[Mesa-dev] [PATCH 3/7] radeonsi: fully enable 2x DCC MSAA for array and non-array textures
Marek Olšák
maraeo at gmail.com
Tue Apr 17 00:41:30 UTC 2018
From: Marek Olšák <marek.olsak at amd.com>
The clear code is exactly the same as for 1 sample buffers -
just clear the whole thing.
---
src/gallium/drivers/radeonsi/si_clear.c | 13 +++++++------
src/gallium/drivers/radeonsi/si_pipe.c | 5 +----
src/gallium/drivers/radeonsi/si_pipe.h | 1 -
src/gallium/drivers/radeonsi/si_texture.c | 15 ++++++++++++---
4 files changed, 20 insertions(+), 14 deletions(-)
diff --git a/src/gallium/drivers/radeonsi/si_clear.c b/src/gallium/drivers/radeonsi/si_clear.c
index b08a9558b4d..f2df9483c12 100644
--- a/src/gallium/drivers/radeonsi/si_clear.c
+++ b/src/gallium/drivers/radeonsi/si_clear.c
@@ -204,33 +204,34 @@ void vi_dcc_clear_level(struct si_context *sctx,
dcc_buffer = &rtex->dcc_separate_buffer->b.b;
dcc_offset = 0;
} else {
dcc_buffer = &rtex->resource.b.b;
dcc_offset = rtex->dcc_offset;
}
if (sctx->chip_class >= GFX9) {
/* Mipmap level clears aren't implemented. */
assert(rtex->resource.b.b.last_level == 0);
- /* MSAA needs a different clear size. */
- assert(rtex->resource.b.b.nr_samples <= 1);
+ /* 4x and 8x MSAA needs a sophisticated compute shader for
+ * the clear. See AMDVLK. */
+ assert(rtex->resource.b.b.nr_samples <= 2);
clear_size = rtex->surface.dcc_size;
} else {
unsigned num_layers = util_num_layers(&rtex->resource.b.b, level);
/* If this is 0, fast clear isn't possible. (can occur with MSAA) */
assert(rtex->surface.u.legacy.level[level].dcc_fast_clear_size);
- /* Layered MSAA DCC fast clears need to clear dcc_fast_clear_size
- * bytes for each layer. This is not currently implemented, and
- * therefore MSAA DCC isn't even enabled with multiple layers.
+ /* Layered 4x and 8x MSAA DCC fast clears need to clear
+ * dcc_fast_clear_size bytes for each layer. A compute shader
+ * would be more efficient than separate per-layer clear operations.
*/
- assert(rtex->resource.b.b.nr_samples <= 1 || num_layers == 1);
+ assert(rtex->resource.b.b.nr_samples <= 2 || num_layers == 1);
dcc_offset += rtex->surface.u.legacy.level[level].dcc_offset;
clear_size = rtex->surface.u.legacy.level[level].dcc_fast_clear_size *
num_layers;
}
si_clear_buffer(sctx, dcc_buffer, dcc_offset, clear_size,
clear_value, SI_COHERENCY_CB_META);
}
diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c
index 466546b76b7..a10f1c9f794 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.c
+++ b/src/gallium/drivers/radeonsi/si_pipe.c
@@ -83,21 +83,20 @@ static const struct debug_named_value debug_options[] = {
{ "dpbb", DBG(DPBB), "Enable DPBB." },
{ "dfsm", DBG(DFSM), "Enable DFSM." },
{ "nohyperz", DBG(NO_HYPERZ), "Disable Hyper-Z" },
{ "norbplus", DBG(NO_RB_PLUS), "Disable RB+." },
{ "no2d", DBG(NO_2D_TILING), "Disable 2D tiling" },
{ "notiling", DBG(NO_TILING), "Disable tiling" },
{ "nodcc", DBG(NO_DCC), "Disable DCC." },
{ "nodccclear", DBG(NO_DCC_CLEAR), "Disable DCC fast clear." },
{ "nodccfb", DBG(NO_DCC_FB), "Disable separate DCC on the main framebuffer" },
{ "nodccmsaa", DBG(NO_DCC_MSAA), "Disable DCC for MSAA" },
- { "dccmsaa", DBG(DCC_MSAA), "Enable DCC for MSAA" },
{ "nofmask", DBG(NO_FMASK), "Disable MSAA compression" },
