On Feb 27, 2016 9:02 AM, "Samuel Pitoiset" <samuel.pitoi...@gmail.com
<mailto:samuel.pitoi...@gmail.com>> wrote:
To not overwrite buffers and surfaces on Kepler, we need a different
offset. This will be currently only used for compute because we have
to bind both surfaces and buffers.
Signed-off-by: Samuel Pitoiset <samuel.pitoi...@gmail.com
<mailto:samuel.pitoi...@gmail.com>>
---
.../drivers/nouveau/codegen/nv50_ir_driver.h | 1 +
.../nouveau/codegen/nv50_ir_lowering_nvc0.cpp | 80
++++++++++++++++------
.../nouveau/codegen/nv50_ir_lowering_nvc0.h | 12 +++-
src/gallium/drivers/nouveau/nvc0/nvc0_program.c | 8 ++-
4 files changed, 73 insertions(+), 28 deletions(-)
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_driver.h
b/src/gallium/drivers/nouveau/codegen/nv50_ir_driver.h
index 4504240..479e426 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_driver.h
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_driver.h
@@ -179,6 +179,7 @@ struct nv50_ir_prog_info
uint8_t resInfoCBSlot; /* cX[] used for tex handles,
surface info */
uint16_t texBindBase; /* base address for tex handles
(nve4) */
uint16_t suInfoBase; /* base address for surface info
(nve4) */
+ uint16_t bufInfoBase; /* base address for buffer info */
uint16_t sampleInfoBase; /* base address for sample
positions */
uint8_t msInfoCBSlot; /* cX[] used for multisample info */
uint16_t msInfoBase; /* base address for multisample
info */
diff --git
a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
index d181f15..8abdd93 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
@@ -1026,7 +1026,7 @@ bool
NVC0LoweringPass::handleSUQ(Instruction *suq)
{
suq->op = OP_MOV;
- suq->setSrc(0, loadResLength32(suq->getIndirect(0, 1),
+ suq->setSrc(0, loadBufLength32(suq->getIndirect(0, 1),
suq->getSrc(0)->reg.fileIndex *
16));
suq->setIndirect(0, 0, NULL);
suq->setIndirect(0, 1, NULL);
@@ -1142,7 +1142,7 @@ NVC0LoweringPass::handleATOM(Instruction *atom)
return true;
default:
assert(atom->src(0).getFile() == FILE_MEMORY_GLOBAL);
- base = loadResInfo64(ind, atom->getSrc(0)->reg.fileIndex * 16);
+ base = loadBufInfo64(ind, atom->getSrc(0)->reg.fileIndex * 16);
assert(base->reg.size == 8);
if (ptr)
base = bld.mkOp2v(OP_ADD, TYPE_U64, base, base, ptr);
@@ -1202,19 +1202,19 @@ NVC0LoweringPass::handleCasExch(Instruction
*cas, bool needCctl)
}
inline Value *
-NVC0LoweringPass::loadResInfo32(Value *ptr, uint32_t off)
+NVC0LoweringPass::loadResInfo32(Value *ptr, uint32_t off, uint16_t
base)
{
uint8_t b = prog->driver->io.resInfoCBSlot;
- off += prog->driver->io.suInfoBase;
+ off += base;
return bld.
