radeonsi: use queries from r600g

Reviewed-by: Michel Dänzer <[email protected]>
Reviewed-by: Tom Stellard <[email protected]>

9 files changed, 40 insertions, 757 deletions

diff --git a/src/gallium/drivers/radeonsi/Makefile.sources b/src/gallium/drivers/radeonsi/Makefile.sources
index 33f1492f6ce..56295720f4d 100644
--- a/src/gallium/drivers/radeonsi/Makefile.sources
+++ b/src/gallium/drivers/radeonsi/Makefile.sources
@@ -7,7 +7,6 @@ C_SOURCES := \
 	si_hw_context.c \
 	si_pipe.c \
 	si_pm4.c \
-	si_query.c \
 	si_resource.c \
 	si_shader.c \
 	si_state.c \
diff --git a/src/gallium/drivers/radeonsi/si.h b/src/gallium/drivers/radeonsi/si.h
index 46184ec33bd..5c69b07a05a 100644
--- a/src/gallium/drivers/radeonsi/si.h
+++ b/src/gallium/drivers/radeonsi/si.h
@@ -33,56 +33,11 @@
 #include "si_resource.h"
 
 struct winsys_handle;
-
-/* R600/R700 STATES */
-struct si_query {
-	union {
-		uint64_t			u64;
-		boolean				b;
-		struct pipe_query_data_so_statistics so;
-	} result;
-	/* The kind of query */
-	unsigned				type;
-	/* Offset of the first result for current query */
-	unsigned				results_start;
-	/* Offset of the next free result after current query data */
-	unsigned				results_end;
-	/* Size of the result in memory for both begin_query and end_query,
-	 * this can be one or two numbers, or it could even be a size of a structure. */
-	unsigned				result_size;
-	/* The buffer where query results are stored. It's used as a ring,
-	 * data blocks for current query are stored sequentially from
-	 * results_start to results_end, with wrapping on the buffer end */
-	struct r600_resource			*buffer;
-	/* The number of dwords for begin_query or end_query. */
-	unsigned				num_cs_dw;
-	/* linked list of queries */
-	struct list_head			list;
-};
-
 struct si_context;
 struct si_screen;
 
-void si_get_backend_mask(struct si_context *ctx);
 void si_context_flush(struct si_context *ctx, unsigned flags);
 void si_begin_new_cs(struct si_context *ctx);
-
-struct si_query *si_context_query_create(struct si_context *ctx, unsigned query_type);
-void si_context_query_destroy(struct si_context *ctx, struct si_query *query);
-boolean si_context_query_result(struct si_context *ctx,
-				struct si_query *query,
-				boolean wait, void *vresult);
-void si_query_begin(struct si_context *ctx, struct si_query *query);
-void si_query_end(struct si_context *ctx, struct si_query *query);
-void si_context_queries_suspend(struct si_context *ctx);
-void si_context_queries_resume(struct si_context *ctx);
-void si_query_predication(struct si_context *ctx, struct si_query *query, int operation,
-			    int flag_wait);
-
-bool si_is_timer_query(unsigned type);
-bool si_query_needs_begin(unsigned type);
 void si_need_cs_space(struct si_context *ctx, unsigned num_dw, boolean count_draw_in);
 
-int si_context_init(struct si_context *ctx);
-
 #endif
diff --git a/src/gallium/drivers/radeonsi/si_blit.c b/src/gallium/drivers/radeonsi/si_blit.c
index ba2ebe7d7f0..250caab64e1 100644
--- a/src/gallium/drivers/radeonsi/si_blit.c
+++ b/src/gallium/drivers/radeonsi/si_blit.c
@@ -51,7 +51,7 @@ static void si_blitter_begin(struct pipe_context *ctx, enum si_blitter_op op)
 {
 	struct si_context *sctx = (struct si_context *)ctx;
 
-	si_context_queries_suspend(sctx);
+	r600_suspend_nontimer_queries(&sctx->b);
 
 	util_blitter_save_blend(sctx->blitter, sctx->queued.named.blend);
 	util_blitter_save_depth_stencil_alpha(sctx->blitter, sctx->queued.named.dsa);
@@ -81,26 +81,18 @@ static void si_blitter_begin(struct pipe_context *ctx, enum si_blitter_op op)
 			sctx->samplers[PIPE_SHADER_FRAGMENT].views.views);
 	}
 
-	if ((op & SI_DISABLE_RENDER_COND) && sctx->current_render_cond) {
-		sctx->saved_render_cond = sctx->current_render_cond;
-		sctx->saved_render_cond_cond = sctx->current_render_cond_cond;
-		sctx->saved_render_cond_mode = sctx->current_render_cond_mode;
-		sctx->b.b.render_condition(&sctx->b.b, NULL, FALSE, 0);
+	if ((op & SI_DISABLE_RENDER_COND) && sctx->b.current_render_cond) {
+		util_blitter_save_render_condition(sctx->blitter,
+                                                   sctx->b.current_render_cond,
+                                                   sctx->b.current_render_cond_cond,
+                                                   sctx->b.current_render_cond_mode);
 	}
-
 }
 
