freedreno/ir3: move binning_pass out of shader variant key

Prep work for a following patch, that introduces a cache to map from
program state (all shader stages) plus variant key to pre-baked hw
state (which could be emit'd via CP_SET_DRAW_STATE, for example).
To do that, we really want the variant key to be immutable, and to
treat the binning pass shader as an extra shader stage, rather than
as a VS variant.

Signed-off-by: Rob Clark <robdclark@gmail.com>

21 files changed, 109 insertions, 75 deletions

diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_draw.c b/src/gallium/drivers/freedreno/a3xx/fd3_draw.c
index 9f148e6179a..e6572d62457 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_draw.c
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_draw.c
@@ -83,7 +83,7 @@ draw_impl(struct fd_context *ctx, struct fd_ringbuffer *ring,
 		primtype = DI_PT_POINTLIST_PSIZE;
 
 	fd_draw_emit(ctx->batch, ring, primtype,
-			emit->key.binning_pass ? IGNORE_VISIBILITY : USE_VISIBILITY,
+			emit->binning_pass ? IGNORE_VISIBILITY : USE_VISIBILITY,
 			info, index_offset);
 }
 
@@ -158,12 +158,12 @@ fd3_draw_vbo(struct fd_context *ctx, const struct pipe_draw_info *info,
 	ctx->stats.vs_regs += ir3_shader_halfregs(vp);
 	ctx->stats.fs_regs += ir3_shader_halfregs(fp);
 
-	emit.key.binning_pass = false;
+	emit.binning_pass = false;
 	emit.dirty = dirty;
 	draw_impl(ctx, ctx->batch->draw, &emit, index_offset);
 
 	/* and now binning pass: */
-	emit.key.binning_pass = true;
+	emit.binning_pass = true;
 	emit.dirty = dirty & ~(FD_DIRTY_BLEND);
 	emit.vp = NULL;   /* we changed key so need to refetch vp */
 	emit.fp = NULL;
diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_emit.c b/src/gallium/drivers/freedreno/a3xx/fd3_emit.c
index b5e20e5a592..77b0339b59e 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_emit.c
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_emit.c
@@ -509,7 +509,7 @@ fd3_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 	}
 
 	if ((dirty & (FD_DIRTY_ZSA | FD_DIRTY_PROG | FD_DIRTY_BLEND_DUAL)) &&
-		!emit->key.binning_pass) {
+		!emit->binning_pass) {
 		uint32_t val = fd3_zsa_stateobj(ctx->zsa)->rb_render_control |
 			fd3_blend_stateobj(ctx->blend)->rb_render_control;
 
@@ -622,7 +622,7 @@ fd3_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 		uint32_t val = fd3_rasterizer_stateobj(ctx->rasterizer)
 				->pc_prim_vtx_cntl;
 
-		if (!emit->key.binning_pass) {
+		if (!emit->binning_pass) {
 			uint32_t stride_in_vpc = align(fp->total_in, 4) / 4;
 			if (stride_in_vpc > 0)
 				stride_in_vpc = MAX2(stride_in_vpc, 2);
@@ -721,7 +721,7 @@ fd3_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 
 	if (emit->prog == &ctx->prog) { /* evil hack to deal sanely with clear path */
 		ir3_emit_vs_consts(vp, ring, ctx, emit->info);
-		if (!emit->key.binning_pass)
+		if (!emit->binning_pass)
 			ir3_emit_fs_consts(fp, ring, ctx);
 	}
 
diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_emit.h b/src/gallium/drivers/freedreno/a3xx/fd3_emit.h
index c827522445f..6a6db4791e4 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_emit.h
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_emit.h
@@ -45,6 +45,7 @@ struct fd3_emit {
 	const struct fd_vertex_state *vtx;
 	const struct fd_program_stateobj *prog;
 	const struct pipe_draw_info *info;
+	bool binning_pass;
 	struct ir3_shader_key key;
 	enum fd_dirty_3d_state dirty;
 
