KonstantinSeurer
/
mesa
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

nv50: fix alphatest for non-blendable formats 

The hardware can only do alphatest when using a blendable format. This
means that the various *16 norm formats didn't work with alphatest. It
appears that Talos Principle uses such formats, as well as alpha tests,
for some internal renders, which made them be incorrect. However this
does not appear to affect the final renders, but in a different game it
easily could.

The approach we take is that when alphatests are enabled and a suitable
format is used (which we anticipate is the vast minority of the time),
we insert code into the shader to perform the comparison and discard.
Once inserted, that code lives in the shader forever, and we re-upload
it each time the function changes with a fixed-up compare. To avoid
re-uploading too often, if we switch back to a blendable format, the
test is (effectively) disabled and the hw alphatest functionality is
used.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
This commit is contained in:
Ilia Mirkin 2016-06-19 16:57:50 -04:00
parent cc46fc3c09
commit 062c6b8e54
14 changed files with 118 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
src/gallium/drivers/nouveau/codegen/nv50_ir_driver.h
View File

@ -166,6 +166,7 @@ struct nv50_ir_prog_info
uint8_t auxCBSlot; /* driver constant buffer slot */
uint16_t ucpBase; /* base address for UCPs */
uint16_t drawInfoBase; /* base address for draw parameters */
uint16_t alphaRefBase; /* base address for alpha test values */
uint8_t pointSize; /* output index for PointSize */
uint8_t instanceId; /* system value index of InstanceID */
uint8_t vertexId; /* system value index of VertexID */
@ -206,7 +207,8 @@ extern void nv50_ir_relocate_code(void *relocData, uint32_t *code,
extern void
nv50_ir_apply_fixups(void *fixupData, uint32_t *code,
bool force_per_sample, bool flatshade);
bool force_per_sample, bool flatshade,
uint8_t alphatest);
/* obtain code that will be shared among programs */
extern void nv50_ir_get_target_library(uint32_t chipset,

26
src/gallium/drivers/nouveau/codegen/nv50_ir_emit_nv50.cpp
View File

@ -1265,6 +1265,28 @@ CodeEmitterNV50::emitISAD(const Instruction *i)
}
}
static void
alphatestSet(const FixupEntry *entry, uint32_t *code, const FixupData& data)
{
int loc = entry->loc;
int enc;
switch (data.alphatest) {
case PIPE_FUNC_NEVER: enc = 0x0; break;
case PIPE_FUNC_LESS: enc = 0x1; break;
case PIPE_FUNC_EQUAL: enc = 0x2; break;
case PIPE_FUNC_LEQUAL: enc = 0x3; break;
case PIPE_FUNC_GREATER: enc = 0x4; break;
case PIPE_FUNC_NOTEQUAL: enc = 0x5; break;
case PIPE_FUNC_GEQUAL: enc = 0x6; break;
default:
case PIPE_FUNC_ALWAYS: enc = 0xf; break;
}
code[loc + 1] &= ~(0x1f << 14);
code[loc + 1] |= enc << 14;
}
void
CodeEmitterNV50::emitSET(const Instruction *i)
{
@ -1294,6 +1316,10 @@ CodeEmitterNV50::emitSET(const Instruction *i)
if (i->src(1).mod.abs()) code[1] |= 0x00080000;
emitForm_MAD(i);
if (i->subOp == 1) {
addInterp(0, 0, alphatestSet);
}
}
void

24
src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp
View File

@ -1006,7 +1006,7 @@ bool Source::scanSource()
if (info->type == PIPE_SHADER_FRAGMENT) {
info->prop.fp.writesDepth = scan.writes_z;
info->prop.fp.usesDiscard = scan.uses_kill;
info->prop.fp.usesDiscard = scan.uses_kill || info->io.alphaRefBase;
} else
if (info->type == PIPE_SHADER_GEOMETRY) {
info->prop.gp.instanceCount = 1; // default value
@ -3818,6 +3818,28 @@ Converter::handleUserClipPlanes()
void
Converter::exportOutputs()
{
if (info->io.alphaRefBase) {
for (unsigned int i = 0; i < info->numOutputs; ++i) {
if (info->out[i].sn != TGSI_SEMANTIC_COLOR ||
info->out[i].si != 0)
continue;
const unsigned int c = 3;
if (!oData.exists(sub.cur->values, i, c))
continue;
Value *val = oData.load(sub.cur->values, i, c, NULL);
if (!val)
continue;
Symbol *ref = mkSymbol(FILE_MEMORY_CONST, info->io.auxCBSlot,
TYPE_U32, info->io.alphaRefBase);
Value *pred = new_LValue(func, FILE_PREDICATE);
mkCmp(OP_SET, CC_TR, TYPE_U32, pred, TYPE_F32, val,
mkLoadv(TYPE_U32, ref, NULL))
->subOp = 1;
mkOp(OP_DISCARD, TYPE_NONE, NULL)->setPredicate(CC_NOT_P, pred);
}
}
for (unsigned int i = 0; i < info->numOutputs; ++i) {
for (unsigned int c = 0; c < 4; ++c) {
if (!oData.exists(sub.cur->values, i, c))