mkLoadv(TYPE_U32, bld.mkSymbol(FILE_MEMORY_CONST, b,
TYPE_U32, off), ptr);
}
inline Value *
-NVC0LoweringPass::loadResInfo64(Value *ptr, uint32_t off)
+NVC0LoweringPass::loadResInfo64(Value *ptr, uint32_t off, uint16_t
base)
{
uint8_t b = prog->driver->io.resInfoCBSlot;
- off += prog->driver->io.suInfoBase;
+ off += base;
if (ptr)
ptr = bld.mkOp2v(OP_SHL, TYPE_U32, bld.getScratch(), ptr,
bld.mkImm(4));
@@ -1224,10 +1224,10 @@ NVC0LoweringPass::loadResInfo64(Value *ptr,
uint32_t off)
}
inline Value *
-NVC0LoweringPass::loadResLength32(Value *ptr, uint32_t off)
+NVC0LoweringPass::loadResLength32(Value *ptr, uint32_t off,
uint16_t base)
{
uint8_t b = prog->driver->io.resInfoCBSlot;
- off += prog->driver->io.suInfoBase;
+ off += base;
if (ptr)
ptr = bld.mkOp2v(OP_SHL, TYPE_U32, bld.getScratch(), ptr,
bld.mkImm(4));
@@ -1237,6 +1237,42 @@ NVC0LoweringPass::loadResLength32(Value *ptr,
uint32_t off)
}
inline Value *
+NVC0LoweringPass::loadSufInfo32(Value *ptr, uint32_t off)
+{
+ return loadResInfo32(ptr, off, prog->driver->io.suInfoBase);
+}
+
+inline Value *
+NVC0LoweringPass::loadSufInfo64(Value *ptr, uint32_t off)
+{
+ return loadResInfo64(ptr, off, prog->driver->io.suInfoBase);
+}
+
+inline Value *
+NVC0LoweringPass::loadSufLength32(Value *ptr, uint32_t off)
+{
+ return loadResLength32(ptr, off, prog->driver->io.suInfoBase);
+}
+
+inline Value *
+NVC0LoweringPass::loadBufInfo32(Value *ptr, uint32_t off)
+{
+ return loadResInfo32(ptr, off, prog->driver->io.bufInfoBase);
+}
+
+inline Value *
+NVC0LoweringPass::loadBufInfo64(Value *ptr, uint32_t off)
+{
+ return loadResInfo64(ptr, off, prog->driver->io.bufInfoBase);
+}
+
+inline Value *
+NVC0LoweringPass::loadBufLength32(Value *ptr, uint32_t off)
+{
+ return loadResLength32(ptr, off, prog->driver->io.bufInfoBase);
+}
+
+inline Value *
NVC0LoweringPass::loadMsInfo32(Value *ptr, uint32_t off)
{
uint8_t b = prog->driver->io.msInfoCBSlot;
@@ -1316,8 +1352,8 @@
NVC0LoweringPass::adjustCoordinatesMS(TexInstruction *tex)
Value *tx = bld.getSSA(), *ty = bld.getSSA(), *ts = bld.getSSA();
- Value *ms_x = loadResInfo32(NULL, base + NVE4_SU_INFO_MS(0));
- Value *ms_y = loadResInfo32(NULL, base + NVE4_SU_INFO_MS(1));
+ Value *ms_x = loadSufInfo32(NULL, base + NVE4_SU_INFO_MS(0));
+ Value *ms_y = loadSufInfo32(NULL, base + NVE4_SU_INFO_MS(1));
bld.mkOp2(OP_SHL, TYPE_U32, tx, x, ms_x);
bld.mkOp2(OP_SHL, TYPE_U32, ty, y, ms_y);
@@ -1370,9 +1406,9 @@
NVC0LoweringPass::processSurfaceCoordsNVE4(TexInstruction *su)
for (c = 0; c < arg; ++c) {
src[c] = bld.getScratch();
if (c == 0 && raw)
- v = loadResInfo32(NULL, base + NVE4_SU_INFO_RAW_X);
+ v = loadSufInfo32(NULL, base + NVE4_SU_INFO_RAW_X);
else
- v = loadResInfo32(NULL, base + NVE4_SU_INFO_DIM(c));
+ v = loadSufInfo32(NULL, base + NVE4_SU_INFO_DIM(c));
bld.mkOp3(OP_SUCLAMP, TYPE_S32, src[c], su->getSrc(c), v, zero)
->subOp = getSuClampSubOp(su, c);
}
@@ -1394,16 +1430,16 @@
NVC0LoweringPass::processSurfaceCoordsNVE4(TexInstruction *su)
bld.mkOp2(OP_AND, TYPE_U32, off, src[0],
bld.loadImm(NULL, 0xffff));
} else
if (dim == 3) {
- v = loadResInfo32(NULL, base + NVE4_SU_INFO_UNK1C);
+ v = loadSufInfo32(NULL, base + NVE4_SU_INFO_UNK1C);
bld.mkOp3(OP_MADSP, TYPE_U32, off, src[2], v, src[1])
->subOp = NV50_IR_SUBOP_MADSP(4,2,8); // u16l u16l u16l
- v = loadResInfo32(NULL, base + NVE4_SU_INFO_PITCH);
+ v = loadSufInfo32(NULL, base + NVE4_SU_INFO_PITCH);
bld.mkOp3(OP_MADSP, TYPE_U32, off, off, v, src[0])
->subOp = NV50_IR_SUBOP_MADSP(0,2,8); // u32 u16l u16l
} else {
assert(dim == 2);
- v = loadResInfo32(NULL, base + NVE4_SU_INFO_PITCH);
+ v = loadSufInfo32(NULL, base + NVE4_SU_INFO_PITCH);
bld.mkOp3(OP_MADSP, TYPE_U32, off, src[1], v, src[0])
->subOp = su->tex.target.isArray() ?