 static void si_blitter_end(struct pipe_context *ctx)
 {
 	struct si_context *sctx = (struct si_context *)ctx;
-	if (sctx->saved_render_cond) {
-		sctx->b.b.render_condition(&sctx->b.b,
-					       sctx->saved_render_cond,
-					       sctx->saved_render_cond_cond,
-					       sctx->saved_render_cond_mode);
-		sctx->saved_render_cond = NULL;
-	}
-	si_context_queries_resume(sctx);
+	r600_resume_nontimer_queries(&sctx->b);
 }
 
 static unsigned u_max_sample(struct pipe_resource *r)
diff --git a/src/gallium/drivers/radeonsi/si_hw_context.c b/src/gallium/drivers/radeonsi/si_hw_context.c
index 517becbfec5..a7536e3fbbf 100644
--- a/src/gallium/drivers/radeonsi/si_hw_context.c
+++ b/src/gallium/drivers/radeonsi/si_hw_context.c
@@ -30,97 +30,6 @@
 #include "util/u_memory.h"
 #include <errno.h>
 
-#define GROUP_FORCE_NEW_BLOCK	0
-
-/* Get backends mask */
-void si_get_backend_mask(struct si_context *ctx)
-{
-	struct radeon_winsys_cs *cs = ctx->b.rings.gfx.cs;
-	struct r600_resource *buffer;
-	uint32_t *results;
-	unsigned num_backends = ctx->screen->b.info.r600_num_backends;
-	unsigned i, mask = 0;
-
-	/* if backend_map query is supported by the kernel */
-	if (ctx->screen->b.info.r600_backend_map_valid) {
-		unsigned num_tile_pipes = ctx->screen->b.info.r600_num_tile_pipes;
-		unsigned backend_map = ctx->screen->b.info.r600_backend_map;
-		unsigned item_width = 4, item_mask = 0x7;
-
-		while(num_tile_pipes--) {
-			i = backend_map & item_mask;
-			mask |= (1<<i);
-			backend_map >>= item_width;
-		}
-		if (mask != 0) {
-			ctx->backend_mask = mask;
-			return;
-		}
-	}
-
-	/* otherwise backup path for older kernels */
-
-	/* create buffer for event data */
-	buffer = si_resource_create_custom(&ctx->screen->b.b,
-					   PIPE_USAGE_STAGING,
-					   ctx->max_db*16);
-	if (!buffer)
-		goto err;
-
-	/* initialize buffer with zeroes */
-	results = ctx->b.ws->buffer_map(buffer->cs_buf, ctx->b.rings.gfx.cs, PIPE_TRANSFER_WRITE);
-	if (results) {
-		uint64_t va = 0;
-
-		memset(results, 0, ctx->max_db * 4 * 4);
-		ctx->b.ws->buffer_unmap(buffer->cs_buf);
-
-		/* emit EVENT_WRITE for ZPASS_DONE */
-		va = r600_resource_va(&ctx->screen->b.b, (void *)buffer);
-		cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE, 2, 0);
-		cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_ZPASS_DONE) | EVENT_INDEX(1);
-		cs->buf[cs->cdw++] = va;
-		cs->buf[cs->cdw++] = va >> 32;
-
-		cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
-		cs->buf[cs->cdw++] = r600_context_bo_reloc(&ctx->b, &ctx->b.rings.gfx, buffer, RADEON_USAGE_WRITE);
-
-		/* analyze results */
-		results = ctx->b.ws->buffer_map(buffer->cs_buf, ctx->b.rings.gfx.cs, PIPE_TRANSFER_READ);
-		if (results) {
-			for(i = 0; i < ctx->max_db; i++) {
-				/* at least highest bit will be set if backend is used */
-				if (results[i*4 + 1])
-					mask |= (1<<i);
-			}
-			ctx->b.ws->buffer_unmap(buffer->cs_buf);
-		}
-	}
-
-	r600_resource_reference(&buffer, NULL);
-
-	if (mask != 0) {
-		ctx->backend_mask = mask;
-		return;
-	}
-
-err:
-	/* fallback to old method - set num_backends lower bits to 1 */
-	ctx->backend_mask = (~((uint32_t)0))>>(32-num_backends);
-	return;
-}
-
-bool si_is_timer_query(unsigned type)
-{
-	return type == PIPE_QUERY_TIME_ELAPSED ||
-		type == PIPE_QUERY_TIMESTAMP ||
-		type == PIPE_QUERY_TIMESTAMP_DISJOINT;
-}
-
-bool si_query_needs_begin(unsigned type)
-{
-	return type != PIPE_QUERY_TIMESTAMP;
-}
 
 /* initialize */
 void si_need_cs_space(struct si_context *ctx, unsigned num_dw,
@@ -146,7 +55,7 @@ void si_need_cs_space(struct si_context *ctx, unsigned num_dw,
 	}
 
 	/* Count in queries_suspend. */
-	num_dw += ctx->num_cs_dw_nontimer_queries_suspend;
+	num_dw += ctx->b.num_cs_dw_nontimer_queries_suspend;
 
 	/* Count in streamout_end at the end of CS. */
 	if (ctx->b.streamout.begin_emitted) {
@@ -154,7 +63,7 @@ void si_need_cs_space(struct si_context *ctx, unsigned num_dw,
 	}
 
 	/* Count in render_condition(NULL) at the end of CS. */
-	if (ctx->predicate_drawing) {
+	if (ctx->b.predicate_drawing) {
 		num_dw += 3;
 	}
 
