_assert(psurf->u.tex.first_layer == psurf->u.tex.last_layer);
offset = fd_resource_offset(rsc, psurf->u.tex.level,
psurf->u.tex.first_layer);
if (bin_w) {
stride = bin_w * rsc->cpp;
if (bases) {
base = bases[i];
}
} else {
stride = slice->pitch * rsc->cpp;
}
}
OUT_PKT0(ring, REG_A4XX_RB_MRT_BUF_INFO(i), 3);
OUT_RING(ring, A4XX_RB_MRT_BUF_INFO_COLOR_FORMAT(format) |
A4XX_RB_MRT_BUF_INFO_COLOR_TILE_MODE(tile_mode) |
A4XX_RB_MRT_BUF_INFO_COLOR_BUF_PITCH(stride) |
A4XX_RB_MRT_BUF_INFO_COLOR_SWAP(swap));
if (bin_w || (i >= nr_bufs)) {
OUT_RING(ring, base);
OUT_RING(ring, A4XX_RB_MRT_CONTROL3_STRIDE(stride));
} else {
OUT_RELOCW(ring, rsc->bo, offset, 0, 0);
/* RB_MRT[i].CONTROL3.STRIDE not emitted by c2d..
* not sure if we need to skip it for bypass or
* not.
*/
OUT_RING(ring, A4XX_RB_MRT_CONTROL3_STRIDE(0));
}
}
}
static int
nv50_surface_set(struct nv50_screen *screen, struct pipe_surface *ps, int dst)
{
struct nv50_miptree *mt = nv50_miptree(ps->texture);
struct nouveau_channel *chan = screen->eng2d->channel;
struct nouveau_grobj *eng2d = screen->eng2d;
struct nouveau_bo *bo = nv50_miptree(ps->texture)->base.bo;
int format, mthd = dst ? NV50_2D_DST_FORMAT : NV50_2D_SRC_FORMAT;
int flags = NOUVEAU_BO_VRAM | (dst ? NOUVEAU_BO_WR : NOUVEAU_BO_RD);
format = nv50_format(ps->format);
if (format < 0)
return 1;
if (!bo->tile_flags) {
BEGIN_RING(chan, eng2d, mthd, 2);
OUT_RING (chan, format);
OUT_RING (chan, 1);
BEGIN_RING(chan, eng2d, mthd + 0x14, 5);
OUT_RING (chan, mt->level[ps->level].pitch);
OUT_RING (chan, ps->width);
OUT_RING (chan, ps->height);
OUT_RELOCh(chan, bo, ps->offset, flags);
OUT_RELOCl(chan, bo, ps->offset, flags);
} else {
BEGIN_RING(chan, eng2d, mthd, 5);
OUT_RING (chan, format);
OUT_RING (chan, 0);
OUT_RING (chan, mt->level[ps->level].tile_mode << 4);
OUT_RING (chan, 1);
OUT_RING (chan, 0);
BEGIN_RING(chan, eng2d, mthd + 0x18, 4);
OUT_RING (chan, ps->width);
OUT_RING (chan, ps->height);
OUT_RELOCh(chan, bo, ps->offset, flags);
OUT_RELOCl(chan, bo, ps->offset, flags);
}
#if 0
if (dst) {
BEGIN_RING(chan, eng2d, NV50_2D_CLIP_X, 4);
OUT_RING (chan, 0);
OUT_RING (chan, 0);
OUT_RING (chan, surf->width);
OUT_RING (chan, surf->height);
}
#endif
return 0;
}
void
fd4_program_emit(struct fd_ringbuffer *ring, struct fd4_emit *emit,
int nr, struct pipe_surface **bufs)
{
struct stage s[MAX_STAGES];
uint32_t pos_regid, posz_regid, psize_regid, color_regid[8];
uint32_t face_regid, coord_regid, zwcoord_regid;
int constmode;
int i, j, k;
debug_assert(nr <= ARRAY_SIZE(color_regid));
setup_stages(emit, s);
/* blob seems to always use constmode currently: */
constmode = 1;
pos_regid = ir3_find_output_regid(s[VS].v, VARYING_SLOT_POS);
posz_regid = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DEPTH);
psize_regid = ir3_find_output_regid(s[VS].v, VARYING_SLOT_PSIZ);
if (s[FS].v->color0_mrt) {
color_regid[0] = color_regid[1] = color_regid[2] = color_regid[3] =
color_regid[4] = color_regid[5] = color_regid[6] = color_regid[7] =
ir3_find_output_regid(s[FS].v, FRAG_RESULT_COLOR);
} else {
color_regid[0] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA0);
color_regid[1] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA1);
color_regid[2] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA2);
color_regid[3] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA3);
color_regid[4] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA4);
color_regid[5] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA5);
color_regid[6] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA6);
color_regid[7] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA7);
}
/* TODO get these dynamically: */
face_regid = s[FS].v->frag_face ? regid(0,0) : regid(63,0);
coord_regid = s[FS].v->frag_coord ? regid(0,0) : regid(63,0);
zwcoord_regid = s[FS].v->frag_coord ? regid(0,2) : regid(63,0);
/* we could probably divide this up into things that need to be
* emitted if frag-prog is dirty vs if vert-prog is dirty..
