[Nouveau] [PATCH 4/5] nvc0: refactor TIC uploads to allow different specifies per generation
Samuel Pitoiset
samuel.pitoiset at gmail.com
Mon Oct 17 12:37:11 UTC 2016
On 10/17/2016 02:24 PM, Ilia Mirkin wrote:
> On Mon, Oct 17, 2016 at 5:46 AM, Samuel Pitoiset
> <samuel.pitoiset at gmail.com> wrote:
>> Few comments below.
>>
>> On 10/16/2016 09:14 PM, Ilia Mirkin wrote:
>>>
>>> This flips GM10x to using the updated format, which is what I tested
>>> with. However GM20x and GP10x also use this TIC format.
>>>
>>> Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu>
>>> ---
>>> src/nvc0_accel.c | 11 ++++++++++
>>> src/nvc0_accel.h | 56 ++++++++++++++++++++++++++++++++++++++++++++++
>>> src/nvc0_exa.c | 22 ++++---------------
>>> src/nvc0_xv.c | 67
>>> +++++++++++++++++++-------------------------------------
>>> 4 files changed, 93 insertions(+), 63 deletions(-)
>>>
>>> diff --git a/src/nvc0_accel.c b/src/nvc0_accel.c
>>> index 0682806..8da5051 100644
>>> --- a/src/nvc0_accel.c
>>> +++ b/src/nvc0_accel.c
>>> @@ -322,6 +322,17 @@ NVAccelInit3D_NVC0(ScrnInfoPtr pScrn)
>>> PUSH_DATA (push, (bo->offset + MISC_OFFSET) >> 32);
>>> PUSH_DATA (push, (bo->offset + MISC_OFFSET));
>>> PUSH_DATA (push, 1);
>>> + } else {
>>> + /* Use new TIC format. Not strictly necessary for GM20x+
>>> */
>>
>>
>> Yes, but it's also enabled by default in mesa, looks fine.
>>
>>
>>> + IMMED_NVC0(push, SUBC_3D(0x0f10), 1);
>>> + if (pNv->dev->chipset >= 0x120) {
>>> + /* Use center sample locations. */
>>> + BEGIN_NVC0(push, SUBC_3D(0x11e0), 4);
>>> + PUSH_DATA (push, 0x88888888);
>>> + PUSH_DATA (push, 0x88888888);
>>> + PUSH_DATA (push, 0x88888888);
>>> + PUSH_DATA (push, 0x88888888);
>>> + }
>>> }
>>>
>>> BEGIN_NVC0(push, NVC0_3D(CODE_ADDRESS_HIGH), 2);
>>> diff --git a/src/nvc0_accel.h b/src/nvc0_accel.h
>>> index 607e97b..9378236 100644
>>> --- a/src/nvc0_accel.h
>>> +++ b/src/nvc0_accel.h
>>> @@ -7,6 +7,7 @@
>>> #include "hwdefs/nvc0_m2mf.xml.h"
>>> #include "hwdefs/nv50_defs.xml.h"
>>> #include "hwdefs/nv50_texture.h"
>>> +#include "hwdefs/gm107_texture.xml.h"
>>> #include "hwdefs/nv_3ddefs.xml.h"
>>>
>>> /* subchannel assignments, compatible with kepler's fixed layout */
>>> @@ -108,4 +109,59 @@ PUSH_DATAu(struct nouveau_pushbuf *push, struct
>>> nouveau_bo *bo,
>>> }
>>> }
>>>
>>> +static __inline__ void
>>> +PUSH_TIC(struct nouveau_pushbuf *push, struct nouveau_bo *bo, unsigned
>>> offset,
>>> + unsigned width, unsigned height, unsigned pitch, unsigned format)
>>> +{
>>> + if (push->client->device->chipset < 0x110) {
>>> + unsigned tic2 = 0xd0001000;
>>> + if (pitch == 0)
>>> + tic2 |= 0x00004000;
>>> + else
>>> + tic2 |= 0x0005c000;
>>> + PUSH_DATA(push, format);
>>> + PUSH_DATA(push, bo->offset + offset);
>>> + PUSH_DATA(push, ((bo->offset + offset) >> 32) |
>>> + (bo->config.nvc0.tile_mode << 18) |
>>> + tic2);
>>> + PUSH_DATA(push, 0x00300000);
>>> + PUSH_DATA(push, 0x80000000 | width);
>>> + PUSH_DATA(push, 0x00010000 | height);
>>> + PUSH_DATA (push, 0x03000000);
>>> + PUSH_DATA (push, 0x00000000);
>>
>>
>> Cosmetic.
>
> Oops, will fix.