@@ -181,10 +90,10 @@ void si_context_flush(struct si_context *ctx, unsigned flags)
 		return;
 
 	/* suspend queries */
-	ctx->nontimer_queries_suspended = false;
-	if (ctx->num_cs_dw_nontimer_queries_suspend) {
-		si_context_queries_suspend(ctx);
-		ctx->nontimer_queries_suspended = true;
+	ctx->b.nontimer_queries_suspended = false;
+	if (ctx->b.num_cs_dw_nontimer_queries_suspend) {
+		r600_suspend_nontimer_queries(&ctx->b);
+		ctx->b.nontimer_queries_suspended = true;
 	}
 
 	ctx->b.streamout.suspended = false;
@@ -270,431 +179,13 @@ void si_begin_new_cs(struct si_context *ctx)
 	}
 
 	/* resume queries */
-	if (ctx->nontimer_queries_suspended) {
-		si_context_queries_resume(ctx);
+	if (ctx->b.nontimer_queries_suspended) {
+		r600_resume_nontimer_queries(&ctx->b);
 	}
 
 	si_all_descriptors_begin_new_cs(ctx);
 }
 
-static unsigned si_query_read_result(char *map, unsigned start_index, unsigned end_index,
-				     bool test_status_bit)
-{
-	uint32_t *current_result = (uint32_t*)map;
-	uint64_t start, end;
-
-	start = (uint64_t)current_result[start_index] |
-		(uint64_t)current_result[start_index+1] << 32;
-	end = (uint64_t)current_result[end_index] |
-	      (uint64_t)current_result[end_index+1] << 32;
-
-	if (!test_status_bit ||
-	    ((start & 0x8000000000000000UL) && (end & 0x8000000000000000UL))) {
-		return end - start;
-	}
-	return 0;
-}
-
-static boolean si_query_result(struct si_context *ctx, struct si_query *query, boolean wait)
-{
-	unsigned results_base = query->results_start;
-	char *map;
-
-	map = ctx->b.ws->buffer_map(query->buffer->cs_buf, ctx->b.rings.gfx.cs,
-				  PIPE_TRANSFER_READ |
-				  (wait ? 0 : PIPE_TRANSFER_DONTBLOCK));
-	if (!map)
-		return FALSE;
-
-	/* count all results across all data blocks */
-	switch (query->type) {
-	case PIPE_QUERY_OCCLUSION_COUNTER:
-		while (results_base != query->results_end) {
-			query->result.u64 +=
-				si_query_read_result(map + results_base, 0, 2, true);
-			results_base = (results_base + 16) % query->buffer->b.b.width0;
-		}
-		break;
-	case PIPE_QUERY_OCCLUSION_PREDICATE:
-		while (results_base != query->results_end) {
-			query->result.b = query->result.b ||
-				si_query_read_result(map + results_base, 0, 2, true) != 0;
-			results_base = (results_base + 16) % query->buffer->b.b.width0;
-		}
-		break;
-	case PIPE_QUERY_TIMESTAMP:
-	{
-		uint32_t *current_result = (uint32_t*)map;
-		query->result.u64 = (uint64_t)current_result[0] | (uint64_t)current_result[1] << 32;
-		break;
-	}
-	case PIPE_QUERY_TIME_ELAPSED:
-		while (results_base != query->results_end) {
-			query->result.u64 +=
-				si_query_read_result(map + results_base, 0, 2, false);
-			results_base = (results_base + query->result_size) % query->buffer->b.b.width0;
-		}
-		break;
-	case PIPE_QUERY_PRIMITIVES_EMITTED:
-		/* SAMPLE_STREAMOUTSTATS stores this structure:
-		 * {
-		 *    u64 NumPrimitivesWritten;
-		 *    u64 PrimitiveStorageNeeded;
-		 * }
-		 * We only need NumPrimitivesWritten here. */
-		while (results_base != query->results_end) {
-			query->result.u64 +=
-				si_query_read_result(map + results_base, 2, 6, true);
-			results_base = (results_base + query->result_size) % query->buffer->b.b.width0;
-		}
-		break;
-	case PIPE_QUERY_PRIMITIVES_GENERATED:
-		/* Here we read PrimitiveStorageNeeded. */
-		while (results_base != query->results_end) {
-			query->result.u64 +=
-				si_query_read_result(map + results_base, 0, 4, true);
-			results_base = (results_base + query->result_size) % query->buffer->b.b.width0;
-		}
-		break;
-	case PIPE_QUERY_SO_STATISTICS:
-		while (results_base != query->results_end) {
-			query->result.so.num_primitives_written +=
-				si_query_read_result(map + results_base, 2, 6, true);
-			query->result.so.primitives_storage_needed +=
-				si_query_read_result(map + results_base, 0, 4, true);
-			results_base = (results_base + query->result_size) % query->buffer->b.b.width0;
-		}
-		break;
-	case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
-		while (results_base != query->results_end) {
-			query->result.b = query->result.b ||
-				si_query_read_result(map + results_base, 2, 6, true) !=
-				si_query_read_result(map + results_base, 0, 4, true);
-			results_base = (results_base + query->result_size) % query->buffer->b.b.width0;
-		}
-		break;
-	default:
-		assert(0);
-	}
-
-	query->results_start = query->results_end;
-	ctx->b.ws->buffer_unmap(query->buffer->cs_buf);
-	return TRUE;
-}
-
-void si_query_begin(struct si_context *ctx, struct si_query *query)
-{
-	struct radeon_winsys_cs *cs = ctx->b.rings.gfx.cs;
-	unsigned new_results_end, i;
-	uint32_t *results;
-	uint64_t va;
-
-	si_need_cs_space(ctx, query->num_cs_dw * 2, TRUE);
-
-	new_results_end = (query->results_end + query->result_size) % query->buffer->b.b.width0;
-
-	/* collect current results if query buffer is full */
-	if (new_results_end == query->results_start) {
-		si_query_result(ctx, query, TRUE);
-	}
-
-	switch (query->type) {
-	case PIPE_QUERY_OCCLUSION_COUNTER:
-	case PIPE_QUERY_OCCLUSION_PREDICATE:
-		results = ctx->b.ws->buffer_map(query->buffer->cs_buf, ctx->b.rings.gfx.cs, PIPE_TRANSFER_WRITE);
-		if (results) {
-			results = (uint32_t*)((char*)results + query->results_end);
-			memset(results, 0, query->result_size);
-
-			/* Set top bits for unused backends */
-			for (i = 0; i < ctx->max_db; i++) {
-				if (!(ctx->backend_mask & (1<<i))) {
-					results[(i * 4)+1] = 0x80000000;
-					results[(i * 4)+3] = 0x80000000;
-				}
-			}
-			ctx->b.ws->buffer_unmap(query->buffer->cs_buf);
-		}
-		break;
-	case PIPE_QUERY_TIME_ELAPSED:
-		break;
-	case PIPE_QUERY_PRIMITIVES_EMITTED:
-	case PIPE_QUERY_PRIMITIVES_GENERATED:
-	case PIPE_QUERY_SO_STATISTICS:
-	case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
-		results = ctx->b.ws->buffer_map(query->buffer->cs_buf, ctx->b.rings.gfx.cs, PIPE_TRANSFER_WRITE);
-		results = (uint32_t*)((char*)results + query->results_end);
-		memset(results, 0, query->result_size);
-		ctx->b.ws->buffer_unmap(query->buffer->cs_buf);
-		break;
-	default:
-		assert(0);
-	}
-
-	/* emit begin query */
-	va = r600_resource_va(&ctx->screen->b.b, (void*)query->buffer);
-	va += query->results_end;
-
-	switch (query->type) {
-	case PIPE_QUERY_OCCLUSION_COUNTER:
-	case PIPE_QUERY_OCCLUSION_PREDICATE:
-		cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE, 2, 0);
-		cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_ZPASS_DONE) | EVENT_INDEX(1);
-		cs->buf[cs->cdw++] = va;
-		cs->buf[cs->cdw++] = (va >> 32UL) & 0xFF;
-		break;
-	case PIPE_QUERY_PRIMITIVES_EMITTED:
-	case PIPE_QUERY_PRIMITIVES_GENERATED:
-	case PIPE_QUERY_SO_STATISTICS:
-	case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
-		cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE, 2, 0);
-		cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_SAMPLE_STREAMOUTSTATS) | EVENT_INDEX(3);
-		cs->buf[cs->cdw++] = va;
-		cs->buf[cs->cdw++] = (va >> 32UL) & 0xFF;
-		break;
-	case PIPE_QUERY_TIME_ELAPSED:
-		cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0);
-		cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5);
-		cs->buf[cs->cdw++] = va;
-		cs->buf[cs->cdw++] = (3 << 29) | ((va >> 32UL) & 0xFF);
-		cs->buf[cs->cdw++] = 0;
-		cs->buf[cs->cdw++] = 0;
-		break;
-	default:
-		assert(0);
-	}
-	cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
-	cs->buf[cs->cdw++] = r600_context_bo_reloc(&ctx->b, &ctx->b.rings.gfx, query->buffer, RADEON_USAGE_WRITE);
-
-	if (!si_is_timer_query(query->type)) {
-		ctx->num_cs_dw_nontimer_queries_suspend += query->num_cs_dw;
-	}
-}
-
-void si_query_end(struct si_context *ctx, struct si_query *query)
-{
-	struct radeon_winsys_cs *cs = ctx->b.rings.gfx.cs;
-	uint64_t va;
-	unsigned new_results_end;
-
-	/* The queries which need begin already called this in begin_query. */
-	if (!si_query_needs_begin(query->type)) {
-		si_need_cs_space(ctx, query->num_cs_dw, TRUE);
-
-		new_results_end = (query->results_end + query->result_size) % query->buffer->b.b.width0;
-
-		/* collect current results if query buffer is full */
-		if (new_results_end == query->results_start) {
-		si_query_result(ctx, query, TRUE);
-		}
-	}
-
-	va = r600_resource_va(&ctx->screen->b.b, (void*)query->buffer);
-	/* emit end query */
-	switch (query->type) {
-	case PIPE_QUERY_OCCLUSION_COUNTER:
-	case PIPE_QUERY_OCCLUSION_PREDICATE:
-		va += query->results_end + 8;
-		cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE, 2, 0);
-		cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_ZPASS_DONE) | EVENT_INDEX(1);
-		cs->buf[cs->cdw++] = va;
-		cs->buf[cs->cdw++] = (va >> 32UL) & 0xFF;
-		break;
-	case PIPE_QUERY_PRIMITIVES_EMITTED:
-	case PIPE_QUERY_PRIMITIVES_GENERATED:
-	case PIPE_QUERY_SO_STATISTICS:
-	case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
-		va += query->results_end + query->result_size/2;
-		cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE, 2, 0);
-		cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_SAMPLE_STREAMOUTSTATS) | EVENT_INDEX(3);
-		cs->buf[cs->cdw++] = va;
-		cs->buf[cs->cdw++] = (va >> 32UL) & 0xFF;
-		break;
-	case PIPE_QUERY_TIME_ELAPSED:
-		va += query->results_end + query->result_size/2;
-		/* fall through */
-	case PIPE_QUERY_TIMESTAMP:
-		cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0);
-		cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5);
-		cs->buf[cs->cdw++] = va;
-		cs->buf[cs->cdw++] = (3 << 29) | ((va >> 32UL) & 0xFF);
-		cs->buf[cs->cdw++] = 0;
-		cs->buf[cs->cdw++] = 0;
-		break;
-	default:
-		assert(0);
-	}
-	cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
-	cs->buf[cs->cdw++] = r600_context_bo_reloc(&ctx->b, &ctx->b.rings.gfx, query->buffer, RADEON_USAGE_WRITE);
-
-	query->results_end = (query->results_end + query->result_size) % query->buffer->b.b.width0;
-
-	if (si_query_needs_begin(query->type) && !si_is_timer_query(query->type)) {
-		ctx->num_cs_dw_nontimer_queries_suspend -= query->num_cs_dw;
-	}
-}
-