*/
OUT_PKT0(ring, REG_A4XX_HLSQ_UPDATE_CONTROL, 1);
OUT_RING(ring, 0x00000003);
OUT_PKT0(ring, REG_A4XX_HLSQ_CONTROL_0_REG, 5);
OUT_RING(ring, A4XX_HLSQ_CONTROL_0_REG_FSTHREADSIZE(FOUR_QUADS) |
A4XX_HLSQ_CONTROL_0_REG_CONSTMODE(constmode) |
A4XX_HLSQ_CONTROL_0_REG_FSSUPERTHREADENABLE |
/* NOTE: I guess SHADERRESTART and CONSTFULLUPDATE maybe
* flush some caches? I think we only need to set those
* bits if we have updated const or shader..
*/
A4XX_HLSQ_CONTROL_0_REG_SPSHADERRESTART |
A4XX_HLSQ_CONTROL_0_REG_SPCONSTFULLUPDATE);
OUT_RING(ring, A4XX_HLSQ_CONTROL_1_REG_VSTHREADSIZE(TWO_QUADS) |
A4XX_HLSQ_CONTROL_1_REG_VSSUPERTHREADENABLE |
A4XX_HLSQ_CONTROL_1_REG_COORDREGID(coord_regid) |
A4XX_HLSQ_CONTROL_1_REG_ZWCOORDREGID(zwcoord_regid));
OUT_RING(ring, A4XX_HLSQ_CONTROL_2_REG_PRIMALLOCTHRESHOLD(63) |
0x3f3f000 | /* XXX */
A4XX_HLSQ_CONTROL_2_REG_FACEREGID(face_regid));
OUT_RING(ring, A4XX_HLSQ_CONTROL_3_REG_REGID(s[FS].v->pos_regid) |
0xfcfcfc00);
OUT_RING(ring, 0x00fcfcfc); /* XXX HLSQ_CONTROL_4 */
OUT_PKT0(ring, REG_A4XX_HLSQ_VS_CONTROL_REG, 5);
OUT_RING(ring, A4XX_HLSQ_VS_CONTROL_REG_CONSTLENGTH(s[VS].constlen) |
A4XX_HLSQ_VS_CONTROL_REG_CONSTOBJECTOFFSET(s[VS].constoff) |
A4XX_HLSQ_VS_CONTROL_REG_INSTRLENGTH(s[VS].instrlen) |
A4XX_HLSQ_VS_CONTROL_REG_SHADEROBJOFFSET(s[VS].instroff));
OUT_RING(ring, A4XX_HLSQ_FS_CONTROL_REG_CONSTLENGTH(s[FS].constlen) |
A4XX_HLSQ_FS_CONTROL_REG_CONSTOBJECTOFFSET(s[FS].constoff) |
A4XX_HLSQ_FS_CONTROL_REG_INSTRLENGTH(s[FS].instrlen) |
A4XX_HLSQ_FS_CONTROL_REG_SHADEROBJOFFSET(s[FS].instroff));
OUT_RING(ring, A4XX_HLSQ_HS_CONTROL_REG_CONSTLENGTH(s[HS].constlen) |
A4XX_HLSQ_HS_CONTROL_REG_CONSTOBJECTOFFSET(s[HS].constoff) |
A4XX_HLSQ_HS_CONTROL_REG_INSTRLENGTH(s[HS].instrlen) |
A4XX_HLSQ_HS_CONTROL_REG_SHADEROBJOFFSET(s[HS].instroff));
OUT_RING(ring, A4XX_HLSQ_DS_CONTROL_REG_CONSTLENGTH(s[DS].constlen) |
A4XX_HLSQ_DS_CONTROL_REG_CONSTOBJECTOFFSET(s[DS].constoff) |
A4XX_HLSQ_DS_CONTROL_REG_INSTRLENGTH(s[DS].instrlen) |
A4XX_HLSQ_DS_CONTROL_REG_SHADEROBJOFFSET(s[DS].instroff));
OUT_RING(ring, A4XX_HLSQ_GS_CONTROL_REG_CONSTLENGTH(s[GS].constlen) |
A4XX_HLSQ_GS_CONTROL_REG_CONSTOBJECTOFFSET(s[GS].constoff) |
A4XX_HLSQ_GS_CONTROL_REG_INSTRLENGTH(s[GS].instrlen) |
A4XX_HLSQ_GS_CONTROL_REG_SHADEROBJOFFSET(s[GS].instroff));
OUT_PKT0(ring, REG_A4XX_SP_SP_CTRL_REG, 1);
OUT_RING(ring, 0x140010 | /* XXX */
COND(emit->key.binning_pass, A4XX_SP_SP_CTRL_REG_BINNING_PASS));
OUT_PKT0(ring, REG_A4XX_SP_INSTR_CACHE_CTRL, 1);
OUT_RING(ring, 0x7f | /* XXX */
COND(s[VS].instrlen, A4XX_SP_INSTR_CACHE_CTRL_VS_BUFFER) |
COND(s[FS].instrlen, A4XX_SP_INSTR_CACHE_CTRL_FS_BUFFER) |
COND(s[VS].instrlen && s[FS].instrlen,
A4XX_SP_INSTR_CACHE_CTRL_INSTR_BUFFER));
OUT_PKT0(ring, REG_A4XX_SP_VS_LENGTH_REG, 1);
OUT_RING(ring, s[VS].v->instrlen); /* SP_VS_LENGTH_REG */
OUT_PKT0(ring, REG_A4XX_SP_VS_CTRL_REG0, 3);
OUT_RING(ring, A4XX_SP_VS_CTRL_REG0_THREADMODE(MULTI) |
A4XX_SP_VS_CTRL_REG0_HALFREGFOOTPRINT(s[VS].i->max_half_reg + 1) |
A4XX_SP_VS_CTRL_REG0_FULLREGFOOTPRINT(s[VS].i->max_reg + 1) |
A4XX_SP_VS_CTRL_REG0_INOUTREGOVERLAP(0) |