>
>>
>>
>>> + } else {
>>> + unsigned tile_mode = bo->config.nvc0.tile_mode;
>>> + PUSH_DATA(push, (format & 0x3f) | ((format & ~0x3f) <<
>>> 1));
>>> + PUSH_DATA(push, bo->offset + offset);
>>> + if (pitch == 0) {
>>> + PUSH_DATA(push, ((bo->offset + offset) >> 32) |
>>> +
>>> GM107_TIC2_2_HEADER_VERSION_BLOCKLINEAR);
>>> + PUSH_DATA(push, GM107_TIC2_3_LOD_ANISO_QUALITY_2 |
>>> + (tile_mode & 0x007) |
>>> + (tile_mode & 0x070 >> (4 - 3)) |
>>> + (tile_mode & 0x700 >> (8 - 6)));
>>> + PUSH_DATA(push,
>>> GM107_TIC2_4_SECTOR_PROMOTION_PROMOTE_TO_2_V |
>>> + GM107_TIC2_4_BORDER_SIZE_SAMPLER_COLOR |
>>> + GM107_TIC2_4_TEXTURE_TYPE_TWO_D |
>>> + (width - 1));
>>> + PUSH_DATA(push, GM107_TIC2_5_NORMALIZED_COORDS |
>>> + ((height - 1) & 0xffff));
>>> + PUSH_DATA(push,
>>> GM107_TIC2_6_ANISO_FINE_SPREAD_FUNC_TWO |
>>> +
>>> GM107_TIC2_6_ANISO_COARSE_SPREAD_FUNC_ONE);
>>> + PUSH_DATA(push, 0x00000000);
>>> + } else {
>>> + PUSH_DATA(push, ((bo->offset + offset) >> 32) |
>>> +
>>> GM107_TIC2_2_HEADER_VERSION_PITCH);
>>> + PUSH_DATA(push, GM107_TIC2_3_LOD_ANISO_QUALITY_2 |
>>> + (pitch >> 5));
>>> + PUSH_DATA(push,
>>> GM107_TIC2_4_BORDER_SIZE_SAMPLER_COLOR |
>>> +
>>> GM107_TIC2_4_TEXTURE_TYPE_TWO_D_NO_MIPMAP |
>>> + (width - 1));
>>> + PUSH_DATA(push, GM107_TIC2_5_NORMALIZED_COORDS |
>>> (height - 1));
>>> + PUSH_DATA(push, 0x000000000);
>>> + PUSH_DATA(push, 0x000000000);
>>> + }
>>> + }
>>> +}
>>> +
>>> #endif
>>> diff --git a/src/nvc0_exa.c b/src/nvc0_exa.c
>>> index a53dfe6..c8ace97 100644
>>> --- a/src/nvc0_exa.c
>>> +++ b/src/nvc0_exa.c
>>> @@ -538,14 +538,8 @@ NVC0EXAPictSolid(NVPtr pNv, PicturePtr ppict,
>>> unsigned unit)
>>> PUSH_DATAu(push, pNv->scratch, SOLID(unit), 1);
>>> PUSH_DATA (push, ppict->pSourcePict->solidFill.color);
>>> PUSH_DATAu(push, pNv->scratch, TIC_OFFSET + (unit * 32), 8);
>>> - PUSH_DATA (push, _(B_C0, G_C1, R_C2, A_C3, 8_8_8_8));
>>> - PUSH_DATA (push, offset);
>>> - PUSH_DATA (push, (offset >> 32) | 0xd005d000);
>>> - PUSH_DATA (push, 0x00300000);
>>> - PUSH_DATA (push, 0x00000001);
>>> - PUSH_DATA (push, 0x00010001);
>>> - PUSH_DATA (push, 0x03000000);
>>> - PUSH_DATA (push, 0x00000000);
>>> + PUSH_TIC (push, pNv->scratch, SOLID(unit), 1, 1, 4,
>>> + _(B_C0, G_C1, R_C2, A_C3, 8_8_8_8));
>>> PUSH_DATAu(push, pNv->scratch, TSC_OFFSET + (unit * 32), 8);
>>> PUSH_DATA (push, NV50TSC_1_0_WRAPS_REPEAT |
>>> NV50TSC_1_0_WRAPT_REPEAT |
>>> @@ -651,16 +645,8 @@ NVC0EXAPictTexture(NVPtr pNv, PixmapPtr ppix,
>>> PicturePtr ppict, unsigned unit)