NV50_IR_SUBOP_MADSP_SD : NV50_IR_SUBOP_MADSP(4,2,8); //
u16l u16l u16l
@@ -1414,7 +1450,7 @@
NVC0LoweringPass::processSurfaceCoordsNVE4(TexInstruction *su)
if (raw) {
bf = src[0];
} else {
- v = loadResInfo32(NULL, base + NVE4_SU_INFO_FMT);
+ v = loadSufInfo32(NULL, base + NVE4_SU_INFO_FMT);
bld.mkOp3(OP_VSHL, TYPE_U32, bf, src[0], v, zero)
->subOp = NV50_IR_SUBOP_V1(7,6,8|2);
}
@@ -1431,7 +1467,7 @@
NVC0LoweringPass::processSurfaceCoordsNVE4(TexInstruction *su)
case 2:
z = off;
if (!su->tex.target.isArray()) {
- z = loadResInfo32(NULL, base + NVE4_SU_INFO_UNK1C);
+ z = loadSufInfo32(NULL, base + NVE4_SU_INFO_UNK1C);
subOp = NV50_IR_SUBOP_SUBFM_3D;
}
break;
@@ -1446,7 +1482,7 @@
NVC0LoweringPass::processSurfaceCoordsNVE4(TexInstruction *su)
}
// part 2
- v = loadResInfo32(NULL, base + NVE4_SU_INFO_ADDR);
+ v = loadSufInfo32(NULL, base + NVE4_SU_INFO_ADDR);
if (su->tex.target == TEX_TARGET_BUFFER) {
eau = v;
@@ -1455,7 +1491,7 @@
NVC0LoweringPass::processSurfaceCoordsNVE4(TexInstruction *su)
}
// add array layer offset
if (su->tex.target.isArray()) {
- v = loadResInfo32(NULL, base + NVE4_SU_INFO_ARRAY);
+ v = loadSufInfo32(NULL, base + NVE4_SU_INFO_ARRAY);
if (dim == 1)
bld.mkOp3(OP_MADSP, TYPE_U32, eau, src[1], v, eau)
->subOp = NV50_IR_SUBOP_MADSP(4,0,0); // u16 u24 u32
@@ -1495,7 +1531,7 @@
NVC0LoweringPass::processSurfaceCoordsNVE4(TexInstruction *su)
// let's just set it 0 for raw access and hope it works
v = raw ?
- bld.mkImm(0) : loadResInfo32(NULL, base + NVE4_SU_INFO_FMT);
+ bld.mkImm(0) : loadSufInfo32(NULL, base + NVE4_SU_INFO_FMT);
// get rid of old coordinate sources, make space for fmt info
and predicate
su->moveSources(arg, 3 - arg);
@@ -1966,12 +2002,12 @@ NVC0LoweringPass::visit(Instruction *i)
i->op = OP_VFETCH;
} else if (i->src(0).getFile() == FILE_MEMORY_GLOBAL) {
Value *ind = i->getIndirect(0, 1);
- Value *ptr = loadResInfo64(ind,
i->getSrc(0)->reg.fileIndex * 16);
+ Value *ptr = loadBufInfo64(ind,
i->getSrc(0)->reg.fileIndex * 16);
// XXX come up with a way not to do this for EVERY little
access but
// rather to batch these up somehow. Unfortunately we've
lost the
// information about the field width by the time we get here.
