similar to: [PATCH] nv50: don't allocate in the param buffer

Hi ! I've been trying to improve NV50 shader generation a bit the last couple of weeks, so here is what I've produced. I don't know if it's usable for you or just a pile of horrible hacks, but at least it makes some mesa demos render more correcly, p.e. the teapot (aside from mip-mapping issues of the floor texture), arbfplight, and I think the gears also didn't appear as they

--- src/gallium/drivers/nv50/nv50_program.c | 162 +++++++++++++++++++++++++------ src/gallium/drivers/nv50/nv50_program.h | 1 + 2 files changed, 132 insertions(+), 31 deletions(-) diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c index 5594560..16bf2f1 100644 --- a/src/gallium/drivers/nv50/nv50_program.c +++

--- src/gallium/drivers/nv50/nv50_context.h | 1 + src/gallium/drivers/nv50/nv50_program.c | 171 ++++++++++++++++++------ src/gallium/drivers/nv50/nv50_program.h | 1 + src/gallium/drivers/nv50/nv50_state_validate.c | 3 + 4 files changed, 138 insertions(+), 38 deletions(-) diff --git a/src/gallium/drivers/nv50/nv50_context.h

Currently on NV30/NV40 an assert will be triggered once 32 queries are outstanding. This violates the OpenGL/Gallium interface, which requires support for an unlimited number of fences. This patch fixes the problem by putting queries in a linked list and waiting on the oldest one if allocation fails. nVidia seems to use a similar strategy, but with 1024 instead of 32 fences. The next patch will

Make use of tgsi_shader_info to determine how many nv50_regs we need to allocate, if program uses KIL, or writes DEPR. --- src/gallium/drivers/nv50/nv50_program.c | 284 ++++++++++++++----------------- 1 files changed, 129 insertions(+), 155 deletions(-) diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c index e2adeca..c8157f1 100644 ---

Currently on NV30/NV40 an assert will be triggered once 32 queries are outstanding. This violates the OpenGL/Gallium interface, which requires support for an unlimited number of fences. This patch fixes the problem by putting queries in a linked list and waiting on the oldest one if allocation fails. nVidia seems to use a similar strategy, but with 1024 instead of 32 fences. The next patch will

This moves construction of the mapping between VP outputs and FP inputs into validation. The map also contains slots for special outputs like clip distance and point size, so we need to at least merge the VP related and FP related parts on validation if we want to support those. Now we match every single FP input component with results from the VP and leave those not read out of the map, or

--- pxzcat.c | 21 +++++++++++++++++++++ 1 file changed, 21 insertions(+) diff --git a/pxzcat.c b/pxzcat.c index 9bcdc36..55ccfc0 100644 --- a/pxzcat.c +++ b/pxzcat.c @@ -44,10 +44,11 @@ #include <sys/types.h> #include <error.h> #include <errno.h> #include <getopt.h> #include <pthread.h> +#include <linux/falloc.h> #include <lzma.h> #define DEBUG

Hi, I'm trying to find a place where objects held in nv40_context->state.hw[] and nv40_screen->state[] are being unreferenced during pipe_context destruction. Currently I'm observing that these objects are not unreferenced and since they hold reference to buffer objects, the buffer objects themselves are not unreferenced as well (for example color buffer or z buffer). In

- unreference state objects so that buffer objects are unreferenced and eventually destroyed - free channel at screen's destruction Index: nv50/nv50_screen.c =================================================================== --- nv50/nv50_screen.c (wersja 32083) +++ nv50/nv50_screen.c (kopia robocza) @@ -162,7 +162,22 @@ nv50_screen_destroy(struct pipe_screen *pscreen) { struct

This fixes the previous patches, and add some debugging output if NV50_PROGRAM_DUMP is un-commented. Will merge this into the patches, later. --- src/gallium/drivers/nv50/nv50_program.c | 94 +++++++++++++++++++------------ 1 files changed, 58 insertions(+), 36 deletions(-) diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c index 4b05075..caf03c9

The hardware only generates vertexid when vertices come from a VBO. This fixes: vertexid-drawelements vertexid-drawarrays Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu> Cc: "11.0" <mesa-stable at lists.freedesktop.org> --- src/gallium/drivers/nouveau/nv50/nv50_program.c | 1 + src/gallium/drivers/nouveau/nv50/nv50_program.h | 1 +

We need to invert the viewport translate/scale parameters when the state tracker thinks we have Y_0_TOP. If these cases, we have do to bypass mode by setting an identity viewport transform for x, z and inversion for y, or p.e. clear_with_quad won't work correctly. Clipping for xy in NDC space needs to be disabled then. --- src/gallium/drivers/nv50/nv50_context.h | 1 +

Functionally identical but much simpler. Should also better integrate with future layer/viewport changes/fixes. Cc: 10.1 <mesa-stable at lists.freedesktop.org> Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu> --- Not *strictly* necessary in stable, but it will make backporting later fixes easier. No regressions in piglit. src/gallium/drivers/nouveau/nv50/nv50_program.c | 5

Fixes 3 piglit tests: general/pos-array shaders/glsl-novertexdata shaders/glsl-vs-point-size and makes shaders/vp-ignore-input not trigger PGRAPH DATA_ERROR --- It's a bit ugly... If there's a way to fix it properly, I'm open to suggestions. --- src/gallium/drivers/nv50/nv50_program.c | 7 +++++++ 1 files changed, 7 insertions(+), 0 deletions(-) diff --git

--- src/gallium/drivers/nv50/nv50_program.c | 54 ++++++++++++++++++++++++++++++- src/gallium/drivers/nv50/nv50_screen.c | 2 +- src/gallium/drivers/nv50/nv50_state.c | 3 ++ 3 files changed, 57 insertions(+), 2 deletions(-) diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c index cf57d5c..883c28b 100644 ---

The block counts in the APM partitions assumed 512 bytes per block, whereas the start block numbers assume 2048 as announced in the APM header. This change divides the affected block counts by 4 to correct the assumption. --- utils/isohybrid.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) diff --git a/utils/isohybrid.c b/utils/isohybrid.c index c5b4281..7d0864e 100644 ---

This comes in handy when checking "NV50_PROG_DEBUG=1" outputs with diff! V2: - Use environmental variable (Karol Herbst) V3: - Use the already populated nv50_ir_prog_info to forward information to the print pass (Pierre Moreau) Signed-off-by: Tobias Klausmann <tobias.johannes.klausmann at mni.thm.de> --- src/gallium/drivers/nouveau/codegen/nv50_ir_driver.h | 1 +

--- src/gallium/drivers/nv50/nv50_program.c | 19 +++++++++++++++++++ src/gallium/drivers/nv50/nv50_state.c | 4 +++- 2 files changed, 22 insertions(+), 1 deletions(-) diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c index feb1504..39c3afc 100644 --- a/src/gallium/drivers/nv50/nv50_program.c +++ b/src/gallium/drivers/nv50/nv50_program.c @@

http://bugs.freedesktop.org/show_bug.cgi?id=23291 Summary: A slight problem in the NV50 nouveau driver in Mesa Product: Mesa Version: unspecified Platform: Other OS/Version: All Status: NEW Severity: minor Priority: medium Component: Drivers/DRI/nouveau AssignedTo: nouveau at lists.freedesktop.org