>>>
>>> PUSH_REFN (push, bo, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
>>> PUSH_DATAu(push, pNv->scratch, TIC_OFFSET + (unit * 32), 8);
>>> - PUSH_DATA (push, format);
>>> - PUSH_DATA (push, bo->offset);
>>> - PUSH_DATA (push, (bo->offset >> 32) |
>>> - (bo->config.nvc0.tile_mode << 18) |
>>> - 0xd0005000);
>>> - PUSH_DATA (push, 0x00300000);
>>> - PUSH_DATA (push, (1 << 31) | ppix->drawable.width);
>>> - PUSH_DATA (push, (1 << 16) | ppix->drawable.height);
>>> - PUSH_DATA (push, 0x03000000);
>>> - PUSH_DATA (push, 0x00000000);
>>> + PUSH_TIC (push, bo, 0, ppix->drawable.width,
>>> ppix->drawable.height, 0,
>>> + format);
>>>
>>> PUSH_DATAu(push, pNv->scratch, TSC_OFFSET + (unit * 32), 8);
>>> if (ppict->repeat) {
>>> diff --git a/src/nvc0_xv.c b/src/nvc0_xv.c
>>> index 129c505..c3e58da 100644
>>> --- a/src/nvc0_xv.c
>>> +++ b/src/nvc0_xv.c
>>> @@ -74,7 +74,6 @@ nvc0_xv_image_put(ScrnInfoPtr pScrn,
>>> { dst, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR },
>>> };
>>> struct nouveau_pushbuf *push = pNv->pushbuf;
>>> - uint32_t mode = 0xd0005000 | (src->config.nvc0.tile_mode << 18);
>>> float X1, X2, Y1, Y2;
>>> BoxPtr pbox;
>>> int nbox;
>>> @@ -105,71 +104,49 @@ nvc0_xv_image_put(ScrnInfoPtr pScrn,
>>>
>>> PUSH_DATAu(push, pNv->scratch, TIC_OFFSET, 16);
>>> if (id == FOURCC_YV12 || id == FOURCC_I420) {
>>> - PUSH_DATA (push, NV50TIC_0_0_MAPA_C0 | NV50TIC_0_0_TYPEA_UNORM |
>>> + PUSH_TIC(push, src, packed_y, width, height, 0,
>>> + NV50TIC_0_0_MAPA_C0 | NV50TIC_0_0_TYPEA_UNORM |
>>> NV50TIC_0_0_MAPB_ZERO | NV50TIC_0_0_TYPEB_UNORM |
>>> NV50TIC_0_0_MAPG_ZERO | NV50TIC_0_0_TYPEG_UNORM |
>>> NV50TIC_0_0_MAPR_ZERO | NV50TIC_0_0_TYPER_UNORM |
>>> NV50TIC_0_0_FMT_8);
>>> - PUSH_DATA (push, ((src->offset + packed_y)));
>>> - PUSH_DATA (push, ((src->offset + packed_y) >> 32) | mode);
>>> - PUSH_DATA (push, 0x00300000);
>>> - PUSH_DATA (push, width);
>>> - PUSH_DATA (push, (1 << NV50TIC_0_5_DEPTH_SHIFT) | height);
>>> - PUSH_DATA (push, 0x03000000);
>>> - PUSH_DATA (push, 0x00000000);
>>> - PUSH_DATA (push, NV50TIC_0_0_MAPA_C1 | NV50TIC_0_0_TYPEA_UNORM |
>>> + PUSH_TIC(push, src, uv, width >> 1, height >> 1, 0,
>>
>>
>> At first look, this seems wrong. Why do you shift width/height here?
>
> See the PUSH_DATA's I remove below. It's because with YV12/I420 are
> sub-sampled, so the U/V have 1 pixel's worth of data to every quad of
> Y data. But ultimately, it's dictated by what the old code had.
Okay, I was just a bit surprised.
Makes sense.