A4XX_SP_VS_CTRL_REG0_THREADSIZE(TWO_QUADS) |
A4XX_SP_VS_CTRL_REG0_SUPERTHREADMODE |
COND(s[VS].v->has_samp, A4XX_SP_VS_CTRL_REG0_PIXLODENABLE));
OUT_RING(ring, A4XX_SP_VS_CTRL_REG1_CONSTLENGTH(s[VS].constlen) |
A4XX_SP_VS_CTRL_REG1_INITIALOUTSTANDING(s[VS].v->total_in));
OUT_RING(ring, A4XX_SP_VS_PARAM_REG_POSREGID(pos_regid) |
A4XX_SP_VS_PARAM_REG_PSIZEREGID(psize_regid) |
A4XX_SP_VS_PARAM_REG_TOTALVSOUTVAR(align(s[FS].v->total_in, 4) / 4));
for (i = 0, j = -1; (i < 16) && (j < (int)s[FS].v->inputs_count); i++) {
uint32_t reg = 0;
OUT_PKT0(ring, REG_A4XX_SP_VS_OUT_REG(i), 1);
j = ir3_next_varying(s[FS].v, j);
if (j < s[FS].v->inputs_count) {
k = ir3_find_output(s[VS].v, s[FS].v->inputs[j].slot);
reg |= A4XX_SP_VS_OUT_REG_A_REGID(s[VS].v->outputs[k].regid);
reg |= A4XX_SP_VS_OUT_REG_A_COMPMASK(s[FS].v->inputs[j].compmask);
}
j = ir3_next_varying(s[FS].v, j);
if (j < s[FS].v->inputs_count) {
k = ir3_find_output(s[VS].v, s[FS].v->inputs[j].slot);
reg |= A4XX_SP_VS_OUT_REG_B_REGID(s[VS].v->outputs[k].regid);
reg |= A4XX_SP_VS_OUT_REG_B_COMPMASK(s[FS].v->inputs[j].compmask);
}
OUT_RING(ring, reg);
}
for (i = 0, j = -1; (i < 8) && (j < (int)s[FS].v->inputs_count); i++) {
uint32_t reg = 0;
OUT_PKT0(ring, REG_A4XX_SP_VS_VPC_DST_REG(i), 1);
j = ir3_next_varying(s[FS].v, j);
if (j < s[FS].v->inputs_count)
reg |= A4XX_SP_VS_VPC_DST_REG_OUTLOC0(s[FS].v->inputs[j].inloc);
j = ir3_next_varying(s[FS].v, j);
if (j < s[FS].v->inputs_count)
reg |= A4XX_SP_VS_VPC_DST_REG_OUTLOC1(s[FS].v->inputs[j].inloc);
j = ir3_next_varying(s[FS].v, j);
if (j < s[FS].v->inputs_count)
reg |= A4XX_SP_VS_VPC_DST_REG_OUTLOC2(s[FS].v->inputs[j].inloc);
j = ir3_next_varying(s[FS].v, j);
if (j < s[FS].v->inputs_count)
reg |= A4XX_SP_VS_VPC_DST_REG_OUTLOC3(s[FS].v->inputs[j].inloc);
OUT_RING(ring, reg);
}
OUT_PKT0(ring, REG_A4XX_SP_VS_OBJ_OFFSET_REG, 2);
OUT_RING(ring, A4XX_SP_VS_OBJ_OFFSET_REG_CONSTOBJECTOFFSET(s[VS].constoff) |
A4XX_SP_VS_OBJ_OFFSET_REG_SHADEROBJOFFSET(s[VS].instroff));
OUT_RELOC(ring, s[VS].v->bo, 0, 0, 0); /* SP_VS_OBJ_START_REG */
OUT_PKT0(ring, REG_A4XX_SP_FS_LENGTH_REG, 1);
OUT_RING(ring, s[FS].v->instrlen); /* SP_FS_LENGTH_REG */
OUT_PKT0(ring, REG_A4XX_SP_FS_CTRL_REG0, 2);
OUT_RING(ring, A4XX_SP_FS_CTRL_REG0_THREADMODE(MULTI) |
COND(s[FS].v->total_in > 0, A4XX_SP_FS_CTRL_REG0_VARYING) |
A4XX_SP_FS_CTRL_REG0_HALFREGFOOTPRINT(s[FS].i->max_half_reg + 1) |
A4XX_SP_FS_CTRL_REG0_FULLREGFOOTPRINT(s[FS].i->max_reg + 1) |
A4XX_SP_FS_CTRL_REG0_INOUTREGOVERLAP(1) |
A4XX_SP_FS_CTRL_REG0_THREADSIZE(FOUR_QUADS) |
A4XX_SP_FS_CTRL_REG0_SUPERTHREADMODE |
COND(s[FS].v->has_samp, A4XX_SP_FS_CTRL_REG0_PIXLODENABLE));
OUT_RING(ring, A4XX_SP_FS_CTRL_REG1_CONSTLENGTH(s[FS].constlen) |
0x80000000 | /* XXX */
COND(s[FS].v->frag_face, A4XX_SP_FS_CTRL_REG1_FACENESS) |
COND(s[FS].v->total_in > 0, A4XX_SP_FS_CTRL_REG1_VARYING) |
COND(s[FS].v->frag_coord, A4XX_SP_FS_CTRL_REG1_FRAGCOORD));
OUT_PKT0(ring, REG_A4XX_SP_FS_OBJ_OFFSET_REG, 2);
OUT_RING(ring, A4XX_SP_FS_OBJ_OFFSET_REG_CONSTOBJECTOFFSET(s[FS].constoff) |