@@ -61,7 +62,8 @@ fd3_emit_get_vp(struct fd3_emit *emit)
 {
 	if (!emit->vp) {
 		struct ir3_shader *shader = emit->prog->vp;
-		emit->vp = ir3_shader_variant(shader, emit->key, emit->debug);
+		emit->vp = ir3_shader_variant(shader, emit->key,
+				emit->binning_pass, emit->debug);
 	}
 	return emit->vp;
 }
@@ -70,13 +72,14 @@ static inline const struct ir3_shader_variant *
 fd3_emit_get_fp(struct fd3_emit *emit)
 {
 	if (!emit->fp) {
-		if (emit->key.binning_pass) {
+		if (emit->binning_pass) {
 			/* use dummy stateobj to simplify binning vs non-binning: */
 			static const struct ir3_shader_variant binning_fp = {};
 			emit->fp = &binning_fp;
 		} else {
 			struct ir3_shader *shader = emit->prog->fp;
-			emit->fp = ir3_shader_variant(shader, emit->key, emit->debug);
+			emit->fp = ir3_shader_variant(shader, emit->key,
+					false, emit->debug);
 		}
 	}
 	return emit->fp;
diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_program.c b/src/gallium/drivers/freedreno/a3xx/fd3_program.c
index 9d5c7b661fd..1a7d3062cf3 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_program.c
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_program.c
@@ -249,7 +249,7 @@ fd3_program_emit(struct fd_ringbuffer *ring, struct fd3_emit *emit,
 
 	OUT_PKT0(ring, REG_A3XX_SP_SP_CTRL_REG, 1);
 	OUT_RING(ring, A3XX_SP_SP_CTRL_REG_CONSTMODE(constmode) |
-			COND(emit->key.binning_pass, A3XX_SP_SP_CTRL_REG_BINNING) |
+			COND(emit->binning_pass, A3XX_SP_SP_CTRL_REG_BINNING) |
 			A3XX_SP_SP_CTRL_REG_SLEEPMODE(1) |
 			A3XX_SP_SP_CTRL_REG_L0MODE(0));
 
@@ -309,7 +309,7 @@ fd3_program_emit(struct fd_ringbuffer *ring, struct fd3_emit *emit,
 			A3XX_SP_VS_OBJ_OFFSET_REG_SHADEROBJOFFSET(0));
 	OUT_RELOC(ring, vp->bo, 0, 0, 0);  /* SP_VS_OBJ_START_REG */
 
-	if (emit->key.binning_pass) {
+	if (emit->binning_pass) {
 		OUT_PKT0(ring, REG_A3XX_SP_FS_LENGTH_REG, 1);
 		OUT_RING(ring, 0x00000000);
 
@@ -367,7 +367,7 @@ fd3_program_emit(struct fd_ringbuffer *ring, struct fd3_emit *emit,
 		OUT_RING(ring, mrt_reg);
 	}
 
-	if (emit->key.binning_pass) {
+	if (emit->binning_pass) {
 		OUT_PKT0(ring, REG_A3XX_VPC_ATTR, 2);
 		OUT_RING(ring, A3XX_VPC_ATTR_THRDASSIGN(1) |
 				A3XX_VPC_ATTR_LMSIZE(1) |
@@ -472,7 +472,7 @@ fd3_program_emit(struct fd_ringbuffer *ring, struct fd3_emit *emit,
 	OUT_PKT0(ring, REG_A3XX_VFD_PERFCOUNTER0_SELECT, 1);
 	OUT_RING(ring, 0x00000000);        /* VFD_PERFCOUNTER0_SELECT */
 
-	if (!emit->key.binning_pass) {
+	if (!emit->binning_pass) {
 		if (fpbuffer == BUFFER)
 			emit_shader(ring, fp);
 
diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_draw.c b/src/gallium/drivers/freedreno/a4xx/fd4_draw.c
index 029c6086f2d..d854ebc4a40 100644
--- a/src/gallium/drivers/freedreno/a4xx/fd4_draw.c
+++ b/src/gallium/drivers/freedreno/a4xx/fd4_draw.c
@@ -67,7 +67,7 @@ draw_impl(struct fd_context *ctx, struct fd_ringbuffer *ring,
 		primtype = DI_PT_POINTLIST_PSIZE;
 
 	fd4_draw_emit(ctx->batch, ring, primtype,
-			emit->key.binning_pass ? IGNORE_VISIBILITY : USE_VISIBILITY,
+			emit->binning_pass ? IGNORE_VISIBILITY : USE_VISIBILITY,
 			info, index_offset);
 }
 
@@ -144,7 +144,7 @@ fd4_draw_vbo(struct fd_context *ctx, const struct pipe_draw_info *info,
 	ctx->stats.vs_regs += ir3_shader_halfregs(vp);
 	ctx->stats.fs_regs += ir3_shader_halfregs(fp);
 
-	emit.key.binning_pass = false;
+	emit.binning_pass = false;
 	emit.dirty = dirty;
 
 	struct fd_ringbuffer *ring = ctx->batch->draw;
@@ -168,7 +168,7 @@ fd4_draw_vbo(struct fd_context *ctx, const struct pipe_draw_info *info,
 	}
 