Value *offset = bld.loadImm(NULL,
i->getSrc(0)->reg.data.offset + typeSizeof(i->sType));
- Value *length = loadResLength32(ind,
i->getSrc(0)->reg.fileIndex * 16);
+ Value *length = loadBufLength32(ind,
i->getSrc(0)->reg.fileIndex * 16);
Value *pred = new_LValue(func, FILE_PREDICATE);
if (i->src(0).isIndirect(0)) {
bld.mkOp2(OP_ADD, TYPE_U64, ptr, ptr,
i->getIndirect(0, 0));
diff --git
a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.h
b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.h
index 6eb8aff..3872f52 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.h
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.h
@@ -117,9 +117,15 @@ private:
void readTessCoord(LValue *dst, int c);
- Value *loadResInfo32(Value *ptr, uint32_t off);
- Value *loadResInfo64(Value *ptr, uint32_t off);
- Value *loadResLength32(Value *ptr, uint32_t off);
+ Value *loadResInfo32(Value *ptr, uint32_t off, uint16_t base);
+ Value *loadResInfo64(Value *ptr, uint32_t off, uint16_t base);
+ Value *loadResLength32(Value *ptr, uint32_t off, uint16_t base);
+ Value *loadSufInfo32(Value *ptr, uint32_t off);
+ Value *loadSufInfo64(Value *ptr, uint32_t off);
+ Value *loadSufLength32(Value *ptr, uint32_t off);
+ Value *loadBufInfo32(Value *ptr, uint32_t off);
+ Value *loadBufInfo64(Value *ptr, uint32_t off);
+ Value *loadBufLength32(Value *ptr, uint32_t off);
Value *loadMsInfo32(Value *ptr, uint32_t off);
Value *loadTexHandle(Value *ptr, unsigned int slot);
diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_program.c
b/src/gallium/drivers/nouveau/nvc0/nvc0_program.c
index 89a7f5c..d01de73 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_program.c
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_program.c
@@ -544,22 +544,24 @@ nvc0_program_translate(struct nvc0_program
*prog, uint16_t chipset,
info->io.texBindBase = NVE4_CP_INPUT_TEX(0);
info->io.suInfoBase = NVE4_CP_INPUT_SUF(0);
info->prop.cp.gridInfoBase = NVE4_CP_INPUT_GRID_INFO(0);
+ info->io.bufInfoBase = 0; /* TODO */
} else {
info->io.resInfoCBSlot = 15;
- info->io.suInfoBase = NVC0_CB_AUX_BUF_INFO(0);
+ info->io.bufInfoBase = NVC0_CB_AUX_BUF_INFO(0);
+ info->io.suInfoBase = 0; /* TODO */
}
info->io.msInfoCBSlot = 0;
info->io.msInfoBase = NVE4_CP_INPUT_MS_OFFSETS;
} else {
if (chipset >= NVISA_GK104_CHIPSET) {
info->io.texBindBase = NVC0_CB_AUX_TEX_INFO(0);
- info->io.suInfoBase = 0; /* TODO */
}
info->io.resInfoCBSlot = 15;
info->io.sampleInfoBase = NVC0_CB_AUX_SAMPLE_INFO;
- info->io.suInfoBase = NVC0_CB_AUX_BUF_INFO(0);
+ info->io.bufInfoBase = NVC0_CB_AUX_BUF_INFO(0);
info->io.msInfoCBSlot = 15;
info->io.msInfoBase = 0; /* TODO */
+ info->io.suInfoBase = 0; /* TODO */
}
info->assignSlots = nvc0_program_assign_varying_slots;
--
2.7.1
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org <mailto:mesa-dev@lists.freedesktop.org>
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