A4XX_SP_FS_OBJ_OFFSET_REG_SHADEROBJOFFSET(s[FS].instroff));
if (emit->key.binning_pass)
OUT_RING(ring, 0x00000000);
else
OUT_RELOC(ring, s[FS].v->bo, 0, 0, 0); /* SP_FS_OBJ_START_REG */
OUT_PKT0(ring, REG_A4XX_SP_HS_OBJ_OFFSET_REG, 1);
OUT_RING(ring, A4XX_SP_HS_OBJ_OFFSET_REG_CONSTOBJECTOFFSET(s[HS].constoff) |
A4XX_SP_HS_OBJ_OFFSET_REG_SHADEROBJOFFSET(s[HS].instroff));
OUT_PKT0(ring, REG_A4XX_SP_DS_OBJ_OFFSET_REG, 1);
OUT_RING(ring, A4XX_SP_DS_OBJ_OFFSET_REG_CONSTOBJECTOFFSET(s[DS].constoff) |
A4XX_SP_DS_OBJ_OFFSET_REG_SHADEROBJOFFSET(s[DS].instroff));
OUT_PKT0(ring, REG_A4XX_SP_GS_OBJ_OFFSET_REG, 1);
OUT_RING(ring, A4XX_SP_GS_OBJ_OFFSET_REG_CONSTOBJECTOFFSET(s[GS].constoff) |
A4XX_SP_GS_OBJ_OFFSET_REG_SHADEROBJOFFSET(s[GS].instroff));
OUT_PKT0(ring, REG_A4XX_RB_RENDER_CONTROL2, 1);
OUT_RING(ring, A4XX_RB_RENDER_CONTROL2_MSAA_SAMPLES(0) |
COND(s[FS].v->total_in > 0, A4XX_RB_RENDER_CONTROL2_VARYING) |
COND(s[FS].v->frag_face, A4XX_RB_RENDER_CONTROL2_FACENESS) |
COND(s[FS].v->frag_coord, A4XX_RB_RENDER_CONTROL2_XCOORD |
A4XX_RB_RENDER_CONTROL2_YCOORD |
// TODO enabling gl_FragCoord.z is causing lockups on 0ad (but seems
// to work everywhere else).
// A4XX_RB_RENDER_CONTROL2_ZCOORD |
A4XX_RB_RENDER_CONTROL2_WCOORD));
OUT_PKT0(ring, REG_A4XX_RB_FS_OUTPUT_REG, 1);
OUT_RING(ring, A4XX_RB_FS_OUTPUT_REG_MRT(MAX2(1, nr)) |
COND(s[FS].v->writes_pos, A4XX_RB_FS_OUTPUT_REG_FRAG_WRITES_Z));
OUT_PKT0(ring, REG_A4XX_SP_FS_OUTPUT_REG, 1);
OUT_RING(ring, A4XX_SP_FS_OUTPUT_REG_MRT(MAX2(1, nr)) |
COND(s[FS].v->writes_pos, A4XX_SP_FS_OUTPUT_REG_DEPTH_ENABLE) |
A4XX_SP_FS_OUTPUT_REG_DEPTH_REGID(posz_regid));
OUT_PKT0(ring, REG_A4XX_SP_FS_MRT_REG(0), 8);
for (i = 0; i < 8; i++) {
enum a4xx_color_fmt format = 0;
bool srgb = false;
if (i < nr) {
format = fd4_emit_format(bufs[i]);
if (bufs[i] && !emit->no_decode_srgb)
srgb = util_format_is_srgb(bufs[i]->format);
}
OUT_RING(ring, A4XX_SP_FS_MRT_REG_REGID(color_regid[i]) |
A4XX_SP_FS_MRT_REG_MRTFORMAT(format) |
COND(srgb, A4XX_SP_FS_MRT_REG_COLOR_SRGB) |
COND(emit->key.half_precision,
A4XX_SP_FS_MRT_REG_HALF_PRECISION));
}
if (emit->key.binning_pass) {
OUT_PKT0(ring, REG_A4XX_VPC_ATTR, 2);
OUT_RING(ring, A4XX_VPC_ATTR_THRDASSIGN(1) |
0x40000000 | /* XXX */
COND(s[VS].v->writes_psize, A4XX_VPC_ATTR_PSIZE));
OUT_RING(ring, 0x00000000);
} else {
uint32_t vinterp[8], vpsrepl[8];
memset(vinterp, 0, sizeof(vinterp));
memset(vpsrepl, 0, sizeof(vpsrepl));
/* looks like we need to do int varyings in the frag
* shader on a4xx (no flatshad reg? or a420.0 bug?):
*
* (sy)(ss)nop
* (sy)ldlv.u32 r0.x,l[r0.x], 1
* ldlv.u32 r0.y,l[r0.x+1], 1
* (ss)bary.f (ei)r63.x, 0, r0.x
* (ss)(rpt1)cov.s32f16 hr0.x, (r)r0.x
* (rpt5)nop
* sam (f16)(xyzw)hr0.x, hr0.x, s#0, t#0
*
* Possibly on later a4xx variants we'll be able to use
* something like the code below instead of workaround
* in the shader:
*/
/* figure out VARYING_INTERP / VARYING_PS_REPL register values: */
for (j = -1; (j = ir3_next_varying(s[FS].v, j)) < (int)s[FS].v->inputs_count; ) {
/* TODO might be cleaner to just +8 in SP_VS_VPC_DST_REG
* instead.. rather than -8 everywhere else..