-void si_query_predication(struct si_context *ctx, struct si_query *query, int operation,
-			  int flag_wait)
-{
-	struct radeon_winsys_cs *cs = ctx->b.rings.gfx.cs;
-	uint64_t va;
-
-	if (operation == PREDICATION_OP_CLEAR) {
-		si_need_cs_space(ctx, 3, FALSE);
-
-		cs->buf[cs->cdw++] = PKT3(PKT3_SET_PREDICATION, 1, 0);
-		cs->buf[cs->cdw++] = 0;
-		cs->buf[cs->cdw++] = PRED_OP(PREDICATION_OP_CLEAR);
-	} else {
-		unsigned results_base = query->results_start;
-		unsigned count;
-		uint32_t op;
-
-		/* find count of the query data blocks */
-		count = (query->buffer->b.b.width0 + query->results_end - query->results_start) % query->buffer->b.b.width0;
-		count /= query->result_size;
-
-		si_need_cs_space(ctx, 5 * count, TRUE);
-
-		op = PRED_OP(operation) | PREDICATION_DRAW_VISIBLE |
-				(flag_wait ? PREDICATION_HINT_WAIT : PREDICATION_HINT_NOWAIT_DRAW);
-		va = r600_resource_va(&ctx->screen->b.b, (void*)query->buffer);
-
-		/* emit predicate packets for all data blocks */
-		while (results_base != query->results_end) {
-			cs->buf[cs->cdw++] = PKT3(PKT3_SET_PREDICATION, 1, 0);
-			cs->buf[cs->cdw++] = (va + results_base) & 0xFFFFFFFFUL;
-			cs->buf[cs->cdw++] = op | (((va + results_base) >> 32UL) & 0xFF);
-			cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
-			cs->buf[cs->cdw++] = r600_context_bo_reloc(&ctx->b, &ctx->b.rings.gfx,
-								   query->buffer, RADEON_USAGE_READ);
-			results_base = (results_base + query->result_size) % query->buffer->b.b.width0;
-
-			/* set CONTINUE bit for all packets except the first */
-			op |= PREDICATION_CONTINUE;
-		}
-	}
-}
-
-struct si_query *si_context_query_create(struct si_context *ctx, unsigned query_type)
-{
-	struct si_query *query;
-	unsigned buffer_size = 4096;
-
-	query = CALLOC_STRUCT(si_query);
-	if (query == NULL)
-		return NULL;
-
-	query->type = query_type;
-
-	switch (query_type) {
-	case PIPE_QUERY_OCCLUSION_COUNTER:
-	case PIPE_QUERY_OCCLUSION_PREDICATE:
-		query->result_size = 16 * ctx->max_db;
-		query->num_cs_dw = 6;
-		break;
-	case PIPE_QUERY_TIMESTAMP:
-		query->result_size = 8;
-		query->num_cs_dw = 8;
-		break;
-	case PIPE_QUERY_TIME_ELAPSED:
-		query->result_size = 16;
-		query->num_cs_dw = 8;
-		break;
-	case PIPE_QUERY_PRIMITIVES_EMITTED:
-	case PIPE_QUERY_PRIMITIVES_GENERATED:
-	case PIPE_QUERY_SO_STATISTICS:
-	case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
-		/* NumPrimitivesWritten, PrimitiveStorageNeeded. */
-		query->result_size = 32;
-		query->num_cs_dw = 6;
-		break;
-	default:
-		assert(0);
-		FREE(query);
-		return NULL;
-	}
-
-	/* adjust buffer size to simplify offsets wrapping math */
-	buffer_size -= buffer_size % query->result_size;
-
-	/* Queries are normally read by the CPU after
-	 * being written by the gpu, hence staging is probably a good
-	 * usage pattern.
-	 */
-	query->buffer = si_resource_create_custom(&ctx->screen->b.b,
-						  PIPE_USAGE_STAGING,
-						  buffer_size);
-	if (!query->buffer) {
-		FREE(query);
-		return NULL;
-	}
-	return query;
-}
-
-void si_context_query_destroy(struct si_context *ctx, struct si_query *query)
-{
-	r600_resource_reference(&query->buffer, NULL);
-	free(query);
-}
-
-boolean si_context_query_result(struct si_context *ctx,
-				struct si_query *query,
-				boolean wait, void *vresult)
-{
-	boolean *result_b = (boolean*)vresult;
-	uint64_t *result_u64 = (uint64_t*)vresult;
-	struct pipe_query_data_so_statistics *result_so =
-		(struct pipe_query_data_so_statistics*)vresult;
-
-	if (!si_query_result(ctx, query, wait))
-		return FALSE;
-
-	switch (query->type) {
-	case PIPE_QUERY_OCCLUSION_COUNTER:
-	case PIPE_QUERY_PRIMITIVES_EMITTED:
-	case PIPE_QUERY_PRIMITIVES_GENERATED:
-		*result_u64 = query->result.u64;
-		break;
-	case PIPE_QUERY_OCCLUSION_PREDICATE:
-	case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
-		*result_b = query->result.b;
-		break;
-	case PIPE_QUERY_TIMESTAMP:
-	case PIPE_QUERY_TIME_ELAPSED:
-		*result_u64 = (1000000 * query->result.u64) / ctx->screen->b.info.r600_clock_crystal_freq;
-		break;
-	case PIPE_QUERY_SO_STATISTICS:
-		*result_so = query->result.so;
-		break;
-	default:
-		assert(0);
-	}
-	return TRUE;
-}
-
-void si_context_queries_suspend(struct si_context *ctx)
-{
-	struct si_query *query;
-
-	LIST_FOR_EACH_ENTRY(query, &ctx->active_nontimer_query_list, list) {
-		si_query_end(ctx, query);
-	}
-	assert(ctx->num_cs_dw_nontimer_queries_suspend == 0);
-}
-
-void si_context_queries_resume(struct si_context *ctx)
-{
-	struct si_query *query;
-
-	assert(ctx->num_cs_dw_nontimer_queries_suspend == 0);
-
-	LIST_FOR_EACH_ENTRY(query, &ctx->active_nontimer_query_list, list) {
-		si_query_begin(ctx, query);
-	}
-}
-
 #if SI_TRACE_CS
 void si_trace_emit(struct si_context *sctx)
 {
diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c
index 662c5211f36..fee6251ee2f 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.c
+++ b/src/gallium/drivers/radeonsi/si_pipe.c
@@ -67,10 +67,10 @@ void si_flush(struct pipe_context *ctx, struct pipe_fence_handle **fence,
 	}
 