>
>>
>>
>>> + NV50TIC_0_0_MAPA_C1 | NV50TIC_0_0_TYPEA_UNORM |
>>> NV50TIC_0_0_MAPB_C0 | NV50TIC_0_0_TYPEB_UNORM |
>>> NV50TIC_0_0_MAPG_ZERO | NV50TIC_0_0_TYPEG_UNORM |
>>> NV50TIC_0_0_MAPR_ZERO | NV50TIC_0_0_TYPER_UNORM |
>>> NV50TIC_0_0_FMT_8_8);
>>> - PUSH_DATA (push, ((src->offset + uv)));
>>> - PUSH_DATA (push, ((src->offset + uv) >> 32) | mode);
>>> - PUSH_DATA (push, 0x00300000);
>>> - PUSH_DATA (push, width >> 1);
>>> - PUSH_DATA (push, (1 << NV50TIC_0_5_DEPTH_SHIFT) | (height >> 1));
>>> - PUSH_DATA (push, 0x03000000);
>>> - PUSH_DATA (push, 0x00000000);
>>> } else {
>>> - if (id == FOURCC_UYVY) {
>>> - PUSH_DATA (push, NV50TIC_0_0_MAPA_C1 | NV50TIC_0_0_TYPEA_UNORM |
>>> + unsigned format;
>>> + if (id == FOURCC_UYVY) {
>>> + format = NV50TIC_0_0_MAPA_C1 | NV50TIC_0_0_TYPEA_UNORM |
>>> NV50TIC_0_0_MAPB_ZERO | NV50TIC_0_0_TYPEB_UNORM |
>>> NV50TIC_0_0_MAPG_ZERO | NV50TIC_0_0_TYPEG_UNORM |
>>> NV50TIC_0_0_MAPR_ZERO | NV50TIC_0_0_TYPER_UNORM |
>>> - NV50TIC_0_0_FMT_8_8);
>>> - } else {
>>> - PUSH_DATA (push, NV50TIC_0_0_MAPA_C0 | NV50TIC_0_0_TYPEA_UNORM |
>>> + NV50TIC_0_0_FMT_8_8;
>>> + } else {
>>> + format = NV50TIC_0_0_MAPA_C0 | NV50TIC_0_0_TYPEA_UNORM |
>>> NV50TIC_0_0_MAPB_ZERO | NV50TIC_0_0_TYPEB_UNORM |
>>> NV50TIC_0_0_MAPG_ZERO | NV50TIC_0_0_TYPEG_UNORM |
>>> NV50TIC_0_0_MAPR_ZERO | NV50TIC_0_0_TYPER_UNORM |
>>> - NV50TIC_0_0_FMT_8_8);
>>> - }
>>> - PUSH_DATA (push, ((src->offset + packed_y)));
>>> - PUSH_DATA (push, ((src->offset + packed_y) >> 32) | mode);
>>> - PUSH_DATA (push, 0x00300000);
>>> - PUSH_DATA (push, width);
>>> - PUSH_DATA (push, (1 << NV50TIC_0_5_DEPTH_SHIFT) | height);
>>> - PUSH_DATA (push, 0x03000000);
>>> - PUSH_DATA (push, 0x00000000);
>>> - if (id == FOURCC_UYVY) {
>>> - PUSH_DATA (push, NV50TIC_0_0_MAPA_C2 | NV50TIC_0_0_TYPEA_UNORM |
>>> + NV50TIC_0_0_FMT_8_8;
>>> + }
>>> + PUSH_TIC(push, src, packed_y, width, height, 0, format);
>>> +
>>> + if (id == FOURCC_UYVY) {
>>> + format = NV50TIC_0_0_MAPA_C2 | NV50TIC_0_0_TYPEA_UNORM |
>>> NV50TIC_0_0_MAPB_C0 | NV50TIC_0_0_TYPEB_UNORM |
>>> NV50TIC_0_0_MAPG_ZERO | NV50TIC_0_0_TYPEG_UNORM |
>>> NV50TIC_0_0_MAPR_ZERO | NV50TIC_0_0_TYPER_UNORM |
>>> - NV50TIC_0_0_FMT_8_8_8_8);
>>> - } else {
>>> - PUSH_DATA (push, NV50TIC_0_0_MAPA_C3 | NV50TIC_0_0_TYPEA_UNORM |
>>> + NV50TIC_0_0_FMT_8_8_8_8;
>>> + } else {
>>> + format = NV50TIC_0_0_MAPA_C3 | NV50TIC_0_0_TYPEA_UNORM |
>>> NV50TIC_0_0_MAPB_C1 | NV50TIC_0_0_TYPEB_UNORM |
>>> NV50TIC_0_0_MAPG_ZERO | NV50TIC_0_0_TYPEG_UNORM |
>>> NV50TIC_0_0_MAPR_ZERO | NV50TIC_0_0_TYPER_UNORM |
>>> - NV50TIC_0_0_FMT_8_8_8_8);
>>> - }
>>> - PUSH_DATA (push, ((src->offset + packed_y)));
>>> - PUSH_DATA (push, ((src->offset + packed_y) >> 32) | mode);
>>> - PUSH_DATA (push, 0x00300000);
>>> - PUSH_DATA (push, (width >> 1));
>>> - PUSH_DATA (push, (1 << NV50TIC_0_5_DEPTH_SHIFT) | height);
>>> - PUSH_DATA (push, 0x03000000);
>>> - PUSH_DATA (push, 0x00000000);
>>> + NV50TIC_0_0_FMT_8_8_8_8;
>>> + }
>>> + PUSH_TIC(push, src, packed_y, width >> 1, height, 0,
>>> format);
>>> }
>>>
>>> PUSH_DATAu(push, pNv->scratch, TSC_OFFSET, 16);
>>>
>>
>> --
>> -Samuel
--
-Samuel
More information about the Nouveau
mailing list