4
src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
View File

@ -196,6 +196,10 @@ LoadPropagation::checkSwapSrc01(Instruction *insn)
return;
if (insn->src(1).getFile() != FILE_GPR)
return;
// This is the special OP_SET used for alphatesting, we can't reverse its
// arguments as that will confuse the fixup code.
if (insn->op == OP_SET && insn->subOp)
return;
Instruction *i0 = insn->getSrc(0)->getInsn();
Instruction *i1 = insn->getSrc(1)->getInsn();

6
src/gallium/drivers/nouveau/codegen/nv50_ir_target.cpp
View File

@ -506,14 +506,16 @@ nv50_ir_relocate_code(void *relocData, uint32_t *code,
void
nv50_ir_apply_fixups(void *fixupData, uint32_t *code,
bool force_persample_interp, bool flatshade)
bool force_persample_interp, bool flatshade,
uint8_t alphatest)
{
nv50_ir::FixupInfo *info = reinterpret_cast<nv50_ir::FixupInfo *>(
fixupData);
// force_persample_interp: all non-flat -> per-sample
// flatshade: all color -> flat
nv50_ir::FixupData data(force_persample_interp, flatshade);
// alphatest: PIPE_FUNC_* to use with alphatest
nv50_ir::FixupData data(force_persample_interp, flatshade, alphatest);
for (unsigned i = 0; i < info->count; ++i)
info->entry[i].apply(&info->entry[i], code, data);
}

5
src/gallium/drivers/nouveau/codegen/nv50_ir_target.h
View File

@ -59,10 +59,11 @@ struct RelocInfo
};
struct FixupData {
FixupData(bool force, bool flat) :
force_persample_interp(force), flatshade(flat) {}
FixupData(bool force, bool flat, uint8_t alphatest) :
force_persample_interp(force), flatshade(flat), alphatest(alphatest) {}
bool force_persample_interp;
bool flatshade;
uint8_t alphatest;
};
struct FixupEntry;

5
src/gallium/drivers/nouveau/nv50/nv50_context.h
View File

@ -97,7 +97,10 @@
/* Sample position pairs for the current output MS level */
#define NV50_CB_AUX_SAMPLE_OFFSET 0x300
#define NV50_CB_AUX_SAMPLE_OFFSET_SIZE (4 * 8 * 2)
/* next spot: 0x340 */
/* Alpha test ref value */
#define NV50_CB_AUX_ALPHATEST_OFFSET 0x340
#define NV50_CB_AUX_ALPHATEST_SIZE (4)
/* next spot: 0x344 */
/* 4 32-bit floats for the vertex runout, put at the end */
#define NV50_CB_AUX_RUNOUT_OFFSET (NV50_CB_AUX_SIZE - 0x10)

5
src/gallium/drivers/nouveau/nv50/nv50_program.c
View File

@ -334,6 +334,8 @@ nv50_program_translate(struct nv50_program *prog, uint16_t chipset,
info->io.auxCBSlot = 15;
info->io.ucpBase = NV50_CB_AUX_UCP_OFFSET;
info->io.genUserClip = prog->vp.clpd_nr;
if (prog->fp.alphatest)
info->io.alphaRefBase = NV50_CB_AUX_ALPHATEST_OFFSET;
info->io.suInfoBase = NV50_CB_AUX_TEX_MS_OFFSET;
info->io.sampleInfoBase = NV50_CB_AUX_SAMPLE_OFFSET;
@ -488,7 +490,8 @@ nv50_program_upload_code(struct nv50_context *nv50, struct nv50_program *prog)
if (prog->interps)
nv50_ir_apply_fixups(prog->interps, prog->code,
prog->fp.force_persample_interp,
false /* flatshade */);
false /* flatshade */,
prog->fp.alphatest - 1);
nv50_sifc_linear_u8(&nv50->base, nv50->screen->code,
(prog_type << NV50_CODE_BO_SIZE_LOG2) + prog->code_base,

1
src/gallium/drivers/nouveau/nv50/nv50_program.h
View File

@ -90,6 +90,7 @@ struct nv50_program {
uint32_t colors; /* 0x1904 */
uint8_t has_samplemask;
uint8_t force_persample_interp;
uint8_t alphatest;
} fp;
struct {

