r600g: modify block to only emit the first few dirty registers. (v2)

This gets me from 2200 to 1978 dwords for a gears frame.

This is due to us having some 32-dwords blocks in the SPI, that we only
modify the first dwords off.

v2: fix dirty reg count from Bas Nieuwenhuizen

Signed-off-by: Dave Airlie <airlied@redhat.com>

4 files changed, 39 insertions, 12 deletions

diff --git a/src/gallium/drivers/r600/r600.h b/src/gallium/drivers/r600/r600.h
index d6050002033..41666f2bff2 100644
--- a/src/gallium/drivers/r600/r600.h
+++ b/src/gallium/drivers/r600/r600.h
@@ -184,7 +184,8 @@ struct r600_block {
 	unsigned		pm4_ndwords;
 	unsigned		pm4_flush_ndwords;
 	unsigned		nbo;
-	unsigned		nreg;
+	u16 		        nreg;
+	u16                     nreg_dirty;
 	u32			*reg;
 	u32			pm4[R600_BLOCK_MAX_REG];
 	unsigned		pm4_bo_index[R600_BLOCK_MAX_REG];
diff --git a/src/gallium/winsys/r600/drm/evergreen_hw_context.c b/src/gallium/winsys/r600/drm/evergreen_hw_context.c
index b287ed54c29..670cbf7804e 100644
--- a/src/gallium/winsys/r600/drm/evergreen_hw_context.c
+++ b/src/gallium/winsys/r600/drm/evergreen_hw_context.c
@@ -664,7 +664,7 @@ static inline void evergreen_context_pipe_state_set_resource(struct r600_context
 		r600_bo_reference(ctx->radeon, &block->reloc[1].bo, state->regs[2].bo);
 		r600_bo_reference(ctx->radeon, &block->reloc[2].bo, state->regs[3].bo);
 	}
-	r600_context_dirty_block(ctx, block, R600_BLOCK_STATUS_DIRTY);
+	r600_context_dirty_block(ctx, block, R600_BLOCK_STATUS_DIRTY, 7);
 }
 
 void evergreen_context_pipe_state_set_ps_resource(struct r600_context *ctx, struct r600_pipe_state *state, unsigned rid)
@@ -711,7 +711,7 @@ static inline void evergreen_context_pipe_state_set_sampler(struct r600_context
 		}
 	}
 
-	r600_context_dirty_block(ctx, block, dirty);
+	r600_context_dirty_block(ctx, block, dirty, 2);
 }
 
 static inline void evergreen_context_pipe_state_set_sampler_border(struct r600_context *ctx, struct r600_pipe_state *state, unsigned offset, unsigned id)
@@ -745,7 +745,7 @@ static inline void evergreen_context_pipe_state_set_sampler_border(struct r600_c
 		}
 	}
 
-	r600_context_dirty_block(ctx, block, dirty);
+	r600_context_dirty_block(ctx, block, dirty, 4);
 }
 
 void evergreen_context_pipe_state_set_ps_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id)
diff --git a/src/gallium/winsys/r600/drm/r600_hw_context.c b/src/gallium/winsys/r600/drm/r600_hw_context.c
index 2d9e40303e0..52abf061cce 100644
--- a/src/gallium/winsys/r600/drm/r600_hw_context.c
+++ b/src/gallium/winsys/r600/drm/r600_hw_context.c
@@ -113,6 +113,7 @@ int r600_context_add_block(struct r600_context *ctx, const struct r600_reg *reg,
 		block->reg = &block->pm4[block->pm4_ndwords];
 		block->pm4_ndwords += n;
 		block->nreg = n;
+		block->nreg_dirty = n;
 		block->flags = 0;
 		LIST_INITHEAD(&block->list);
 
@@ -844,7 +845,7 @@ void r600_context_pipe_state_set(struct r600_context *ctx, struct r600_pipe_stat
 	unsigned new_val;
 	int dirty;
 	for (int i = 0; i < state->nregs; i++) {
-		unsigned id;
+		unsigned id, reloc_id;
 
 		range = &ctx->range[CTX_RANGE_ID(ctx, state->regs[i].offset)];
 		block = range->blocks[CTX_BLOCK_ID(ctx, state->regs[i].offset)];
@@ -863,14 +864,14 @@ void r600_context_pipe_state_set(struct r600_context *ctx, struct r600_pipe_stat
 			dirty |= R600_BLOCK_STATUS_DIRTY;
 		if (block->pm4_bo_index[id]) {
 			/* find relocation */
-			id = block->pm4_bo_index[id];
-			r600_bo_reference(ctx->radeon, &block->reloc[id].bo, state->regs[i].bo);
+			reloc_id = block->pm4_bo_index[id];
+			r600_bo_reference(ctx->radeon, &block->reloc[reloc_id].bo, state->regs[i].bo);
 			state->regs[i].bo->fence = ctx->radeon->fence;
 			/* always force dirty for relocs for now */
 			dirty |= R600_BLOCK_STATUS_DIRTY;
 		}
 
