drm/radeon/kms: optimize CS state checking for r100->r500
The colorbuffer, zbuffer, and texture states are checked only once when
they get changed. This improves performance in the apps which emit
lots of draw packets and few state changes.
This drops performance in glxgears by a 1% or so, but glxgears is not
a benchmark we care about.
The time spent in the kernel when running Torcs dropped from 33% to 23%
and the frame rate is higher, which is a good thing.
r600 might need something like this as well.
Signed-off-by: Marek Olšák <maraeo@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
diff --git a/drivers/gpu/drm/radeon/r100.c b/drivers/gpu/drm/radeon/r100.c
index 5f15820..fdf4bc6 100644
--- a/drivers/gpu/drm/radeon/r100.c
+++ b/drivers/gpu/drm/radeon/r100.c
@@ -1427,6 +1427,7 @@
}
track->zb.robj = reloc->robj;
track->zb.offset = idx_value;
+ track->zb_dirty = true;
ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
break;
case RADEON_RB3D_COLOROFFSET:
@@ -1439,6 +1440,7 @@
}
track->cb[0].robj = reloc->robj;
track->cb[0].offset = idx_value;
+ track->cb_dirty = true;
ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
break;
case RADEON_PP_TXOFFSET_0:
@@ -1454,6 +1456,7 @@
}
ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
track->textures[i].robj = reloc->robj;
+ track->tex_dirty = true;
break;
case RADEON_PP_CUBIC_OFFSET_T0_0:
case RADEON_PP_CUBIC_OFFSET_T0_1:
@@ -1471,6 +1474,7 @@
track->textures[0].cube_info[i].offset = idx_value;
ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
track->textures[0].cube_info[i].robj = reloc->robj;
+ track->tex_dirty = true;
break;
case RADEON_PP_CUBIC_OFFSET_T1_0:
case RADEON_PP_CUBIC_OFFSET_T1_1:
@@ -1488,6 +1492,7 @@
track->textures[1].cube_info[i].offset = idx_value;
ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
track->textures[1].cube_info[i].robj = reloc->robj;
+ track->tex_dirty = true;
break;
case RADEON_PP_CUBIC_OFFSET_T2_0:
case RADEON_PP_CUBIC_OFFSET_T2_1:
@@ -1505,9 +1510,12 @@
track->textures[2].cube_info[i].offset = idx_value;
ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
track->textures[2].cube_info[i].robj = reloc->robj;
+ track->tex_dirty = true;
break;
case RADEON_RE_WIDTH_HEIGHT:
track->maxy = ((idx_value >> 16) & 0x7FF);
+ track->cb_dirty = true;
+ track->zb_dirty = true;
break;
case RADEON_RB3D_COLORPITCH:
r = r100_cs_packet_next_reloc(p, &reloc);
@@ -1528,9 +1536,11 @@
ib[idx] = tmp;
track->cb[0].pitch = idx_value & RADEON_COLORPITCH_MASK;
+ track->cb_dirty = true;
break;
case RADEON_RB3D_DEPTHPITCH:
track->zb.pitch = idx_value & RADEON_DEPTHPITCH_MASK;
+ track->zb_dirty = true;
break;
case RADEON_RB3D_CNTL:
switch ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f) {
@@ -1555,6 +1565,8 @@
return -EINVAL;
}
track->z_enabled = !!(idx_value & RADEON_Z_ENABLE);
+ track->cb_dirty = true;
+ track->zb_dirty = true;
break;
case RADEON_RB3D_ZSTENCILCNTL:
switch (idx_value & 0xf) {
@@ -1572,6 +1584,7 @@
default:
break;
}
+ track->zb_dirty = true;
break;
case RADEON_RB3D_ZPASS_ADDR:
r = r100_cs_packet_next_reloc(p, &reloc);
@@ -1588,6 +1601,7 @@
uint32_t temp = idx_value >> 4;
for (i = 0; i < track->num_texture; i++)
track->textures[i].enabled = !!(temp & (1 << i));
+ track->tex_dirty = true;
}
break;
case RADEON_SE_VF_CNTL:
@@ -1602,12 +1616,14 @@
i = (reg - RADEON_PP_TEX_SIZE_0) / 8;
track->textures[i].width = (idx_value & RADEON_TEX_USIZE_MASK) + 1;
track->textures[i].height = ((idx_value & RADEON_TEX_VSIZE_MASK) >> RADEON_TEX_VSIZE_SHIFT) + 1;
+ track->tex_dirty = true;
break;
case RADEON_PP_TEX_PITCH_0:
case RADEON_PP_TEX_PITCH_1:
case RADEON_PP_TEX_PITCH_2:
i = (reg - RADEON_PP_TEX_PITCH_0) / 8;
track->textures[i].pitch = idx_value + 32;
+ track->tex_dirty = true;
break;
case RADEON_PP_TXFILTER_0:
case RADEON_PP_TXFILTER_1:
@@ -1621,6 +1637,7 @@
tmp = (idx_value >> 27) & 0x7;
if (tmp == 2 || tmp == 6)
track->textures[i].roundup_h = false;
+ track->tex_dirty = true;
break;
case RADEON_PP_TXFORMAT_0:
case RADEON_PP_TXFORMAT_1:
@@ -1673,6 +1690,7 @@
}
track->textures[i].cube_info[4].width = 1 << ((idx_value >> 16) & 0xf);
track->textures[i].cube_info[4].height = 1 << ((idx_value >> 20) & 0xf);
+ track->tex_dirty = true;
break;
case RADEON_PP_CUBIC_FACES_0:
case RADEON_PP_CUBIC_FACES_1:
@@ -1683,6 +1701,7 @@
track->textures[i].cube_info[face].width = 1 << ((tmp >> (face * 8)) & 0xf);
track->textures[i].cube_info[face].height = 1 << ((tmp >> ((face * 8) + 4)) & 0xf);
}
+ track->tex_dirty = true;
break;
default:
printk(KERN_ERR "Forbidden register 0x%04X in cs at %d\n",
@@ -3318,9 +3337,9 @@
unsigned long size;
unsigned prim_walk;
unsigned nverts;
- unsigned num_cb = track->num_cb;
+ unsigned num_cb = track->cb_dirty ? track->num_cb : 0;
- if (!track->zb_cb_clear && !track->color_channel_mask &&
+ if (num_cb && !track->zb_cb_clear && !track->color_channel_mask &&
!track->blend_read_enable)
num_cb = 0;
@@ -3341,7 +3360,9 @@
return -EINVAL;
}
}
- if (track->z_enabled) {
+ track->cb_dirty = false;
+
+ if (track->zb_dirty && track->z_enabled) {
if (track->zb.robj == NULL) {
DRM_ERROR("[drm] No buffer for z buffer !\n");
return -EINVAL;
@@ -3358,6 +3379,8 @@
return -EINVAL;
}
}
+ track->zb_dirty = false;
+
prim_walk = (track->vap_vf_cntl >> 4) & 0x3;
if (track->vap_vf_cntl & (1 << 14)) {
nverts = track->vap_alt_nverts;
@@ -3417,13 +3440,22 @@
prim_walk);
return -EINVAL;
}
- return r100_cs_track_texture_check(rdev, track);
+
+ if (track->tex_dirty) {
+ track->tex_dirty = false;
+ return r100_cs_track_texture_check(rdev, track);
+ }
+ return 0;
}
void r100_cs_track_clear(struct radeon_device *rdev, struct r100_cs_track *track)
{
unsigned i, face;
+ track->cb_dirty = true;
+ track->zb_dirty = true;
+ track->tex_dirty = true;
+
if (rdev->family < CHIP_R300) {
track->num_cb = 1;
if (rdev->family <= CHIP_RS200)