36
src/gallium/drivers/nouveau/nv50/nv50_shader_state.c
View File

@ -174,6 +174,42 @@ nv50_fragprog_validate(struct nv50_context *nv50)
struct nv50_program *fp = nv50->fragprog;
struct pipe_rasterizer_state *rast = &nv50->rast->pipe;
if (nv50->zsa && nv50->zsa->pipe.alpha.enabled) {
struct pipe_framebuffer_state *fb = &nv50->framebuffer;
bool blendable = fb->nr_cbufs == 0 || !fb->cbufs[0] ||
nv50->screen->base.base.is_format_supported(
&nv50->screen->base.base,
fb->cbufs[0]->format,
fb->cbufs[0]->texture->target,
fb->cbufs[0]->texture->nr_samples,
PIPE_BIND_BLENDABLE);
/* If we already have alphatest code, we have to keep updating
* it. However we only have to have different code if the current RT0 is
* non-blendable. Otherwise we just set it to always pass and use the
* hardware alpha test.
*/
if (fp->fp.alphatest || !blendable) {
uint8_t alphatest = PIPE_FUNC_ALWAYS + 1;
if (!blendable)
alphatest = nv50->zsa->pipe.alpha.func + 1;
if (!fp->fp.alphatest)
nv50_program_destroy(nv50, fp);
else if (fp->mem && fp->fp.alphatest != alphatest)
nouveau_heap_free(&fp->mem);
fp->fp.alphatest = alphatest;
}
} else if (fp->fp.alphatest && fp->fp.alphatest != PIPE_FUNC_ALWAYS + 1) {
/* Alpha test is disabled but we have a shader where it's filled
* in. Make sure to reset the function to 'always', otherwise it'll end
* up discarding fragments incorrectly.
*/
if (fp->mem)
nouveau_heap_free(&fp->mem);
fp->fp.alphatest = PIPE_FUNC_ALWAYS + 1;
}
if (fp->fp.force_persample_interp != rast->force_persample_interp) {
/* Force the program to be reuploaded, which will trigger interp fixups
* to get applied

5
src/gallium/drivers/nouveau/nv50/nv50_state.c
View File

@ -416,6 +416,11 @@ nv50_zsa_state_create(struct pipe_context *pipe,
SB_DATA (so, 0);
}
SB_BEGIN_3D(so, CB_ADDR, 1);
SB_DATA (so, NV50_CB_AUX_ALPHATEST_OFFSET << (8 - 2) | NV50_CB_AUX);
SB_BEGIN_3D(so, CB_DATA(0), 1);
SB_DATA (so, fui(cso->alpha.ref_value));
assert(so->size <= ARRAY_SIZE(so->state));
return (void *)so;
}

3
src/gallium/drivers/nouveau/nv50/nv50_state_validate.c
View File

@ -522,7 +522,8 @@ validate_list_3d[] = {
{ nv50_vertprog_validate, NV50_NEW_3D_VERTPROG },
{ nv50_gmtyprog_validate, NV50_NEW_3D_GMTYPROG },
{ nv50_fragprog_validate, NV50_NEW_3D_FRAGPROG | NV50_NEW_3D_RASTERIZER |
NV50_NEW_3D_MIN_SAMPLES },
NV50_NEW_3D_MIN_SAMPLES | NV50_NEW_3D_ZSA |
NV50_NEW_3D_FRAMEBUFFER},
{ nv50_fp_linkage_validate, NV50_NEW_3D_FRAGPROG | NV50_NEW_3D_VERTPROG |
NV50_NEW_3D_GMTYPROG | NV50_NEW_3D_RASTERIZER },
{ nv50_gp_linkage_validate, NV50_NEW_3D_GMTYPROG | NV50_NEW_3D_VERTPROG },

2
src/gallium/drivers/nouveau/nv50/nv50_stateobj.h
View File

@ -31,7 +31,7 @@ struct nv50_rasterizer_stateobj {
struct nv50_zsa_stateobj {
struct pipe_depth_stencil_alpha_state pipe;
int size;
uint32_t state[34];
uint32_t state[38];
};
struct nv50_constbuf {

3
src/gallium/drivers/nouveau/nvc0/nvc0_program.c
View File

@ -748,7 +748,8 @@ nvc0_program_upload_code(struct nvc0_context *nvc0, struct nvc0_program *prog)
if (prog->fixups) {
nv50_ir_apply_fixups(prog->fixups, prog->code,
prog->fp.force_persample_interp,
prog->fp.flatshade);
prog->fp.flatshade,
0 /* alphatest */);
for (int i = 0; i < 2; i++) {
unsigned mask = prog->fp.color_interp[i] >> 4;
unsigned interp = prog->fp.color_interp[i] & 3;