 	/* and now binning pass: */
-	emit.key.binning_pass = true;
+	emit.binning_pass = true;
 	emit.dirty = dirty & ~(FD_DIRTY_BLEND);
 	emit.vp = NULL;   /* we changed key so need to refetch vp */
 	emit.fp = NULL;
diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_emit.c b/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
index 34f8ef1b306..49ce6353526 100644
--- a/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
+++ b/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
@@ -510,7 +510,7 @@ fd4_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 
 	emit_marker(ring, 5);
 
-	if ((dirty & FD_DIRTY_FRAMEBUFFER) && !emit->key.binning_pass) {
+	if ((dirty & FD_DIRTY_FRAMEBUFFER) && !emit->binning_pass) {
 		struct pipe_framebuffer_state *pfb = &ctx->batch->framebuffer;
 		unsigned char mrt_comp[A4XX_MAX_RENDER_TARGETS] = {0};
 
@@ -686,7 +686,7 @@ fd4_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 
 	if (emit->prog == &ctx->prog) { /* evil hack to deal sanely with clear path */
 		ir3_emit_vs_consts(vp, ring, ctx, emit->info);
-		if (!emit->key.binning_pass)
+		if (!emit->binning_pass)
 			ir3_emit_fs_consts(fp, ring, ctx);
 	}
 
diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_emit.h b/src/gallium/drivers/freedreno/a4xx/fd4_emit.h
index 376523d45b8..4d27f7076ee 100644
--- a/src/gallium/drivers/freedreno/a4xx/fd4_emit.h
+++ b/src/gallium/drivers/freedreno/a4xx/fd4_emit.h
@@ -45,6 +45,7 @@ struct fd4_emit {
 	const struct fd_vertex_state *vtx;
 	const struct fd_program_stateobj *prog;
 	const struct pipe_draw_info *info;
+	bool binning_pass;
 	struct ir3_shader_key key;
 	enum fd_dirty_3d_state dirty;
 
@@ -70,7 +71,8 @@ fd4_emit_get_vp(struct fd4_emit *emit)
 {
 	if (!emit->vp) {
 		struct ir3_shader *shader = emit->prog->vp;
-		emit->vp = ir3_shader_variant(shader, emit->key, emit->debug);
+		emit->vp = ir3_shader_variant(shader, emit->key,
+				emit->binning_pass, emit->debug);
 	}
 	return emit->vp;
 }
@@ -79,13 +81,14 @@ static inline const struct ir3_shader_variant *
 fd4_emit_get_fp(struct fd4_emit *emit)
 {
 	if (!emit->fp) {
-		if (emit->key.binning_pass) {
+		if (emit->binning_pass) {
 			/* use dummy stateobj to simplify binning vs non-binning: */
 			static const struct ir3_shader_variant binning_fp = {};
 			emit->fp = &binning_fp;
 		} else {
 			struct ir3_shader *shader = emit->prog->fp;
-			emit->fp = ir3_shader_variant(shader, emit->key, emit->debug);
+			emit->fp = ir3_shader_variant(shader, emit->key,
+					false, emit->debug);
 		}
 	}
 	return emit->fp;
diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_program.c b/src/gallium/drivers/freedreno/a4xx/fd4_program.c
index 860f615e31d..dac96de10ed 100644
--- a/src/gallium/drivers/freedreno/a4xx/fd4_program.c
+++ b/src/gallium/drivers/freedreno/a4xx/fd4_program.c
@@ -207,7 +207,7 @@ fd4_program_emit(struct fd_ringbuffer *ring, struct fd4_emit *emit,
 
 	debug_assert(nr <= ARRAY_SIZE(color_regid));
 
-	if (emit->key.binning_pass)
+	if (emit->binning_pass)
 		nr = 0;
 
 	setup_stages(emit, s);
@@ -299,7 +299,7 @@ fd4_program_emit(struct fd_ringbuffer *ring, struct fd4_emit *emit,
 
 	OUT_PKT0(ring, REG_A4XX_SP_SP_CTRL_REG, 1);
 	OUT_RING(ring, 0x140010 | /* XXX */
-			COND(emit->key.binning_pass, A4XX_SP_SP_CTRL_REG_BINNING_PASS));
+			COND(emit->binning_pass, A4XX_SP_SP_CTRL_REG_BINNING_PASS));
 
 	OUT_PKT0(ring, REG_A4XX_SP_INSTR_CACHE_CTRL, 1);
 	OUT_RING(ring, 0x7f | /* XXX */
@@ -362,7 +362,7 @@ fd4_program_emit(struct fd_ringbuffer *ring, struct fd4_emit *emit,
 			A4XX_SP_VS_OBJ_OFFSET_REG_SHADEROBJOFFSET(s[VS].instroff));
 	OUT_RELOC(ring, s[VS].v->bo, 0, 0, 0);  /* SP_VS_OBJ_START_REG */
 
