radeonsi: emit_guardband packets optimization

Signed-off-by: Sonny Jiang <[email protected]>
Signed-off-by: Marek Olšák <[email protected]>

4 files changed, 50 insertions, 8 deletions

diff --git a/src/gallium/drivers/radeonsi/si_build_pm4.h b/src/gallium/drivers/radeonsi/si_build_pm4.h
index 0c92b1a35c0..b339cd57ce7 100644
--- a/src/gallium/drivers/radeonsi/si_build_pm4.h
+++ b/src/gallium/drivers/radeonsi/si_build_pm4.h
@@ -181,4 +181,37 @@ static inline void radeon_opt_set_context_reg3(struct si_context *sctx, unsigned
 	}
 }
 
+/**
+ * Set 4 consecutive registers if any registers value is different.
+ */
+static inline void radeon_opt_set_context_reg4(struct si_context *sctx, unsigned offset,
+					       enum si_tracked_reg reg, unsigned value1,
+					       unsigned value2, unsigned value3,
+					       unsigned value4)
+{
+	struct radeon_cmdbuf *cs = sctx->gfx_cs;
+
+	if (!(sctx->tracked_regs.reg_saved & (1 << reg)) ||
+	    !(sctx->tracked_regs.reg_saved & (1 << (reg + 1))) ||
+	    !(sctx->tracked_regs.reg_saved & (1 << (reg + 2))) ||
+	    !(sctx->tracked_regs.reg_saved & (1 << (reg + 3))) ||
+	    sctx->tracked_regs.reg_value[reg] != value1 ||
+	    sctx->tracked_regs.reg_value[reg+1] != value2 ||
+	    sctx->tracked_regs.reg_value[reg+2] != value3 ||
+	    sctx->tracked_regs.reg_value[reg+3] != value4 ) {
+
+		radeon_set_context_reg_seq(cs, offset, 4);
+		radeon_emit(cs, value1);
+		radeon_emit(cs, value2);
+		radeon_emit(cs, value3);
+		radeon_emit(cs, value4);
+
+		sctx->tracked_regs.reg_value[reg] = value1;
+		sctx->tracked_regs.reg_value[reg+1] = value2;
+		sctx->tracked_regs.reg_value[reg+2] = value3;
+		sctx->tracked_regs.reg_value[reg+3] = value4;
+		sctx->tracked_regs.reg_saved |= 0xf << reg;
+	}
+}
+
 #endif
diff --git a/src/gallium/drivers/radeonsi/si_gfx_cs.c b/src/gallium/drivers/radeonsi/si_gfx_cs.c
index ac4909a847a..628b6c50e46 100644
--- a/src/gallium/drivers/radeonsi/si_gfx_cs.c
+++ b/src/gallium/drivers/radeonsi/si_gfx_cs.c
@@ -342,6 +342,10 @@ void si_begin_new_gfx_cs(struct si_context *ctx)
 		ctx->tracked_regs.reg_value[SI_TRACKED_PA_CL_CLIP_CNTL]	= 0x00090000;
 		ctx->tracked_regs.reg_value[SI_TRACKED_PA_SC_BINNER_CNTL_0] = 0x00000003;
 		ctx->tracked_regs.reg_value[SI_TRACKED_DB_DFSM_CONTROL]	= 0x00000000;
+		ctx->tracked_regs.reg_value[SI_TRACKED_PA_CL_GB_VERT_CLIP_ADJ]	= 0x3f800000;
+		ctx->tracked_regs.reg_value[SI_TRACKED_PA_CL_GB_VERT_DISC_ADJ]	= 0x3f800000;
+		ctx->tracked_regs.reg_value[SI_TRACKED_PA_CL_GB_HORZ_CLIP_ADJ]	= 0x3f800000;
+		ctx->tracked_regs.reg_value[SI_TRACKED_PA_CL_GB_HORZ_DISC_ADJ]	= 0x3f800000;
 
 		/* Set all saved registers state to saved. */
 		ctx->tracked_regs.reg_saved = 0xffffffff;
diff --git a/src/gallium/drivers/radeonsi/si_state.h b/src/gallium/drivers/radeonsi/si_state.h
index f8748bdfffb..71056c76c38 100644
--- a/src/gallium/drivers/radeonsi/si_state.h
+++ b/src/gallium/drivers/radeonsi/si_state.h
@@ -276,6 +276,11 @@ enum si_tracked_reg {
 	SI_TRACKED_PA_SC_BINNER_CNTL_0,
 	SI_TRACKED_DB_DFSM_CONTROL,
 
+	SI_TRACKED_PA_CL_GB_VERT_CLIP_ADJ, /* 4 consecutive registers */
+	SI_TRACKED_PA_CL_GB_VERT_DISC_ADJ,
+	SI_TRACKED_PA_CL_GB_HORZ_CLIP_ADJ,
+	SI_TRACKED_PA_CL_GB_HORZ_DISC_ADJ,
+
 	SI_NUM_TRACKED_REGS,
 };
 
diff --git a/src/gallium/drivers/radeonsi/si_state_viewport.c b/src/gallium/drivers/radeonsi/si_state_viewport.c
index 3b16bdcb17f..4183be0c880 100644
--- a/src/gallium/drivers/radeonsi/si_state_viewport.c
+++ b/src/gallium/drivers/radeonsi/si_state_viewport.c
@@ -140,7 +140,6 @@ static void si_emit_guardband(struct si_context *ctx)
 {
 	const struct si_signed_scissor *vp_as_scissor;
 	struct si_signed_scissor max_vp_scissor;
-	struct radeon_cmdbuf *cs = ctx->gfx_cs;
 	struct pipe_viewport_state vp;
 	float left, top, right, bottom, max_range, guardband_x, guardband_y;
 	float discard_x, discard_y;
@@ -214,13 +213,14 @@ static void si_emit_guardband(struct si_context *ctx)
 		discard_y = MIN2(discard_y, guardband_y);
 	}
 
-	/* If any of the GB registers is updated, all of them must be updated. */
-	radeon_set_context_reg_seq(cs, R_028BE8_PA_CL_GB_VERT_CLIP_ADJ, 4);
-
-	radeon_emit(cs, fui(guardband_y)); /* R_028BE8_PA_CL_GB_VERT_CLIP_ADJ */
-	radeon_emit(cs, fui(discard_y));   /* R_028BEC_PA_CL_GB_VERT_DISC_ADJ */
-	radeon_emit(cs, fui(guardband_x)); /* R_028BF0_PA_CL_GB_HORZ_CLIP_ADJ */
-	radeon_emit(cs, fui(discard_x));   /* R_028BF4_PA_CL_GB_HORZ_DISC_ADJ */
+	/* If any of the GB registers is updated, all of them must be updated.
+	 * R_028BE8_PA_CL_GB_VERT_CLIP_ADJ, R_028BEC_PA_CL_GB_VERT_DISC_ADJ
+	 * R_028BF0_PA_CL_GB_HORZ_CLIP_ADJ, R_028BF4_PA_CL_GB_HORZ_DISC_ADJ
+	 */
+	radeon_opt_set_context_reg4(ctx, R_028BE8_PA_CL_GB_VERT_CLIP_ADJ,
+				    SI_TRACKED_PA_CL_GB_VERT_CLIP_ADJ,
+				    fui(guardband_y), fui(discard_y),
+				    fui(guardband_x), fui(discard_x));
 }
 
 static void si_emit_scissors(struct si_context *ctx)