1 files changed, 83 insertions, 136 deletions

diff --git a/src/gallium/winsys/r600/drm/radeon_ctx.c b/src/gallium/winsys/r600/drm/radeon_ctx.c
index 1a12c1023e..50c5e4741e 100644
--- a/src/gallium/winsys/r600/drm/radeon_ctx.c
+++ b/src/gallium/winsys/r600/drm/radeon_ctx.c
@@ -30,21 +30,16 @@
 #include "radeon_drm.h"
 #include "bof.h"
 
-int radeon_ctx_set_bo_new(struct radeon_ctx *ctx, struct radeon_bo *bo)
+static int radeon_ctx_set_bo_new(struct radeon_ctx *ctx, struct radeon_bo *bo)
 {
-	void *ptr;
-
-	ptr = realloc(ctx->bo, sizeof(struct radeon_bo) * (ctx->nbo + 1));
-	if (ptr == NULL) {
-		return -ENOMEM;
-	}
-	ctx->bo = ptr;
+	if (ctx->nbo >= RADEON_CTX_MAX_PM4)
+		return -EBUSY;
 	ctx->bo[ctx->nbo] = bo;
 	ctx->nbo++;
 	return 0;
 }
 
-struct radeon_bo *radeon_ctx_get_bo(struct radeon_ctx *ctx, unsigned reloc)
+static struct radeon_bo *radeon_ctx_get_bo(struct radeon_ctx *ctx, unsigned reloc)
 {
 	struct radeon_cs_reloc *greloc;
 	unsigned i;
@@ -59,7 +54,7 @@ struct radeon_bo *radeon_ctx_get_bo(struct radeon_ctx *ctx, unsigned reloc)
 	return NULL;
 }
 
-void radeon_ctx_get_placement(struct radeon_ctx *ctx, unsigned reloc, u32 *placement)
+static void radeon_ctx_get_placement(struct radeon_ctx *ctx, unsigned reloc, u32 *placement)
 {
 	struct radeon_cs_reloc *greloc;
 	unsigned i;
@@ -76,50 +71,58 @@ void radeon_ctx_get_placement(struct radeon_ctx *ctx, unsigned reloc, u32 *place
 	}
 }
 
-struct radeon_ctx *radeon_ctx(struct radeon *radeon)
+void radeon_ctx_clear(struct radeon_ctx *ctx)
 {
-	struct radeon_ctx *ctx;
-
-	if (radeon == NULL)
-		return NULL;
-	ctx = calloc(1, sizeof(*ctx));
-	if (ctx == NULL)
-		return NULL;
-	ctx->radeon = radeon_incref(radeon);
-	return ctx;
+	for (int i = 0; i < ctx->nbo; i++) {
+		ctx->bo[i] = radeon_bo_decref(ctx->radeon, ctx->bo[i]);
+	}
+	ctx->ndwords = RADEON_CTX_MAX_PM4;
+	ctx->cdwords = 0;
+	ctx->nreloc = 0;
+	ctx->nbo = 0;
 }
 
-struct radeon_ctx *radeon_ctx_incref(struct radeon_ctx *ctx)
+int radeon_ctx_init(struct radeon_ctx *ctx, struct radeon *radeon)
 {
-	ctx->refcount++;
-	return ctx;
+	if (radeon == NULL)
+		return -EINVAL;
+	memset(ctx, 0, sizeof(struct radeon_ctx));
+	ctx->radeon = radeon_incref(radeon);
+	radeon_ctx_clear(ctx);
+	ctx->pm4 = malloc(RADEON_CTX_MAX_PM4 * 4);
+	if (ctx->pm4 == NULL) {
+		radeon_ctx_fini(ctx);
+		return -ENOMEM;
+	}
+	ctx->reloc = malloc(sizeof(struct radeon_cs_reloc) * RADEON_CTX_MAX_PM4);
+	if (ctx->reloc == NULL) {
+		radeon_ctx_fini(ctx);
+		return -ENOMEM;
+	}
+	ctx->bo = malloc(sizeof(void *) * RADEON_CTX_MAX_PM4);
+	if (ctx->bo == NULL) {
+		radeon_ctx_fini(ctx);
+		return -ENOMEM;
+	}
+	return 0;
 }
 