-	if (emit->key.binning_pass) {
+	if (emit->binning_pass) {
 		OUT_PKT0(ring, REG_A4XX_SP_FS_LENGTH_REG, 1);
 		OUT_RING(ring, 0x00000000);         /* SP_FS_LENGTH_REG */
 
@@ -452,7 +452,7 @@ fd4_program_emit(struct fd_ringbuffer *ring, struct fd4_emit *emit,
 					A4XX_SP_FS_MRT_REG_HALF_PRECISION));
 	}
 
-	if (emit->key.binning_pass) {
+	if (emit->binning_pass) {
 		OUT_PKT0(ring, REG_A4XX_VPC_ATTR, 2);
 		OUT_RING(ring, A4XX_VPC_ATTR_THRDASSIGN(1) |
 				0x40000000 |      /* XXX */
@@ -561,7 +561,7 @@ fd4_program_emit(struct fd_ringbuffer *ring, struct fd4_emit *emit,
 	if (s[VS].instrlen)
 		emit_shader(ring, s[VS].v);
 
-	if (!emit->key.binning_pass)
+	if (!emit->binning_pass)
 		if (s[FS].instrlen)
 			emit_shader(ring, s[FS].v);
 }
diff --git a/src/gallium/drivers/freedreno/a5xx/fd5_compute.c b/src/gallium/drivers/freedreno/a5xx/fd5_compute.c
index 66ed7a4af57..1e084fd4c3b 100644
--- a/src/gallium/drivers/freedreno/a5xx/fd5_compute.c
+++ b/src/gallium/drivers/freedreno/a5xx/fd5_compute.c
@@ -188,7 +188,7 @@ fd5_launch_grid(struct fd_context *ctx, const struct pipe_grid_info *info)
 
 	emit_setup(ctx);
 
-	v = ir3_shader_variant(so->shader, key, &ctx->debug);
+	v = ir3_shader_variant(so->shader, key, false, &ctx->debug);
 	if (!v)
 		return;
 
diff --git a/src/gallium/drivers/freedreno/a5xx/fd5_draw.c b/src/gallium/drivers/freedreno/a5xx/fd5_draw.c
index 96ff1a35945..bbb12897b85 100644
--- a/src/gallium/drivers/freedreno/a5xx/fd5_draw.c
+++ b/src/gallium/drivers/freedreno/a5xx/fd5_draw.c
@@ -60,9 +60,9 @@ draw_impl(struct fd_context *ctx, struct fd_ringbuffer *ring,
 	OUT_RING(ring, info->primitive_restart ? /* PC_RESTART_INDEX */
 			info->restart_index : 0xffffffff);
 
-	fd5_emit_render_cntl(ctx, false, emit->key.binning_pass);
+	fd5_emit_render_cntl(ctx, false, emit->binning_pass);
 	fd5_draw_emit(ctx->batch, ring, primtype,
-			emit->key.binning_pass ? IGNORE_VISIBILITY : USE_VISIBILITY,
+			emit->binning_pass ? IGNORE_VISIBILITY : USE_VISIBILITY,
 			info, index_offset);
 }
 
@@ -144,13 +144,13 @@ fd5_draw_vbo(struct fd_context *ctx, const struct pipe_draw_info *info,
 	 */
 	emit.no_lrz_write = fp->writes_pos || fp->has_kill;
 
-	emit.key.binning_pass = false;
+	emit.binning_pass = false;
 	emit.dirty = dirty;
 
 	draw_impl(ctx, ctx->batch->draw, &emit, index_offset);
 
 	/* and now binning pass: */
-	emit.key.binning_pass = true;
+	emit.binning_pass = true;
 	emit.dirty = dirty & ~(FD_DIRTY_BLEND);
 	emit.vp = NULL;   /* we changed key so need to refetch vp */
 	emit.fp = NULL;
diff --git a/src/gallium/drivers/freedreno/a5xx/fd5_emit.c b/src/gallium/drivers/freedreno/a5xx/fd5_emit.c
index e3bf9e26ba4..c666b25f137 100644
--- a/src/gallium/drivers/freedreno/a5xx/fd5_emit.c
+++ b/src/gallium/drivers/freedreno/a5xx/fd5_emit.c
@@ -524,7 +524,7 @@ fd5_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 
 	emit_marker5(ring, 5);
 