*/
uint32_t inloc = s[FS].v->inputs[j].inloc - 8;
/* currently assuming varyings aligned to 4 (not
* packed):
*/
debug_assert((inloc % 4) == 0);
if ((s[FS].v->inputs[j].interpolate == INTERP_QUALIFIER_FLAT) ||
(s[FS].v->inputs[j].rasterflat && emit->rasterflat)) {
uint32_t loc = inloc;
for (i = 0; i < 4; i++, loc++) {
vinterp[loc / 16] |= 1 << ((loc % 16) * 2);
//flatshade[loc / 32] |= 1 << (loc % 32);
}
}
gl_varying_slot slot = s[FS].v->inputs[j].slot;
/* since we don't enable PIPE_CAP_TGSI_TEXCOORD: */
if (slot >= VARYING_SLOT_VAR0) {
unsigned texmask = 1 << (slot - VARYING_SLOT_VAR0);
/* Replace the .xy coordinates with S/T from the point sprite. Set
* interpolation bits for .zw such that they become .01
*/
if (emit->sprite_coord_enable & texmask) {
vpsrepl[inloc / 16] |= (emit->sprite_coord_mode ? 0x0d : 0x09)
<< ((inloc % 16) * 2);
vinterp[(inloc + 2) / 16] |= 2 << (((inloc + 2) % 16) * 2);
vinterp[(inloc + 3) / 16] |= 3 << (((inloc + 3) % 16) * 2);
}
}
}
OUT_PKT0(ring, REG_A4XX_VPC_ATTR, 2);
OUT_RING(ring, A4XX_VPC_ATTR_TOTALATTR(s[FS].v->total_in) |
A4XX_VPC_ATTR_THRDASSIGN(1) |
COND(s[FS].v->total_in > 0, A4XX_VPC_ATTR_ENABLE) |
0x40000000 | /* XXX */
COND(s[VS].v->writes_psize, A4XX_VPC_ATTR_PSIZE));
OUT_RING(ring, A4XX_VPC_PACK_NUMFPNONPOSVAR(s[FS].v->total_in) |
A4XX_VPC_PACK_NUMNONPOSVSVAR(s[FS].v->total_in));
OUT_PKT0(ring, REG_A4XX_VPC_VARYING_INTERP_MODE(0), 8);
for (i = 0; i < 8; i++)
OUT_RING(ring, vinterp[i]); /* VPC_VARYING_INTERP[i].MODE */
OUT_PKT0(ring, REG_A4XX_VPC_VARYING_PS_REPL_MODE(0), 8);
for (i = 0; i < 8; i++)
OUT_RING(ring, vpsrepl[i]); /* VPC_VARYING_PS_REPL[i] */
}
if (s[VS].instrlen)
emit_shader(ring, s[VS].v);
if (!emit->key.binning_pass)
if (s[FS].instrlen)
emit_shader(ring, s[FS].v);
}
void
fd3_program_emit(struct fd_ringbuffer *ring, struct fd3_emit *emit,
int nr, struct pipe_surface **bufs)
{
const struct ir3_shader_variant *vp, *fp;
const struct ir3_info *vsi, *fsi;
enum a3xx_instrbuffermode fpbuffer, vpbuffer;
uint32_t fpbuffersz, vpbuffersz, fsoff;
uint32_t pos_regid, posz_regid, psize_regid;
uint32_t vcoord_regid, face_regid, coord_regid, zwcoord_regid;
uint32_t color_regid[4] = {0};
int constmode;
int i, j;
debug_assert(nr <= ARRAY_SIZE(color_regid));
vp = fd3_emit_get_vp(emit);
fp = fd3_emit_get_fp(emit);
vsi = &vp->info;
fsi = &fp->info;
fpbuffer = BUFFER;
vpbuffer = BUFFER;
fpbuffersz = fp->instrlen;
vpbuffersz = vp->instrlen;
/*
* Decide whether to use BUFFER or CACHE mode for VS and FS. It
* appears like 256 is the hard limit, but when the combined size
* exceeds 128 then blob will try to keep FS in BUFFER mode and
* switch to CACHE for VS until VS is too large. The blob seems
* to switch FS out of BUFFER mode at slightly under 128. But
* a bit fuzzy on the decision tree, so use slightly conservative
* limits.
