From: Marek Olšák Date: Wed, 9 Nov 2011 00:42:13 +0000 (+0100) Subject: r600g: consolidate checking whether CS is full X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=11bdd28796e303eda70f1d73bd2469597bd561bc;p=mesa.git r600g: consolidate checking whether CS is full This adds a new function r600_need_cs_space. Currently, it's easy to overflow the CS - queries are not counted in. I guess that's not the only case where the driver may crap out. --- diff --git a/src/gallium/drivers/r600/evergreen_hw_context.c b/src/gallium/drivers/r600/evergreen_hw_context.c index f7ed6ac290d..f5f20a12c76 100644 --- a/src/gallium/drivers/r600/evergreen_hw_context.c +++ b/src/gallium/drivers/r600/evergreen_hw_context.c @@ -1159,12 +1159,10 @@ void evergreen_context_draw(struct r600_context *ctx, const struct r600_draw *dr * reserved for flushing the destination caches */ ctx->pm4_ndwords = RADEON_MAX_CMDBUF_DWORDS - ctx->num_dest_buffers * 7 - 16; - if ((ctx->pm4_dirty_cdwords + ndwords + ctx->pm4_cdwords) > ctx->pm4_ndwords) { - /* need to flush */ - r600_context_flush(ctx, RADEON_FLUSH_ASYNC); - } - /* at that point everythings is flushed and ctx->pm4_cdwords = 0 */ - if ((ctx->pm4_dirty_cdwords + ndwords) > ctx->pm4_ndwords) { + r600_need_cs_space(ctx, ndwords); + + /* at this point everything is flushed and ctx->pm4_cdwords = 0 */ + if (unlikely((ctx->pm4_dirty_cdwords + ndwords) > ctx->pm4_ndwords)) { R600_ERR("context is too big to be scheduled\n"); return; } diff --git a/src/gallium/drivers/r600/r600_hw_context.c b/src/gallium/drivers/r600/r600_hw_context.c index c04af902752..72d80b2da2a 100644 --- a/src/gallium/drivers/r600/r600_hw_context.c +++ b/src/gallium/drivers/r600/r600_hw_context.c @@ -929,15 +929,24 @@ out_err: return r; } -/* Flushes all surfaces */ -void r600_context_flush_all(struct r600_context *ctx, unsigned flush_flags) +void r600_need_cs_space(struct r600_context *ctx, unsigned num_dw) { - unsigned ndwords = 5; + /* The number of dwords we already used in the CS so far. */ + num_dw += ctx->pm4_cdwords; + + /* The number of dwords all the dirty states would take. */ + num_dw += ctx->pm4_dirty_cdwords; - if ((ctx->pm4_dirty_cdwords + ndwords + ctx->pm4_cdwords) > ctx->pm4_ndwords) { - /* need to flush */ + /* Flush if there's not enough space. */ + if (num_dw > ctx->pm4_ndwords) { r600_context_flush(ctx, RADEON_FLUSH_ASYNC); } +} + +/* Flushes all surfaces */ +void r600_context_flush_all(struct r600_context *ctx, unsigned flush_flags) +{ + r600_need_cs_space(ctx, 5); ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SURFACE_SYNC, 3, ctx->predicate_drawing); ctx->pm4[ctx->pm4_cdwords++] = flush_flags; /* CP_COHER_CNTL */ @@ -1434,15 +1443,14 @@ void r600_context_draw(struct r600_context *ctx, const struct r600_draw *draw) * reserved for flushing the destination caches */ ctx->pm4_ndwords = RADEON_MAX_CMDBUF_DWORDS - ctx->num_dest_buffers * 7 - 16; - if ((ctx->pm4_dirty_cdwords + ndwords + ctx->pm4_cdwords) > ctx->pm4_ndwords) { - /* need to flush */ - r600_context_flush(ctx, RADEON_FLUSH_ASYNC); - } - /* at that point everythings is flushed and ctx->pm4_cdwords = 0 */ - if ((ctx->pm4_dirty_cdwords + ndwords) > ctx->pm4_ndwords) { + r600_need_cs_space(ctx, ndwords); + + /* at this point everything is flushed and ctx->pm4_cdwords = 0 */ + if (unlikely((ctx->pm4_dirty_cdwords + ndwords) > ctx->pm4_ndwords)) { R600_ERR("context is too big to be scheduled\n"); return; } + /* enough room to copy packet */ LIST_FOR_EACH_ENTRY_SAFE(dirty_block, next_block, &ctx->dirty, list) { r600_context_block_emit_dirty(ctx, dirty_block); @@ -1545,12 +1553,7 @@ void r600_context_flush(struct r600_context *ctx, unsigned flags) void r600_context_emit_fence(struct r600_context *ctx, struct r600_resource *fence_bo, unsigned offset, unsigned value) { - unsigned ndwords = 10; - - if ((ctx->pm4_dirty_cdwords + ndwords + ctx->pm4_cdwords) > ctx->pm4_ndwords) { - /* need to flush */ - r600_context_flush(ctx, RADEON_FLUSH_ASYNC); - } + r600_need_cs_space(ctx, 10); ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0, 0); ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4); @@ -1634,10 +1637,7 @@ void r600_query_begin(struct r600_context *ctx, struct r600_query *query) return; } - if ((required_space + ctx->pm4_cdwords) > ctx->pm4_ndwords) { - /* need to flush */ - r600_context_flush(ctx, RADEON_FLUSH_ASYNC); - } + r600_need_cs_space(ctx, required_space); new_results_end = (query->results_end + query->result_size) % query->buffer->b.b.b.width0; @@ -1723,8 +1723,7 @@ void r600_query_predication(struct r600_context *ctx, struct r600_query *query, int flag_wait) { if (operation == PREDICATION_OP_CLEAR) { - if (ctx->pm4_cdwords + 3 > ctx->pm4_ndwords) - r600_context_flush(ctx, RADEON_FLUSH_ASYNC); + r600_need_cs_space(ctx, 3); ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_PREDICATION, 1, 0); ctx->pm4[ctx->pm4_cdwords++] = 0; @@ -1738,8 +1737,7 @@ void r600_query_predication(struct r600_context *ctx, struct r600_query *query, count = (query->buffer->b.b.b.width0 + query->results_end - query->results_start) % query->buffer->b.b.b.width0; count /= query->result_size; - if (ctx->pm4_cdwords + 5 * count > ctx->pm4_ndwords) - r600_context_flush(ctx, RADEON_FLUSH_ASYNC); + r600_need_cs_space(ctx, 5 * count); op = PRED_OP(operation) | PREDICATION_DRAW_VISIBLE | (flag_wait ? PREDICATION_HINT_WAIT : PREDICATION_HINT_NOWAIT_DRAW); diff --git a/src/gallium/drivers/r600/r600_hw_context_priv.h b/src/gallium/drivers/r600/r600_hw_context_priv.h index 0cbd94bb4c8..182f68d98e6 100644 --- a/src/gallium/drivers/r600/r600_hw_context_priv.h +++ b/src/gallium/drivers/r600/r600_hw_context_priv.h @@ -54,6 +54,7 @@ struct r600_reg { /* * r600_hw_context.c */ +void r600_need_cs_space(struct r600_context *ctx, unsigned num_dw); void r600_context_bo_flush(struct r600_context *ctx, unsigned flush_flags, unsigned flush_mask, struct r600_resource *rbo); struct r600_resource *r600_context_reg_bo(struct r600_context *ctx, unsigned offset);