-	if ((dirty & FD_DIRTY_FRAMEBUFFER) && !emit->key.binning_pass) {
+	if ((dirty & FD_DIRTY_FRAMEBUFFER) && !emit->binning_pass) {
 		unsigned char mrt_comp[A5XX_MAX_RENDER_TARGETS] = {0};
 
 		for (unsigned i = 0; i < A5XX_MAX_RENDER_TARGETS; i++) {
@@ -566,7 +566,7 @@ fd5_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 
 			if (emit->no_lrz_write || !rsc->lrz || !rsc->lrz_valid)
 				gras_lrz_cntl = 0;
-			else if (emit->key.binning_pass && blend->lrz_write && zsa->lrz_write)
+			else if (emit->binning_pass && blend->lrz_write && zsa->lrz_write)
 				gras_lrz_cntl |= A5XX_GRAS_LRZ_CNTL_LRZ_WRITE;
 
 			OUT_PKT4(ring, REG_A5XX_GRAS_LRZ_CNTL, 1);
@@ -685,7 +685,7 @@ fd5_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 		uint32_t posz_regid = ir3_find_output_regid(fp, FRAG_RESULT_DEPTH);
 		unsigned nr = pfb->nr_cbufs;
 
-		if (emit->key.binning_pass)
+		if (emit->binning_pass)
 			nr = 0;
 		else if (ctx->rasterizer->rasterizer_discard)
 			nr = 0;
@@ -701,7 +701,7 @@ fd5_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 	}
 
 	ir3_emit_vs_consts(vp, ring, ctx, emit->info);
-	if (!emit->key.binning_pass)
+	if (!emit->binning_pass)
 		ir3_emit_fs_consts(fp, ring, ctx);
 
 	struct pipe_stream_output_info *info = &vp->shader->stream_output;
diff --git a/src/gallium/drivers/freedreno/a5xx/fd5_emit.h b/src/gallium/drivers/freedreno/a5xx/fd5_emit.h
index bed52d4e87f..69ea3fa06a4 100644
--- a/src/gallium/drivers/freedreno/a5xx/fd5_emit.h
+++ b/src/gallium/drivers/freedreno/a5xx/fd5_emit.h
@@ -44,6 +44,7 @@ struct fd5_emit {
 	const struct fd_vertex_state *vtx;
 	const struct fd_program_stateobj *prog;
 	const struct pipe_draw_info *info;
+	bool binning_pass;
 	struct ir3_shader_key key;
 	enum fd_dirty_3d_state dirty;
 
@@ -77,7 +78,8 @@ fd5_emit_get_vp(struct fd5_emit *emit)
 {
 	if (!emit->vp) {
 		struct ir3_shader *shader = emit->prog->vp;
-		emit->vp = ir3_shader_variant(shader, emit->key, emit->debug);
+		emit->vp = ir3_shader_variant(shader, emit->key,
+				emit->binning_pass, emit->debug);
 	}
 	return emit->vp;
 }
@@ -86,13 +88,14 @@ static inline const struct ir3_shader_variant *
 fd5_emit_get_fp(struct fd5_emit *emit)
 {
 	if (!emit->fp) {
-		if (emit->key.binning_pass) {
+		if (emit->binning_pass) {
 			/* use dummy stateobj to simplify binning vs non-binning: */
 			static const struct ir3_shader_variant binning_fp = {};
 			emit->fp = &binning_fp;
 		} else {
 			struct ir3_shader *shader = emit->prog->fp;
-			emit->fp = ir3_shader_variant(shader, emit->key, emit->debug);
+			emit->fp = ir3_shader_variant(shader, emit->key,
+					false, emit->debug);
 		}
 	}
 	return emit->fp;
diff --git a/src/gallium/drivers/freedreno/a5xx/fd5_program.c b/src/gallium/drivers/freedreno/a5xx/fd5_program.c
index 2a6e3334aed..a30678d0477 100644
--- a/src/gallium/drivers/freedreno/a5xx/fd5_program.c
+++ b/src/gallium/drivers/freedreno/a5xx/fd5_program.c
@@ -448,7 +448,7 @@ fd5_program_emit(struct fd_context *ctx, struct fd_ringbuffer *ring,
 	ir3_link_shaders(&l, s[VS].v, s[FS].v);
 
 	if ((s[VS].v->shader->stream_output.num_outputs > 0) &&
-			!emit->key.binning_pass)
+			!emit->binning_pass)
 		link_stream_out(&l, s[VS].v);
 
 	BITSET_DECLARE(varbs, 128) = {0};
@@ -474,7 +474,7 @@ fd5_program_emit(struct fd_context *ctx, struct fd_ringbuffer *ring,
 	}
 
 	if ((s[VS].v->shader->stream_output.num_outputs > 0) &&
-			!emit->key.binning_pass) {
+			!emit->binning_pass) {
 		emit_stream_out(ring, s[VS].v, &l);
 