-struct radeon_ctx *radeon_ctx_decref(struct radeon_ctx *ctx)
+void radeon_ctx_fini(struct radeon_ctx *ctx)
 {
 	unsigned i;
 
 	if (ctx == NULL)
-		return NULL;
-	if (--ctx->refcount > 0) {
-		return NULL;
-	}
+		return;
 
-	for (i = 0; i < ctx->ndraw; i++) {
-		ctx->draw[i] = radeon_draw_decref(ctx->draw[i]);
-	}
 	for (i = 0; i < ctx->nbo; i++) {
 		ctx->bo[i] = radeon_bo_decref(ctx->radeon, ctx->bo[i]);
 	}
 	ctx->radeon = radeon_decref(ctx->radeon);
-	free(ctx->state);
-	free(ctx->draw);
 	free(ctx->bo);
 	free(ctx->pm4);
 	free(ctx->reloc);
 	memset(ctx, 0, sizeof(*ctx));
 	free(ctx);
-	return NULL;
 }
 
 static int radeon_ctx_state_bo(struct radeon_ctx *ctx, struct radeon_state *state)
@@ -152,17 +155,17 @@ int radeon_ctx_submit(struct radeon_ctx *ctx)
 	uint64_t chunk_array[2];
 	int r = 0;
 
-	if (!ctx->cpm4)
+	if (!ctx->cdwords)
 		return 0;
 #if 0
-	for (r = 0; r < ctx->cpm4; r++) {
+	for (r = 0; r < ctx->cdwords; r++) {
 		fprintf(stderr, "0x%08X\n", ctx->pm4[r]);
 	}
 #endif
 	drmib.num_chunks = 2;
 	drmib.chunks = (uint64_t)(uintptr_t)chunk_array;
 	chunks[0].chunk_id = RADEON_CHUNK_ID_IB;
-	chunks[0].length_dw = ctx->cpm4;
+	chunks[0].length_dw = ctx->cdwords;
 	chunks[0].chunk_data = (uint64_t)(uintptr_t)ctx->pm4;
 	chunks[1].chunk_id = RADEON_CHUNK_ID_RELOCS;
 	chunks[1].length_dw = ctx->nreloc * sizeof(struct radeon_cs_reloc) / 4;
@@ -180,7 +183,6 @@ static int radeon_ctx_reloc(struct radeon_ctx *ctx, struct radeon_bo *bo,
 			unsigned id, unsigned *placement)
 {
 	unsigned i;
-	struct radeon_cs_reloc *ptr;
 
 	for (i = 0; i < ctx->nreloc; i++) {
 		if (ctx->reloc[i].handle == bo->handle) {
@@ -188,14 +190,13 @@ static int radeon_ctx_reloc(struct radeon_ctx *ctx, struct radeon_bo *bo,
 			return 0;
 		}
 	}
-	ptr = realloc(ctx->reloc, sizeof(struct radeon_cs_reloc) * (ctx->nreloc + 1));
-	if (ptr == NULL)
-		return -ENOMEM;
-	ctx->reloc = ptr;
-	ptr[ctx->nreloc].handle = bo->handle;
-	ptr[ctx->nreloc].read_domain = placement[0] | placement [1];
-	ptr[ctx->nreloc].write_domain = placement[0] | placement [1];
-	ptr[ctx->nreloc].flags = 0;
+	if (ctx->nreloc >= RADEON_CTX_MAX_PM4) {
+		return -EBUSY;
+	}
+	ctx->reloc[ctx->nreloc].handle = bo->handle;
+	ctx->reloc[ctx->nreloc].read_domain = placement[0] | placement [1];
+	ctx->reloc[ctx->nreloc].write_domain = placement[0] | placement [1];
+	ctx->reloc[ctx->nreloc].flags = 0;
 	ctx->pm4[id] = ctx->nreloc * sizeof(struct radeon_cs_reloc) / 4;
 	ctx->nreloc++;
 	return 0;
@@ -208,11 +209,14 @@ static int radeon_ctx_state_schedule(struct radeon_ctx *ctx, struct radeon_state
 
