@@ -184,6 +184,11 @@ struct nv50_wndw_atom {
} i;
} xlut;
+ struct {
+ u32 matrix[12];
+ bool valid;
+ } csc;
+
struct {
u8 mode:2;
u8 interval:4;
@@ -221,6 +226,7 @@ struct nv50_wndw_atom {
bool ntfy:1;
bool sema:1;
bool xlut:1;
+ bool csc:1;
bool image:1;
bool scale:1;
bool point:1;
@@ -83,6 +83,68 @@ base907c_ilut(struct nv50_wndw *wndw, struct nv50_wndw_atom *asyw)
asyw->xlut.i.load = head907d_olut_load;
}
+static inline u32
+csc_drm_to_base(u64 in)
+{
+ /* base takes a 19-bit 2's complement value in S3.16 format */
+ bool sign = in & BIT_ULL(63);
+ u32 integer = (in >> 32) & 0x7fffffff;
+ u32 fraction = in & 0xffffffff;
+
+ if (integer >= 4) {
+ return (1 << 18) - (sign ? 0 : 1);
+ } else {
+ u32 ret = (integer << 16) | (fraction >> 16);
+ if (sign)
+ ret = -ret;
+ return ret & GENMASK(18, 0);
+ }
+}
+
+static void
+base907c_csc(struct nv50_wndw *wndw, struct nv50_wndw_atom *asyw,
+ const struct drm_color_ctm *ctm)
+{
+ int i, j;
+
+ for (j = 0; j < 3; j++) {
+ for (i = 0; i < 4; i++) {
+ u32 *val = &asyw->csc.matrix[j * 4 + i];
+ /* DRM does not support constant offset, while
+ * HW CSC does. Skip it. */
+ if (i == 3) {
+ *val = 0;
+ } else {
+ *val = csc_drm_to_base(ctm->matrix[j * 3 + i]);
+ }
+ }
+ }
+}
+
+static void
+base907c_csc_clr(struct nv50_wndw *wndw)
+{
+ u32 *push;
+ if ((push = evo_wait(&wndw->wndw, 2))) {
+ evo_mthd(push, 0x0140, 1);
+ evo_data(push, 0x00000000);
+ evo_kick(push, &wndw->wndw);
+ }
+}
+
+static void
+base907c_csc_set(struct nv50_wndw *wndw, struct nv50_wndw_atom *asyw)
+{
+ u32 *push, i;
+ if ((push = evo_wait(&wndw->wndw, 13))) {
+ evo_mthd(push, 0x0140, 12);
+ evo_data(push, asyw->csc.matrix[0] | 0x80000000);
+ for (i = 1; i < 12; i++)
+ evo_data(push, asyw->csc.matrix[i]);
+ evo_kick(push, &wndw->wndw);
+ }
+}
+
const struct nv50_wndw_func
base907c = {
.acquire = base507c_acquire,
@@ -94,6 +156,9 @@ base907c = {
.ntfy_clr = base507c_ntfy_clr,
.ntfy_wait_begun = base507c_ntfy_wait_begun,
.ilut = base907c_ilut,
+ .csc = base907c_csc,
+ .csc_set = base907c_csc_set,
+ .csc_clr = base907c_csc_clr,
.olut_core = true,
.xlut_set = base907c_xlut_set,
.xlut_clr = base907c_xlut_clr,
@@ -118,6 +118,7 @@ nv50_wndw_flush_clr(struct nv50_wndw *wndw, u32 *interlock, bool flush,
if (clr.sema ) wndw->func-> sema_clr(wndw);
if (clr.ntfy ) wndw->func-> ntfy_clr(wndw);
if (clr.xlut ) wndw->func-> xlut_clr(wndw);
+ if (clr.csc ) wndw->func-> csc_clr(wndw);
if (clr.image) wndw->func->image_clr(wndw);
interlock[wndw->interlock.type] |= wndw->interlock.data;
@@ -145,6 +146,7 @@ nv50_wndw_flush_set(struct nv50_wndw *wndw, u32 *interlock,