 		OUT_PKT4(ring, REG_A5XX_VPC_SO_OVERRIDE, 1);
@@ -534,7 +534,7 @@ fd5_program_emit(struct fd_context *ctx, struct fd_ringbuffer *ring,
 
 	fd5_context(ctx)->max_loc = l.max_loc;
 
-	if (emit->key.binning_pass) {
+	if (emit->binning_pass) {
 		OUT_PKT4(ring, REG_A5XX_SP_FS_OBJ_START_LO, 2);
 		OUT_RING(ring, 0x00000000);    /* SP_FS_OBJ_START_LO */
 		OUT_RING(ring, 0x00000000);    /* SP_FS_OBJ_START_HI */
@@ -613,7 +613,7 @@ fd5_program_emit(struct fd_context *ctx, struct fd_ringbuffer *ring,
 	OUT_RING(ring, A5XX_VPC_PACK_NUMNONPOSVAR(s[FS].v->total_in) |
 			A5XX_VPC_PACK_PSIZELOC(psize_loc));
 
-	if (!emit->key.binning_pass) {
+	if (!emit->binning_pass) {
 		uint32_t vinterp[8], vpsrepl[8];
 
 		memset(vinterp, 0, sizeof(vinterp));
@@ -704,7 +704,7 @@ fd5_program_emit(struct fd_context *ctx, struct fd_ringbuffer *ring,
 			OUT_RING(ring, vpsrepl[i]);   /* VPC_VARYING_PS_REPL[i] */
 	}
 
-	if (!emit->key.binning_pass)
+	if (!emit->binning_pass)
 		if (s[FS].instrlen)
 			fd5_emit_shader(ring, s[FS].v);
 
diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_draw.c b/src/gallium/drivers/freedreno/a6xx/fd6_draw.c
index d9e363575db..be13f5f6098 100644
--- a/src/gallium/drivers/freedreno/a6xx/fd6_draw.c
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_draw.c
@@ -156,11 +156,11 @@ draw_impl(struct fd_context *ctx, struct fd_ringbuffer *ring,
 
 	if (info->indirect) {
 		draw_emit_indirect(ctx->batch, ring, primtype,
-						   emit->key.binning_pass ? IGNORE_VISIBILITY : USE_VISIBILITY,
+						   emit->binning_pass ? IGNORE_VISIBILITY : USE_VISIBILITY,
 						   info, index_offset);
 	} else {
 		draw_emit(ctx->batch, ring, primtype,
-				  emit->key.binning_pass ? IGNORE_VISIBILITY : USE_VISIBILITY,
+				  emit->binning_pass ? IGNORE_VISIBILITY : USE_VISIBILITY,
 				  info, index_offset);
 	}
 
@@ -248,13 +248,13 @@ fd6_draw_vbo(struct fd_context *ctx, const struct pipe_draw_info *info,
 	 */
 	emit.no_lrz_write = fp->writes_pos || fp->has_kill;
 
-	emit.key.binning_pass = false;
+	emit.binning_pass = false;
 	emit.dirty = dirty;
 
 	draw_impl(ctx, ctx->batch->draw, &emit, index_offset);
 
 	/* and now binning pass: */
-	emit.key.binning_pass = true;
+	emit.binning_pass = true;
 	emit.dirty = dirty & ~(FD_DIRTY_BLEND);
 	emit.vp = NULL;   /* we changed key so need to refetch vp */
 	emit.fp = NULL;
diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_emit.c b/src/gallium/drivers/freedreno/a6xx/fd6_emit.c
index eb24fb96cfb..706386af9cb 100644
--- a/src/gallium/drivers/freedreno/a6xx/fd6_emit.c
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_emit.c
@@ -595,7 +595,7 @@ fd6_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 
 	emit_marker6(ring, 5);
 
-	if ((dirty & FD_DIRTY_FRAMEBUFFER) && !emit->key.binning_pass) {
+	if ((dirty & FD_DIRTY_FRAMEBUFFER) && !emit->binning_pass) {
 		unsigned char mrt_comp[A6XX_MAX_RENDER_TARGETS] = {0};
 
 		for (unsigned i = 0; i < pfb->nr_cbufs; i++) {
@@ -649,7 +649,7 @@ fd6_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 			if (emit->no_lrz_write || !rsc->lrz || !rsc->lrz_valid) {
 				gras_lrz_cntl = 0;
 				rb_lrz_cntl = 0;
-			} else if (emit->key.binning_pass && zsa->lrz_write) {
+			} else if (emit->binning_pass && zsa->lrz_write) {
 				gras_lrz_cntl |= A6XX_GRAS_LRZ_CNTL_LRZ_WRITE;
 			}
 