*
* TODO check if these thresholds for BUFFER vs CACHE mode are the
* same for all a3xx or whether we need to consider the gpuid
*/
if ((fpbuffersz + vpbuffersz) > 128) {
if (fpbuffersz < 112) {
/* FP:BUFFER VP:CACHE */
vpbuffer = CACHE;
vpbuffersz = 256 - fpbuffersz;
} else if (vpbuffersz < 112) {
/* FP:CACHE VP:BUFFER */
fpbuffer = CACHE;
fpbuffersz = 256 - vpbuffersz;
} else {
/* FP:CACHE VP:CACHE */
vpbuffer = fpbuffer = CACHE;
vpbuffersz = fpbuffersz = 192;
}
}
if (fpbuffer == BUFFER) {
fsoff = 128 - fpbuffersz;
} else {
fsoff = 256 - fpbuffersz;
}
/* seems like vs->constlen + fs->constlen > 256, then CONSTMODE=1 */
constmode = ((vp->constlen + fp->constlen) > 256) ? 1 : 0;
pos_regid = ir3_find_output_regid(vp, VARYING_SLOT_POS);
posz_regid = ir3_find_output_regid(fp, FRAG_RESULT_DEPTH);
psize_regid = ir3_find_output_regid(vp, VARYING_SLOT_PSIZ);
if (fp->color0_mrt) {
color_regid[0] = color_regid[1] = color_regid[2] = color_regid[3] =
ir3_find_output_regid(fp, FRAG_RESULT_COLOR);
} else {
color_regid[0] = ir3_find_output_regid(fp, FRAG_RESULT_DATA0);
color_regid[1] = ir3_find_output_regid(fp, FRAG_RESULT_DATA1);
color_regid[2] = ir3_find_output_regid(fp, FRAG_RESULT_DATA2);
color_regid[3] = ir3_find_output_regid(fp, FRAG_RESULT_DATA3);
}
face_regid = ir3_find_sysval_regid(fp, SYSTEM_VALUE_FRONT_FACE);
coord_regid = ir3_find_sysval_regid(fp, SYSTEM_VALUE_FRAG_COORD);
zwcoord_regid = (coord_regid == regid(63,0)) ? regid(63,0) : (coord_regid + 2);
vcoord_regid = ir3_find_sysval_regid(fp, SYSTEM_VALUE_BARYCENTRIC_PIXEL);
/* adjust regids for alpha output formats. there is no alpha render
* format, so it's just treated like red
*/
for (i = 0; i < nr; i++)
if (util_format_is_alpha(pipe_surface_format(bufs[i])))
color_regid[i] += 3;
/* we could probably divide this up into things that need to be
* emitted if frag-prog is dirty vs if vert-prog is dirty..
*/
OUT_PKT0(ring, REG_A3XX_HLSQ_CONTROL_0_REG, 6);
OUT_RING(ring, A3XX_HLSQ_CONTROL_0_REG_FSTHREADSIZE(FOUR_QUADS) |
A3XX_HLSQ_CONTROL_0_REG_FSSUPERTHREADENABLE |
A3XX_HLSQ_CONTROL_0_REG_CONSTMODE(constmode) |
/* NOTE: I guess SHADERRESTART and CONSTFULLUPDATE maybe
* flush some caches? I think we only need to set those
* bits if we have updated const or shader..
*/
A3XX_HLSQ_CONTROL_0_REG_SPSHADERRESTART |
A3XX_HLSQ_CONTROL_0_REG_SPCONSTFULLUPDATE);
OUT_RING(ring, A3XX_HLSQ_CONTROL_1_REG_VSTHREADSIZE(TWO_QUADS) |
A3XX_HLSQ_CONTROL_1_REG_VSSUPERTHREADENABLE |
A3XX_HLSQ_CONTROL_1_REG_FRAGCOORDXYREGID(coord_regid) |
A3XX_HLSQ_CONTROL_1_REG_FRAGCOORDZWREGID(zwcoord_regid));
OUT_RING(ring, A3XX_HLSQ_CONTROL_2_REG_PRIMALLOCTHRESHOLD(31) |
A3XX_HLSQ_CONTROL_2_REG_FACENESSREGID(face_regid));
OUT_RING(ring, A3XX_HLSQ_CONTROL_3_REG_REGID(vcoord_regid));
OUT_RING(ring, A3XX_HLSQ_VS_CONTROL_REG_CONSTLENGTH(vp->constlen) |
A3XX_HLSQ_VS_CONTROL_REG_CONSTSTARTOFFSET(0) |
A3XX_HLSQ_VS_CONTROL_REG_INSTRLENGTH(vpbuffersz));
OUT_RING(ring, A3XX_HLSQ_FS_CONTROL_REG_CONSTLENGTH(fp->constlen) |
A3XX_HLSQ_FS_CONTROL_REG_CONSTSTARTOFFSET(128) |
A3XX_HLSQ_FS_CONTROL_REG_INSTRLENGTH(fpbuffersz));
OUT_PKT0(ring, REG_A3XX_SP_SP_CTRL_REG, 1);