/* Tests: */
{ "testdma", DBG(TEST_DMA), "Invoke SDMA tests and exit." },
{ "testvmfaultcp", DBG(TEST_VMFAULT_CP), "Invoke a CP VM fault test and exit." },
{ "testvmfaultsdma", DBG(TEST_VMFAULT_SDMA), "Invoke a SDMA VM fault test and exit." },
{ "testvmfaultshader", DBG(TEST_VMFAULT_SHADER), "Invoke a shader VM fault test and exit." },
DEBUG_NAMED_VALUE_END /* must be last */
};
@@ -1004,23 +1003,21 @@ struct pipe_screen *radeonsi_screen_create(struct radeon_winsys *ws,
sscreen->has_rbplus = true;
sscreen->rbplus_allowed =
!(sscreen->debug_flags & DBG(NO_RB_PLUS)) &&
(sscreen->info.family == CHIP_STONEY ||
sscreen->info.family == CHIP_VEGA12 ||
sscreen->info.family == CHIP_RAVEN);
}
sscreen->dcc_msaa_allowed =
- !(sscreen->debug_flags & DBG(NO_DCC_MSAA)) &&
- (sscreen->debug_flags & DBG(DCC_MSAA) ||
- sscreen->info.chip_class == VI);
+ !(sscreen->debug_flags & DBG(NO_DCC_MSAA));
sscreen->cpdma_prefetch_writes_memory = sscreen->info.chip_class <= VI;
(void) mtx_init(&sscreen->shader_parts_mutex, mtx_plain);
sscreen->use_monolithic_shaders =
(sscreen->debug_flags & DBG(MONOLITHIC_SHADERS)) != 0;
sscreen->barrier_flags.cp_to_L2 = SI_CONTEXT_INV_SMEM_L1 |
SI_CONTEXT_INV_VMEM_L1;
if (sscreen->info.chip_class <= VI) {
diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index 351c9f4cd38..f73e0d1aac3 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -147,21 +147,20 @@ enum {
DBG_DPBB,
DBG_DFSM,
DBG_NO_HYPERZ,
DBG_NO_RB_PLUS,
DBG_NO_2D_TILING,
DBG_NO_TILING,
DBG_NO_DCC,
DBG_NO_DCC_CLEAR,
DBG_NO_DCC_FB,
DBG_NO_DCC_MSAA,
- DBG_DCC_MSAA,
DBG_NO_FMASK,
/* Tests: */
DBG_TEST_DMA,
DBG_TEST_VMFAULT_CP,
DBG_TEST_VMFAULT_SDMA,
DBG_TEST_VMFAULT_SHADER,
};
#define DBG_ALL_SHADERS (((1 << (DBG_CS + 1)) - 1))
diff --git a/src/gallium/drivers/radeonsi/si_texture.c b/src/gallium/drivers/radeonsi/si_texture.c
index b41a0d1b925..675e78766ad 100644
--- a/src/gallium/drivers/radeonsi/si_texture.c
+++ b/src/gallium/drivers/radeonsi/si_texture.c
@@ -261,23 +261,32 @@ static int si_init_surface(struct si_screen *sscreen,
flags |= RADEON_SURF_TC_COMPATIBLE_HTILE;
}
if (is_stencil)
flags |= RADEON_SURF_SBUFFER;
}
if (sscreen->info.chip_class >= VI &&
(ptex->flags & SI_RESOURCE_FLAG_DISABLE_DCC ||
ptex->format == PIPE_FORMAT_R9G9B9E5_FLOAT ||
- /* DCC MSAA array textures are disallowed due to incomplete clear impl. */
- (ptex->nr_samples >= 2 &&
- (!sscreen->dcc_msaa_allowed || ptex->array_size > 1))))
+ (ptex->nr_samples >= 2 && !sscreen->dcc_msaa_allowed)))
+ flags |= RADEON_SURF_DISABLE_DCC;
+
+ /* VI: DCC clear for 4x and 8x MSAA array textures unimplemented. */
+ if (sscreen->info.chip_class == VI &&
+ ptex->nr_samples >= 4 &&
+ ptex->array_size > 1)
+ flags |= RADEON_SURF_DISABLE_DCC;
+
+ /* GFX9: DCC clear for 4x and 8x MSAA textures unimplemented. */
+ if (sscreen->info.chip_class >= GFX9 &&
+ ptex->nr_samples >= 4)
flags |= RADEON_SURF_DISABLE_DCC;
if (ptex->bind & PIPE_BIND_SCANOUT || is_scanout) {
/* This should catch bugs in gallium users setting incorrect flags. */
assert(ptex->nr_samples <= 1 &&
ptex->array_size == 1 &&
ptex->depth0 == 1 &&
ptex->last_level == 0 &&
!(flags & RADEON_SURF_Z_OR_SBUFFER));
--
2.17.0
More information about the mesa-dev
mailing list