@@ -783,7 +783,7 @@ fd6_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 		uint32_t posz_regid = ir3_find_output_regid(fp, FRAG_RESULT_DEPTH);
 		unsigned nr = pfb->nr_cbufs;
 
-		if (emit->key.binning_pass)
+		if (emit->binning_pass)
 			nr = 0;
 		else if (ctx->rasterizer->rasterizer_discard)
 			nr = 0;
@@ -812,7 +812,7 @@ fd6_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 	}
 
 	if ((ctx->dirty_shader[PIPE_SHADER_FRAGMENT] & DIRTY_CONST) &&
-			!emit->key.binning_pass) {
+			!emit->binning_pass) {
 		struct fd_ringbuffer *fsconstobj =
 			fd_ringbuffer_new_flags(ctx->pipe, 0x1000,
 					FD_RINGBUFFER_OBJECT | FD_RINGBUFFER_STREAMING);
diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_emit.h b/src/gallium/drivers/freedreno/a6xx/fd6_emit.h
index 005952750f8..c05b0a45c96 100644
--- a/src/gallium/drivers/freedreno/a6xx/fd6_emit.h
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_emit.h
@@ -61,6 +61,7 @@ struct fd6_emit {
 	const struct fd_vertex_state *vtx;
 	const struct fd_program_stateobj *prog;
 	const struct pipe_draw_info *info;
+	bool binning_pass;
 	struct ir3_shader_key key;
 	enum fd_dirty_3d_state dirty;
 
@@ -90,7 +91,8 @@ fd6_emit_get_vp(struct fd6_emit *emit)
 {
 	if (!emit->vp) {
 		struct ir3_shader *shader = emit->prog->vp;
-		emit->vp = ir3_shader_variant(shader, emit->key, emit->debug);
+		emit->vp = ir3_shader_variant(shader, emit->key,
+				emit->binning_pass, emit->debug);
 	}
 	return emit->vp;
 }
@@ -99,13 +101,14 @@ static inline const struct ir3_shader_variant *
 fd6_emit_get_fp(struct fd6_emit *emit)
 {
 	if (!emit->fp) {
-		if (emit->key.binning_pass) {
+		if (emit->binning_pass) {
 			/* use dummy stateobj to simplify binning vs non-binning: */
 			static const struct ir3_shader_variant binning_fp = {};
 			emit->fp = &binning_fp;
 		} else {
 			struct ir3_shader *shader = emit->prog->fp;
-			emit->fp = ir3_shader_variant(shader, emit->key,emit->debug);
+			emit->fp = ir3_shader_variant(shader, emit->key,
+					false, emit->debug);
 		}
 	}
 	return emit->fp;
diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_program.c b/src/gallium/drivers/freedreno/a6xx/fd6_program.c
index c6d062a3a9a..33f5962ad13 100644
--- a/src/gallium/drivers/freedreno/a6xx/fd6_program.c
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_program.c
@@ -393,7 +393,7 @@ fd6_program_emit(struct fd_context *ctx, struct fd_ringbuffer *ring,
 	ir3_link_shaders(&l, s[VS].v, s[FS].v);
 
 	if ((s[VS].v->shader->stream_output.num_outputs > 0) &&
-			!emit->key.binning_pass)
+			!emit->binning_pass)
 		link_stream_out(&l, s[VS].v);
 
 	BITSET_DECLARE(varbs, 128) = {0};
@@ -419,7 +419,7 @@ fd6_program_emit(struct fd_context *ctx, struct fd_ringbuffer *ring,
 	}
 
 	if ((s[VS].v->shader->stream_output.num_outputs > 0) &&
-			!emit->key.binning_pass) {
+			!emit->binning_pass) {
 		setup_stream_out(ctx, s[VS].v, &l);
 	}
 
@@ -478,7 +478,7 @@ fd6_program_emit(struct fd_context *ctx, struct fd_ringbuffer *ring,
 	OUT_RING(ring, A6XX_PC_PRIMITIVE_CNTL_1_STRIDE_IN_VPC(l.max_loc) |
 			 COND(psize_regid != regid(63,0), 0x100));
 