 	/* Disable render condition. */
-	if (sctx->current_render_cond) {
-		render_cond = sctx->current_render_cond;
-		render_cond_cond = sctx->current_render_cond_cond;
-		render_cond_mode = sctx->current_render_cond_mode;
+	if (sctx->b.current_render_cond) {
+		render_cond = sctx->b.current_render_cond;
+		render_cond_cond = sctx->b.current_render_cond_cond;
+		render_cond_mode = sctx->b.current_render_cond_mode;
 		ctx->render_condition(ctx, NULL, FALSE, 0);
 	}
 
@@ -142,7 +142,6 @@ static struct pipe_context *si_create_context(struct pipe_screen *screen, void *
 		goto fail;
 
 	si_init_blit_functions(sctx);
-	si_init_query_functions(sctx);
 	si_init_context_resource_functions(sctx);
 	si_init_compute_functions(sctx);
 
@@ -169,8 +168,6 @@ static struct pipe_context *si_create_context(struct pipe_screen *screen, void *
 	case SI:
 	case CIK:
 		si_init_state_functions(sctx);
-		LIST_INITHEAD(&sctx->active_nontimer_query_list);
-		sctx->max_db = 8;
 		si_init_config(sctx);
 		break;
 	default:
@@ -192,7 +189,7 @@ static struct pipe_context *si_create_context(struct pipe_screen *screen, void *
 
 	/* these must be last */
 	si_begin_new_cs(sctx);
-	si_get_backend_mask(sctx);
+	r600_query_init_backend_mask(&sctx->b); /* this emits commands and must be last */
 
 	/* CIK cannot unbind a constant buffer (S_BUFFER_LOAD is buggy
 	 * with a NULL buffer). We need to use a dummy buffer instead. */
diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index 56b704bfa5d..ecc66993230 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -127,12 +127,6 @@ struct si_context {
 	struct si_pipe_shader_selector	*ps_shader;
 	struct si_pipe_shader_selector	*vs_shader;
 	struct si_cs_shader_state	cs_shader_state;
-	struct pipe_query		*current_render_cond;
-	unsigned			current_render_cond_mode;
-	boolean				current_render_cond_cond;
-	struct pipe_query		*saved_render_cond;
-	unsigned			saved_render_cond_mode;
-	boolean				saved_render_cond_cond;
 	/* shader information */
 	unsigned			sprite_coord_enable;
 	unsigned			export_16bpc;
@@ -148,16 +142,6 @@ struct si_context {
 	 */
 	unsigned		pm4_dirty_cdwords;
 