OUT_RING(ring, A3XX_SP_SP_CTRL_REG_CONSTMODE(constmode) |
COND(emit->binning_pass, A3XX_SP_SP_CTRL_REG_BINNING) |
A3XX_SP_SP_CTRL_REG_SLEEPMODE(1) |
A3XX_SP_SP_CTRL_REG_L0MODE(0));
OUT_PKT0(ring, REG_A3XX_SP_VS_LENGTH_REG, 1);
OUT_RING(ring, A3XX_SP_VS_LENGTH_REG_SHADERLENGTH(vp->instrlen));
OUT_PKT0(ring, REG_A3XX_SP_VS_CTRL_REG0, 3);
OUT_RING(ring, A3XX_SP_VS_CTRL_REG0_THREADMODE(MULTI) |
A3XX_SP_VS_CTRL_REG0_INSTRBUFFERMODE(vpbuffer) |
COND(vpbuffer == CACHE, A3XX_SP_VS_CTRL_REG0_CACHEINVALID) |
A3XX_SP_VS_CTRL_REG0_HALFREGFOOTPRINT(vsi->max_half_reg + 1) |
A3XX_SP_VS_CTRL_REG0_FULLREGFOOTPRINT(vsi->max_reg + 1) |
A3XX_SP_VS_CTRL_REG0_THREADSIZE(TWO_QUADS) |
A3XX_SP_VS_CTRL_REG0_SUPERTHREADMODE |
A3XX_SP_VS_CTRL_REG0_LENGTH(vpbuffersz));
OUT_RING(ring, A3XX_SP_VS_CTRL_REG1_CONSTLENGTH(vp->constlen) |
A3XX_SP_VS_CTRL_REG1_INITIALOUTSTANDING(vp->total_in) |
A3XX_SP_VS_CTRL_REG1_CONSTFOOTPRINT(MAX2(vp->constlen + 1, 0)));
OUT_RING(ring, A3XX_SP_VS_PARAM_REG_POSREGID(pos_regid) |
A3XX_SP_VS_PARAM_REG_PSIZEREGID(psize_regid) |
A3XX_SP_VS_PARAM_REG_TOTALVSOUTVAR(fp->varying_in));
struct ir3_shader_linkage l = {0};
ir3_link_shaders(&l, vp, fp);
for (i = 0, j = 0; (i < 16) && (j < l.cnt); i++) {
uint32_t reg = 0;
OUT_PKT0(ring, REG_A3XX_SP_VS_OUT_REG(i), 1);
reg |= A3XX_SP_VS_OUT_REG_A_REGID(l.var[j].regid);
reg |= A3XX_SP_VS_OUT_REG_A_COMPMASK(l.var[j].compmask);
j++;
reg |= A3XX_SP_VS_OUT_REG_B_REGID(l.var[j].regid);
reg |= A3XX_SP_VS_OUT_REG_B_COMPMASK(l.var[j].compmask);
j++;
OUT_RING(ring, reg);
}
for (i = 0, j = 0; (i < 8) && (j < l.cnt); i++) {
uint32_t reg = 0;
OUT_PKT0(ring, REG_A3XX_SP_VS_VPC_DST_REG(i), 1);
reg |= A3XX_SP_VS_VPC_DST_REG_OUTLOC0(l.var[j++].loc + 8);
reg |= A3XX_SP_VS_VPC_DST_REG_OUTLOC1(l.var[j++].loc + 8);
reg |= A3XX_SP_VS_VPC_DST_REG_OUTLOC2(l.var[j++].loc + 8);
reg |= A3XX_SP_VS_VPC_DST_REG_OUTLOC3(l.var[j++].loc + 8);
OUT_RING(ring, reg);
}
OUT_PKT0(ring, REG_A3XX_SP_VS_OBJ_OFFSET_REG, 2);
OUT_RING(ring, A3XX_SP_VS_OBJ_OFFSET_REG_CONSTOBJECTOFFSET(0) |
A3XX_SP_VS_OBJ_OFFSET_REG_SHADEROBJOFFSET(0));
OUT_RELOC(ring, vp->bo, 0, 0, 0); /* SP_VS_OBJ_START_REG */
if (emit->binning_pass) {
OUT_PKT0(ring, REG_A3XX_SP_FS_LENGTH_REG, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A3XX_SP_FS_CTRL_REG0, 2);
OUT_RING(ring, A3XX_SP_FS_CTRL_REG0_THREADMODE(MULTI) |
A3XX_SP_FS_CTRL_REG0_INSTRBUFFERMODE(BUFFER));
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A3XX_SP_FS_OBJ_OFFSET_REG, 1);
OUT_RING(ring, A3XX_SP_FS_OBJ_OFFSET_REG_CONSTOBJECTOFFSET(128) |
A3XX_SP_FS_OBJ_OFFSET_REG_SHADEROBJOFFSET(0));
} else {
OUT_PKT0(ring, REG_A3XX_SP_FS_LENGTH_REG, 1);
OUT_RING(ring, A3XX_SP_FS_LENGTH_REG_SHADERLENGTH(fp->instrlen));
OUT_PKT0(ring, REG_A3XX_SP_FS_CTRL_REG0, 2);
OUT_RING(ring, A3XX_SP_FS_CTRL_REG0_THREADMODE(MULTI) |
A3XX_SP_FS_CTRL_REG0_INSTRBUFFERMODE(fpbuffer) |
COND(fpbuffer == CACHE, A3XX_SP_FS_CTRL_REG0_CACHEINVALID) |
A3XX_SP_FS_CTRL_REG0_HALFREGFOOTPRINT(fsi->max_half_reg + 1) |
A3XX_SP_FS_CTRL_REG0_FULLREGFOOTPRINT(fsi->max_reg + 1) |
A3XX_SP_FS_CTRL_REG0_INOUTREGOVERLAP |
A3XX_SP_FS_CTRL_REG0_THREADSIZE(FOUR_QUADS) |
A3XX_SP_FS_CTRL_REG0_SUPERTHREADMODE |