wndw->func->xlut_set(wndw, asyw);
}
+ if (asyw->set.csc ) wndw->func->csc_set (wndw, asyw);
if (asyw->set.scale) wndw->func->scale_set(wndw, asyw);
if (asyw->set.point) {
if (asyw->set.point = false, asyw->set.mask)
@@ -334,6 +336,8 @@ nv50_wndw_atomic_check_lut(struct nv50_wndw *wndw,
asyw->xlut.handle = wndw->wndw.vram.handle;
asyw->xlut.i.buffer = !asyw->xlut.i.buffer;
asyw->set.xlut = true;
+ } else {
+ asyw->clr.xlut = armw->xlut.handle != 0;;
}
/* Handle setting base SET_OUTPUT_LUT_LO_ENABLE_USE_CORE_LUT. */
@@ -341,6 +345,16 @@ nv50_wndw_atomic_check_lut(struct nv50_wndw *wndw,
(!armw->visible || (armw->xlut.handle && !asyw->xlut.handle)))
asyw->set.xlut = true;
+ if (wndw->func->csc && asyh->state.ctm) {
+ const struct drm_color_ctm *ctm = asyh->state.ctm->data;
+ wndw->func->csc(wndw, asyw, ctm);
+ asyw->csc.valid = true;
+ asyw->set.csc = true;
+ } else {
+ asyw->csc.valid = false;
+ asyw->clr.csc = armw->csc.valid;
+ }
+
/* Can't do an immediate flip while changing the LUT. */
asyh->state.pageflip_flags &= ~DRM_MODE_PAGE_FLIP_ASYNC;
}
@@ -410,6 +424,7 @@ nv50_wndw_atomic_check(struct drm_plane *plane, struct drm_plane_state *state)
asyw->clr.ntfy = armw->ntfy.handle != 0;
asyw->clr.sema = armw->sema.handle != 0;
asyw->clr.xlut = armw->xlut.handle != 0;
+ asyw->clr.csc = armw->csc.valid;
if (wndw->func->image_clr)
asyw->clr.image = armw->image.handle[0] != 0;
}
@@ -501,6 +516,7 @@ nv50_wndw_atomic_duplicate_state(struct drm_plane *plane)
asyw->ntfy = armw->ntfy;
asyw->ilut = NULL;
asyw->xlut = armw->xlut;
+ asyw->csc = armw->csc;
asyw->image = armw->image;
asyw->point = armw->point;
asyw->clr.mask = 0;
@@ -65,6 +65,10 @@ struct nv50_wndw_func {
int (*ntfy_wait_begun)(struct nouveau_bo *, u32 offset,
struct nvif_device *);
void (*ilut)(struct nv50_wndw *, struct nv50_wndw_atom *);
+ void (*csc)(struct nv50_wndw *, struct nv50_wndw_atom *,
+ const struct drm_color_ctm *);
+ void (*csc_set)(struct nv50_wndw *, struct nv50_wndw_atom *);
+ void (*csc_clr)(struct nv50_wndw *);
bool ilut_identity;
bool olut_core;
void (*xlut_set)(struct nv50_wndw *, struct nv50_wndw_atom *);
This adds support on GF119:GV100 (exclusive) for CTM (aka CSC). Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu> --- v1 -> v2: - ctm -> csc - mark csc.valid = false when there is no ctm property drivers/gpu/drm/nouveau/dispnv50/atom.h | 6 ++ drivers/gpu/drm/nouveau/dispnv50/base907c.c | 65 +++++++++++++++++++++ drivers/gpu/drm/nouveau/dispnv50/wndw.c | 16 +++++ drivers/gpu/drm/nouveau/dispnv50/wndw.h | 4 ++ 4 files changed, 91 insertions(+)