 	if (state == NULL)
 		return 0;
-	memcpy(&ctx->pm4[ctx->id], state->pm4, state->cpm4 * 4);
+	if (state->cpm4 > ctx->ndwords) {
+		return -EBUSY;
+	}
+	memcpy(&ctx->pm4[ctx->cdwords], state->pm4, state->cpm4 * 4);
 	for (i = 0; i < state->nreloc; i++) {
 		rid = state->reloc_pm4_id[i];
 		bid = state->reloc_bo_id[i];
-		cid = ctx->id + rid;
+		cid = ctx->cdwords + rid;
 		r = radeon_ctx_reloc(ctx, state->bo[bid], cid,
 					&state->placement[bid * 2]);
 		if (r) {
@@ -220,120 +224,65 @@ static int radeon_ctx_state_schedule(struct radeon_ctx *ctx, struct radeon_state
 			return r;
 		}
 	}
-	ctx->id += state->cpm4;
+	ctx->cdwords += state->cpm4;
+	ctx->ndwords -= state->cpm4;
 	return 0;
 }
 
 int radeon_ctx_set_query_state(struct radeon_ctx *ctx, struct radeon_state *state)
 {
-	void *tmp;
 	int r = 0;
 
 	/* !!! ONLY ACCEPT QUERY STATE HERE !!! */
-	if (state->stype->stype != R600_STATE_QUERY_BEGIN && state->stype->stype != R600_STATE_QUERY_END) {
-		return -EINVAL;
-	}
 	r = radeon_state_pm4(state);
 	if (r)
 		return r;
-	if ((ctx->draw_cpm4 + state->cpm4) > RADEON_CTX_MAX_PM4) {
-		/* need to flush */
-		return -EBUSY;
-	}
-	if (state->cpm4 >= RADEON_CTX_MAX_PM4) {
-		fprintf(stderr, "%s single state too big %d, max %d\n",
-			__func__, state->cpm4, RADEON_CTX_MAX_PM4);
-		return -EINVAL;
-	}
-	tmp = realloc(ctx->state, (ctx->nstate + 1) * sizeof(void*));
-	if (tmp == NULL)
-		return -ENOMEM;
-	ctx->state = tmp;
-	ctx->state[ctx->nstate++] = radeon_state_incref(state);
 	/* BEGIN/END query are balanced in the same cs so account for END
 	 * END query when scheduling BEGIN query
 	 */
-	if (state->stype->stype == R600_STATE_QUERY_BEGIN) {
-		ctx->draw_cpm4 += state->cpm4 * 2;
+	switch (state->stype->stype) {
+	case R600_STATE_QUERY_BEGIN:
+		/* is there enough place for begin & end */
+		if ((state->cpm4 * 2) > ctx->ndwords)
+			return -EBUSY;
+		ctx->ndwords -= state->cpm4;
+		break;
+	case R600_STATE_QUERY_END:
+		ctx->ndwords += state->cpm4;
+		break;
+	default:
+		return -EINVAL;
 	}
-	return 0;
+	return radeon_ctx_state_schedule(ctx, state);
 }
 
-int radeon_ctx_set_draw_new(struct radeon_ctx *ctx, struct radeon_draw *draw)
+int radeon_ctx_set_draw(struct radeon_ctx *ctx, struct radeon_draw *draw)
 {
-	struct radeon_draw *pdraw = NULL;
-	struct radeon_draw **ndraw;
-	struct radeon_state *nstate, *ostate;
-	unsigned cpm4, i, cstate;
-	void *tmp;
+	unsigned previous_cdwords;
 	int r = 0;
 