-	/* The list of active queries. Only one query of each type can be active. */
-	struct list_head	active_nontimer_query_list;
-	unsigned		num_cs_dw_nontimer_queries_suspend;
-	/* If queries have been suspended. */
-	bool			nontimer_queries_suspended;
-
-	unsigned		backend_mask;
-	unsigned                max_db; /* for OQ */
-	boolean                 predicate_drawing;
-
 	/* Vertex and index buffers. */
 	bool			vertex_buffers_dirty;
 	struct pipe_index_buffer index_buffer;
@@ -192,9 +176,6 @@ void si_flush(struct pipe_context *ctx, struct pipe_fence_handle **fence,
 	       unsigned flags);
 const char *si_get_llvm_processor_name(enum radeon_family family);
 
-/* si_query.c */
-void si_init_query_functions(struct si_context *sctx);
-
 /* si_resource.c */
 void si_init_context_resource_functions(struct si_context *sctx);
 
diff --git a/src/gallium/drivers/radeonsi/si_query.c b/src/gallium/drivers/radeonsi/si_query.c
deleted file mode 100644
index d694eca8d92..00000000000
--- a/src/gallium/drivers/radeonsi/si_query.c
+++ /dev/null
@@ -1,147 +0,0 @@
-/*
- * Copyright 2010 Jerome Glisse <[email protected]>
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * on the rights to use, copy, modify, merge, publish, distribute, sub
- * license, and/or sell copies of the Software, and to permit persons to whom
- * the Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice (including the next
- * paragraph) shall be included in all copies or substantial portions of the
- * Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
- * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
- * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
- * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
- * USE OR OTHER DEALINGS IN THE SOFTWARE.
- */
-#include "si_pipe.h"
-#include "sid.h"
-
-static struct pipe_query *si_create_query(struct pipe_context *ctx, unsigned query_type)
-{
-	struct si_context *sctx = (struct si_context *)ctx;
-
-	return (struct pipe_query*)si_context_query_create(sctx, query_type);
-}
-
-static void si_destroy_query(struct pipe_context *ctx, struct pipe_query *query)
-{
-	struct si_context *sctx = (struct si_context *)ctx;
-
-	si_context_query_destroy(sctx, (struct si_query *)query);
-}
-
-static void si_begin_query(struct pipe_context *ctx, struct pipe_query *query)
-{
-	struct si_context *sctx = (struct si_context *)ctx;
-	struct si_query *rquery = (struct si_query *)query;
-
-	if (!si_query_needs_begin(rquery->type)) {
-		assert(0);
-		return;
-	}
-
-	memset(&rquery->result, 0, sizeof(rquery->result));
-	rquery->results_start = rquery->results_end;
-	si_query_begin(sctx, (struct si_query *)query);
-
-	if (!si_is_timer_query(rquery->type)) {
-		LIST_ADDTAIL(&rquery->list, &sctx->active_nontimer_query_list);
-	}
-}
-
-static void si_end_query(struct pipe_context *ctx, struct pipe_query *query)
-{
-	struct si_context *sctx = (struct si_context *)ctx;
-	struct si_query *rquery = (struct si_query *)query;
-
-	if (!si_query_needs_begin(rquery->type)) {
-		memset(&rquery->result, 0, sizeof(rquery->result));
-	}
-
-	si_query_end(sctx, rquery);
-
-	if (si_query_needs_begin(rquery->type) && !si_is_timer_query(rquery->type)) {
-		LIST_DELINIT(&rquery->list);
-	}
-}
-
-static boolean si_get_query_result(struct pipe_context *ctx,
-				   struct pipe_query *query,
-				   boolean wait, union pipe_query_result *vresult)
-{
-	struct si_context *sctx = (struct si_context *)ctx;
-	struct si_query *rquery = (struct si_query *)query;
-
-	return si_context_query_result(sctx, rquery, wait, vresult);
-}
-
-static void si_render_condition(struct pipe_context *ctx,
-				struct pipe_query *query,
-				boolean condition,
-				uint mode)
-{
-	struct si_context *sctx = (struct si_context *)ctx;
-	struct si_query *rquery = (struct si_query *)query;
-	int wait_flag = 0;
-
-	/* If we already have nonzero result, render unconditionally */
-	if (query != NULL && rquery->result.u64 != 0) {
-		if (sctx->current_render_cond) {
-			si_render_condition(ctx, NULL, FALSE, 0);
-		}
-		return;
-	}
-
-	sctx->current_render_cond = query;
-	sctx->current_render_cond_cond = condition;
-	sctx->current_render_cond_mode = mode;
-
-	if (query == NULL) {
-		if (sctx->predicate_drawing) {
-			sctx->predicate_drawing = false;
-			si_query_predication(sctx, NULL, PREDICATION_OP_CLEAR, 1);
-		}
-		return;
-	}
-
-	if (mode == PIPE_RENDER_COND_WAIT ||
-	    mode == PIPE_RENDER_COND_BY_REGION_WAIT) {
-		wait_flag = 1;
-	}
-
-	sctx->predicate_drawing = true;
-
-	switch (rquery->type) {
-	case PIPE_QUERY_OCCLUSION_COUNTER:
-	case PIPE_QUERY_OCCLUSION_PREDICATE:
-		si_query_predication(sctx, rquery, PREDICATION_OP_ZPASS, wait_flag);
-		break;
-	case PIPE_QUERY_PRIMITIVES_EMITTED:
-	case PIPE_QUERY_PRIMITIVES_GENERATED:
-	case PIPE_QUERY_SO_STATISTICS:
-	case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
-		si_query_predication(sctx, rquery, PREDICATION_OP_PRIMCOUNT, wait_flag);
-		break;
-	default:
-		assert(0);
-	}
-}
-
-void si_init_query_functions(struct si_context *sctx)
-{
-	sctx->b.b.create_query = si_create_query;
-	sctx->b.b.destroy_query = si_destroy_query;
-	sctx->b.b.begin_query = si_begin_query;
-	sctx->b.b.end_query = si_end_query;
-	sctx->b.b.get_query_result = si_get_query_result;
-
-	if (sctx->screen->b.info.r600_num_backends > 0)
-	    sctx->b.b.render_condition = si_render_condition;
-}
diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c
index a269335b6b9..8396ef51210 100644
--- a/src/gallium/drivers/radeonsi/si_state.c
+++ b/src/gallium/drivers/radeonsi/si_state.c
@@ -3028,6 +3028,19 @@ static boolean si_dma_copy(struct pipe_context *ctx,
 	return FALSE;
 }
 