-		r600_context_dirty_block(ctx, block, dirty);
+		r600_context_dirty_block(ctx, block, dirty, id);
 	}
 }
 
@@ -909,7 +910,7 @@ static inline void r600_context_pipe_state_set_resource(struct r600_context *ctx
 		state->regs[2].bo->fence = ctx->radeon->fence;
 		state->regs[3].bo->fence = ctx->radeon->fence;
 	}
-	r600_context_dirty_block(ctx, block, R600_BLOCK_STATUS_DIRTY);
+	r600_context_dirty_block(ctx, block, R600_BLOCK_STATUS_DIRTY, 6);
 }
 
 void r600_context_pipe_state_set_ps_resource(struct r600_context *ctx, struct r600_pipe_state *state, unsigned rid)
@@ -955,7 +956,7 @@ static inline void r600_context_pipe_state_set_sampler(struct r600_context *ctx,
 		}
 	}
 
-	r600_context_dirty_block(ctx, block, dirty);
+	r600_context_dirty_block(ctx, block, dirty, 2);
 }
 
 static inline void r600_context_pipe_state_set_sampler_border(struct r600_context *ctx, struct r600_pipe_state *state, unsigned offset)
@@ -983,7 +984,7 @@ static inline void r600_context_pipe_state_set_sampler_border(struct r600_contex
 		}
 	}
 
-	r600_context_dirty_block(ctx, block, dirty);
+	r600_context_dirty_block(ctx, block, dirty, 3);
 }
 
 void r600_context_pipe_state_set_ps_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id)
@@ -1211,6 +1212,7 @@ void r600_context_flush(struct r600_context *ctx)
 			}
 			ctx->pm4_dirty_cdwords += ctx->blocks[i]->pm4_ndwords + ctx->blocks[i]->pm4_flush_ndwords;
 			ctx->blocks[i]->status |= R600_BLOCK_STATUS_DIRTY;
+			ctx->blocks[i]->nreg_dirty = ctx->blocks[i]->nreg;
 		}
 	}
 }
diff --git a/src/gallium/winsys/r600/drm/r600_priv.h b/src/gallium/winsys/r600/drm/r600_priv.h
index 534df11f185..faf47e82ab8 100644
--- a/src/gallium/winsys/r600/drm/r600_priv.h
+++ b/src/gallium/winsys/r600/drm/r600_priv.h
@@ -36,6 +36,9 @@
 #include <os/os_thread.h>
 #include "r600.h"
 
+#define PKT_COUNT_C                     0xC000FFFF
+#define PKT_COUNT_S(x)                  (((x) & 0x3FFF) << 16)
+
 struct r600_bomgr;
 struct r600_bo;
 
@@ -198,9 +201,13 @@ static void inline r600_context_reg(struct r600_context *ctx,
 }
 
 static inline void r600_context_dirty_block(struct r600_context *ctx, struct r600_block *block,
-					    int dirty)
+					    int dirty, int index)
 {
+	if (dirty && (index + 1) > block->nreg_dirty)
+		block->nreg_dirty = index + 1;
+
 	if ((dirty != (block->status & R600_BLOCK_STATUS_DIRTY)) || !(block->status & R600_BLOCK_STATUS_ENABLED)) {
+
 		block->status |= R600_BLOCK_STATUS_ENABLED;
 		block->status |= R600_BLOCK_STATUS_DIRTY;
 		ctx->pm4_dirty_cdwords += block->pm4_ndwords + block->pm4_flush_ndwords;
@@ -212,6 +219,10 @@ static inline void r600_context_block_emit_dirty(struct r600_context *ctx, struc
 {
 	int id;
 
+	if (block->nreg_dirty == 0 && block->nbo == 0 && !(block->flags & REG_FLAG_DIRTY_ALWAYS)) {
+		goto out;
+	}
+
 	for (int j = 0; j < block->nreg; j++) {
 		if (block->pm4_bo_index[j]) {
 			/* find relocation */
@@ -227,7 +238,20 @@ static inline void r600_context_block_emit_dirty(struct r600_context *ctx, struc
 	}
 	memcpy(&ctx->pm4[ctx->pm4_cdwords], block->pm4, block->pm4_ndwords * 4);
 	ctx->pm4_cdwords += block->pm4_ndwords;
+
+	if (block->nreg_dirty != block->nreg && block->nbo == 0 && !(block->flags & REG_FLAG_DIRTY_ALWAYS)) {
+		int new_dwords = block->nreg_dirty;
+		uint32_t oldword, newword;
+		ctx->pm4_cdwords -= block->pm4_ndwords;
+		newword = oldword = ctx->pm4[ctx->pm4_cdwords];
+		newword &= PKT_COUNT_C;
+		newword |= PKT_COUNT_S(new_dwords);
+		ctx->pm4[ctx->pm4_cdwords] = newword;
+		ctx->pm4_cdwords += new_dwords + 2;
+	}
+out:
 	block->status ^= R600_BLOCK_STATUS_DIRTY;
+	block->nreg_dirty = 0;
 	LIST_DELINIT(&block->list);
 }