COND(fp->num_samp > 0, A3XX_SP_FS_CTRL_REG0_PIXLODENABLE) |
A3XX_SP_FS_CTRL_REG0_LENGTH(fpbuffersz));
OUT_RING(ring, A3XX_SP_FS_CTRL_REG1_CONSTLENGTH(fp->constlen) |
A3XX_SP_FS_CTRL_REG1_INITIALOUTSTANDING(fp->total_in) |
A3XX_SP_FS_CTRL_REG1_CONSTFOOTPRINT(MAX2(fp->constlen + 1, 0)) |
A3XX_SP_FS_CTRL_REG1_HALFPRECVAROFFSET(63));
OUT_PKT0(ring, REG_A3XX_SP_FS_OBJ_OFFSET_REG, 2);
OUT_RING(ring, A3XX_SP_FS_OBJ_OFFSET_REG_CONSTOBJECTOFFSET(
MAX2(128, vp->constlen)) |
A3XX_SP_FS_OBJ_OFFSET_REG_SHADEROBJOFFSET(fsoff));
OUT_RELOC(ring, fp->bo, 0, 0, 0); /* SP_FS_OBJ_START_REG */
}
OUT_PKT0(ring, REG_A3XX_SP_FS_OUTPUT_REG, 1);
OUT_RING(ring,
COND(fp->writes_pos, A3XX_SP_FS_OUTPUT_REG_DEPTH_ENABLE) |
A3XX_SP_FS_OUTPUT_REG_DEPTH_REGID(posz_regid) |
A3XX_SP_FS_OUTPUT_REG_MRT(MAX2(1, nr) - 1));
OUT_PKT0(ring, REG_A3XX_SP_FS_MRT_REG(0), 4);
for (i = 0; i < 4; i++) {
uint32_t mrt_reg = A3XX_SP_FS_MRT_REG_REGID(color_regid[i]) |
COND(fp->key.half_precision, A3XX_SP_FS_MRT_REG_HALF_PRECISION);
if (i < nr) {
enum pipe_format fmt = pipe_surface_format(bufs[i]);
mrt_reg |= COND(util_format_is_pure_uint(fmt), A3XX_SP_FS_MRT_REG_UINT) |
COND(util_format_is_pure_sint(fmt), A3XX_SP_FS_MRT_REG_SINT);
}
OUT_RING(ring, mrt_reg);
}
if (emit->binning_pass) {
OUT_PKT0(ring, REG_A3XX_VPC_ATTR, 2);
OUT_RING(ring, A3XX_VPC_ATTR_THRDASSIGN(1) |
A3XX_VPC_ATTR_LMSIZE(1) |
COND(vp->writes_psize, A3XX_VPC_ATTR_PSIZE));
OUT_RING(ring, 0x00000000);
} else {
uint32_t vinterp[4], flatshade[2], vpsrepl[4];
memset(vinterp, 0, sizeof(vinterp));
memset(flatshade, 0, sizeof(flatshade));
memset(vpsrepl, 0, sizeof(vpsrepl));
/* figure out VARYING_INTERP / FLAT_SHAD register values: */
for (j = -1; (j = ir3_next_varying(fp, j)) < (int)fp->inputs_count; ) {
/* NOTE: varyings are packed, so if compmask is 0xb
* then first, third, and fourth component occupy
* three consecutive varying slots:
*/
unsigned compmask = fp->inputs[j].compmask;
uint32_t inloc = fp->inputs[j].inloc;
if ((fp->inputs[j].interpolate == INTERP_MODE_FLAT) ||
(fp->inputs[j].rasterflat && emit->rasterflat)) {
uint32_t loc = inloc;
for (i = 0; i < 4; i++) {
if (compmask & (1 << i)) {
vinterp[loc / 16] |= FLAT << ((loc % 16) * 2);
flatshade[loc / 32] |= 1 << (loc % 32);
loc++;
}
}
}
gl_varying_slot slot = fp->inputs[j].slot;
/* since we don't enable PIPE_CAP_TGSI_TEXCOORD: */
if (slot >= VARYING_SLOT_VAR0) {
unsigned texmask = 1 << (slot - VARYING_SLOT_VAR0);
/* Replace the .xy coordinates with S/T from the point sprite. Set
* interpolation bits for .zw such that they become .01
*/
if (emit->sprite_coord_enable & texmask) {
/* mask is two 2-bit fields, where:
* '01' -> S
* '10' -> T
* '11' -> 1 - T (flip mode)
*/
unsigned mask = emit->sprite_coord_mode ? 0b1101 : 0b1001;
uint32_t loc = inloc;
if (compmask & 0x1) {
vpsrepl[loc / 16] |= ((mask >> 0) & 0x3) << ((loc % 16) * 2);
loc++;
}
if (compmask & 0x2) {
vpsrepl[loc / 16] |= ((mask >> 2) & 0x3) << ((loc % 16) * 2);
loc++;
}
if (compmask & 0x4) {
/* .z <- 0.0f */
vinterp[loc / 16] |= 0b10 << ((loc % 16) * 2);
loc++;
}
if (compmask & 0x8) {
/* .w <- 1.0f */
vinterp[loc / 16] |= 0b11 << ((loc % 16) * 2);
loc++;
}
}