+static void si_set_occlusion_query_state(struct pipe_context *ctx, bool enable)
+{
+	/* XXX Turn this into a proper state. Right now the queries are
+	 * enabled in draw_vbo, which snoops r600_common_context to see
+	 * if any occlusion queries are active. */
+}
+
+static void si_need_gfx_cs_space(struct pipe_context *ctx, unsigned num_dw,
+				 bool include_draw_vbo)
+{
+	si_need_cs_space((struct si_context*)ctx, num_dw, include_draw_vbo);
+}
+
 void si_init_state_functions(struct si_context *sctx)
 {
 	int i;
@@ -3090,6 +3103,8 @@ void si_init_state_functions(struct si_context *sctx)
 	sctx->b.b.create_surface = r600_create_surface;
 	sctx->b.b.surface_destroy = r600_surface_destroy;
 	sctx->b.dma_copy = si_dma_copy;
+	sctx->b.set_occlusion_query_state = si_set_occlusion_query_state;
+	sctx->b.need_gfx_cs_space = si_need_gfx_cs_space;
 
 	sctx->b.b.draw_vbo = si_draw_vbo;
 }
diff --git a/src/gallium/drivers/radeonsi/si_state_draw.c b/src/gallium/drivers/radeonsi/si_state_draw.c
index 4d347ed0774..90f9703cfd3 100644
--- a/src/gallium/drivers/radeonsi/si_state_draw.c
+++ b/src/gallium/drivers/radeonsi/si_state_draw.c
@@ -545,7 +545,7 @@ static void si_state_draw(struct si_context *sctx,
 
 	/* queries need some special values
 	 * (this is non-zero if any query is active) */
-	if (sctx->num_cs_dw_nontimer_queries_suspend) {
+	if (sctx->b.num_occlusion_queries > 0) {
 		if (sctx->b.chip_class >= CIK) {
 			si_pm4_set_reg(pm4, R_028004_DB_COUNT_CONTROL,
 				       S_028004_PERFECT_ZPASS_COUNTS(1) |
@@ -592,11 +592,11 @@ static void si_state_draw(struct si_context *sctx,
 		si_pm4_cmd_add(pm4, V_028A7C_VGT_INDEX_16 | (SI_BIG_ENDIAN ?
 				V_028A7C_VGT_DMA_SWAP_16_BIT : 0));
 	}
-	si_pm4_cmd_end(pm4, sctx->predicate_drawing);
+	si_pm4_cmd_end(pm4, sctx->b.predicate_drawing);
 
 	si_pm4_cmd_begin(pm4, PKT3_NUM_INSTANCES);
 	si_pm4_cmd_add(pm4, info->instance_count);
-	si_pm4_cmd_end(pm4, sctx->predicate_drawing);
+	si_pm4_cmd_end(pm4, sctx->b.predicate_drawing);
 
 	if (info->indexed) {
 		uint32_t max_size = (ib->buffer->width0 - ib->offset) /
@@ -608,11 +608,11 @@ static void si_state_draw(struct si_context *sctx,
 		si_pm4_add_bo(pm4, (struct r600_resource *)ib->buffer, RADEON_USAGE_READ);
 		si_cmd_draw_index_2(pm4, max_size, va, info->count,
 				    V_0287F0_DI_SRC_SEL_DMA,
-				    sctx->predicate_drawing);
+				    sctx->b.predicate_drawing);
 	} else {
 		uint32_t initiator = V_0287F0_DI_SRC_SEL_AUTO_INDEX;
 		initiator |= S_0287F0_USE_OPAQUE(!!info->count_from_stream_output);
-		si_cmd_draw_index_auto(pm4, info->count, initiator, sctx->predicate_drawing);
+		si_cmd_draw_index_auto(pm4, info->count, initiator, sctx->b.predicate_drawing);
 	}
 
 	si_pm4_set_state(sctx, draw, pm4);