-	ndraw = realloc(ctx->draw, sizeof(void*) * (ctx->ndraw + 1));
-	if (ndraw == NULL)
-		return -ENOMEM;
-	ctx->draw = ndraw;
-	for (i = 0; i < draw->nstate; i++) {
+	for (int i = 0; i < (ctx->radeon->nstate_per_shader * R600_SHADER_MAX); i++) {
 		r = radeon_ctx_state_bo(ctx, draw->state[i]);
 		if (r)
 			return r;
 	}
-	r = radeon_draw_check(draw);
-	if (r)
-		return r;
-	if (draw->cpm4 >= RADEON_CTX_MAX_PM4) {
-		fprintf(stderr, "%s single draw too big %d, max %d\n",
-			__func__, draw->cpm4, RADEON_CTX_MAX_PM4);
-		return -EINVAL;
-	}
-	tmp = realloc(ctx->state, (ctx->nstate + draw->nstate) * sizeof(void*));
-	if (tmp == NULL)
-		return -ENOMEM;
-	ctx->state = tmp;
-	pdraw = ctx->cdraw;
-	for (i = 0, cpm4 = 0, cstate = ctx->nstate; i < draw->nstate - 1; i++) {
-		nstate = draw->state[i];
-		if (nstate) {
-			if (pdraw && pdraw->state[i]) {
-				ostate = pdraw->state[i];
-				if (ostate->pm4_crc != nstate->pm4_crc) {
-					ctx->state[cstate++] = nstate;
-					cpm4 += nstate->cpm4;
+	previous_cdwords = ctx->cdwords;
+	for (int i = 0, id = 0; i < ctx->radeon->nstate_per_shader; i++) {
+		for (int j = 0; j < R600_SHADER_MAX; j++) {
+			id = j * ctx->radeon->nstate_per_shader + i;
+			if (draw->state[id]) {
+				r = radeon_ctx_state_schedule(ctx, draw->state[id]);
+				if (r) {
+					ctx->cdwords = previous_cdwords;
+					return r;
 				}
-			} else {
-				ctx->state[cstate++] = nstate;
-				cpm4 += nstate->cpm4;
 			}
 		}
 	}
-	/* The last state is the draw state always add it */
-	if (draw->state[i] == NULL) {
-		fprintf(stderr, "%s no draw command\n", __func__);
-		return -EINVAL;
-	}
-	ctx->state[cstate++] = draw->state[i];
-	cpm4 += draw->state[i]->cpm4;
-	if ((ctx->draw_cpm4 + cpm4) > RADEON_CTX_MAX_PM4) {
-		/* need to flush */
-		return -EBUSY;
-	}
-	ctx->draw_cpm4 += cpm4;
-	ctx->nstate = cstate;
-	ctx->draw[ctx->ndraw++] = draw;
-	ctx->cdraw = draw;
 	return 0;
 }
 
-int radeon_ctx_set_draw(struct radeon_ctx *ctx, struct radeon_draw *draw)
-{
-	int r;
-
-	radeon_draw_incref(draw);
-	r = radeon_ctx_set_draw_new(ctx, draw);
-	if (r)
-		radeon_draw_decref(draw);
-	return r;
-}
-
+#if 0
 int radeon_ctx_pm4(struct radeon_ctx *ctx)
 {
 	unsigned i;
@@ -345,9 +294,6 @@ int radeon_ctx_pm4(struct radeon_ctx *ctx)
 	if (ctx->pm4 == NULL)
 		return -EINVAL;
 	for (i = 0, ctx->id = 0; i < ctx->nstate; i++) {
-		r = radeon_ctx_state_schedule(ctx, ctx->state[i]);
-		if (r)
-			return r;
 	}
 	if (ctx->id != ctx->draw_cpm4) {
 		fprintf(stderr, "%s miss predicted pm4 size %d for %d\n",
@@ -357,6 +303,7 @@ int radeon_ctx_pm4(struct radeon_ctx *ctx)
 	ctx->cpm4 = ctx->draw_cpm4;
 	return 0;
 }
+#endif
 
 void radeon_ctx_dump_bof(struct radeon_ctx *ctx, const char *file)
 {
@@ -384,8 +331,8 @@ printf("%d relocs\n", ctx->nreloc);
 	bof_decref(blob);
 	blob = NULL;
 	/* dump cs */
-printf("%d pm4\n", ctx->cpm4);
-	blob = bof_blob(ctx->cpm4 * 4, ctx->pm4);
+printf("%d pm4\n", ctx->cdwords);
+	blob = bof_blob(ctx->cdwords * 4, ctx->pm4);
 	if (blob == NULL)
 		goto out_err;
 	if (bof_object_set(root, "pm4", blob))