-	if (emit->key.binning_pass) {
+	if (emit->binning_pass) {
 		OUT_PKT4(ring, REG_A6XX_SP_FS_OBJ_START_LO, 2);
 		OUT_RING(ring, 0x00000000);    /* SP_FS_OBJ_START_LO */
 		OUT_RING(ring, 0x00000000);    /* SP_FS_OBJ_START_HI */
@@ -559,7 +559,7 @@ fd6_program_emit(struct fd_context *ctx, struct fd_ringbuffer *ring,
 			 A6XX_VPC_PACK_PSIZELOC(psize_loc) |
 			 A6XX_VPC_PACK_STRIDE_IN_VPC(l.max_loc));
 
-	if (!emit->key.binning_pass) {
+	if (!emit->binning_pass) {
 		uint32_t vinterp[8], vpsrepl[8];
 
 		memset(vinterp, 0, sizeof(vinterp));
@@ -650,7 +650,7 @@ fd6_program_emit(struct fd_context *ctx, struct fd_ringbuffer *ring,
 			OUT_RING(ring, vpsrepl[i]);   /* VPC_VARYING_PS_REPL[i] */
 	}
 
-	if (!emit->key.binning_pass)
+	if (!emit->binning_pass)
 		if (s[FS].instrlen)
 			fd6_emit_shader(ring, s[FS].v);
 
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_cmdline.c b/src/gallium/drivers/freedreno/ir3/ir3_cmdline.c
index b41c32d3756..bf801953d00 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_cmdline.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_cmdline.c
@@ -289,7 +289,7 @@ int main(int argc, char **argv)
 
 		if (!strcmp(argv[n], "--binning-pass")) {
 			debug_printf(" %s", argv[n]);
-			key.binning_pass = true;
+			v.binning_pass = true;
 			n++;
 			continue;
 		}
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c b/src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c
index 9f6a06cf333..9474f75a9f7 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c
@@ -182,10 +182,9 @@ compile_init(struct ir3_compiler *compiler,
 	NIR_PASS_V(ctx->s, nir_convert_from_ssa, true);
 
 	if (fd_mesa_debug & FD_DBG_DISASM) {
-		DBG("dump nir%dv%d: type=%d, k={bp=%u,cts=%u,hp=%u}",
+		DBG("dump nir%dv%d: type=%d, k={cts=%u,hp=%u}",
 			so->shader->id, so->id, so->type,
-			so->key.binning_pass, so->key.color_two_side,
-			so->key.half_precision);
+			so->key.color_two_side, so->key.half_precision);
 		nir_print_shader(ctx->s, stdout);
 	}
 
@@ -3197,7 +3196,7 @@ emit_function(struct ir3_context *ctx, nir_function_impl *impl)
 	 */
 	if ((ctx->compiler->gpu_id < 500) &&
 			(ctx->so->shader->stream_output.num_outputs > 0) &&
-			!ctx->so->key.binning_pass) {
+			!ctx->so->binning_pass) {
 		debug_assert(ctx->so->type == SHADER_VERTEX);
 		emit_stream_out(ctx);
 	}
@@ -3600,7 +3599,7 @@ ir3_compile_shader_nir(struct ir3_compiler *compiler,
 		fixup_frag_inputs(ctx);
 
 	/* at this point, for binning pass, throw away unneeded outputs: */
-	if (so->key.binning_pass) {
+	if (so->binning_pass) {
 		for (i = 0, j = 0; i < so->outputs_count; i++) {
 			unsigned slot = so->outputs[i].slot;
 
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_shader.c b/src/gallium/drivers/freedreno/ir3/ir3_shader.c
index ee063f84d73..9bf0a7f999c 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_shader.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_shader.c
@@ -146,7 +146,7 @@ assemble_variant(struct ir3_shader_variant *v)
 	if (fd_mesa_debug & FD_DBG_DISASM) {
 		struct ir3_shader_key key = v->key;
 		printf("disassemble: type=%d, k={bp=%u,cts=%u,hp=%u}", v->type,
-			key.binning_pass, key.color_two_side, key.half_precision);
+			v->binning_pass, key.color_two_side, key.half_precision);
 		ir3_shader_disasm(v, bin, stdout);
 	}
 
@@ -194,7 +194,8 @@ dump_shader_info(struct ir3_shader_variant *v, struct pipe_debug_callback *debug
 }
 
 static struct ir3_shader_variant *
-create_variant(struct ir3_shader *shader, struct ir3_shader_key key)
+create_variant(struct ir3_shader *shader, struct ir3_shader_key key,
+		bool binning_pass)
 {
 	struct ir3_shader_variant *v = CALLOC_STRUCT(ir3_shader_variant);
 	int ret;
@@ -204,6 +205,7 @@ create_variant(struct ir3_shader *shader, struct ir3_shader_key key)
 
 	v->id = ++shader->variant_count;
 	v->shader = shader;
+	v->binning_pass = binning_pass;
 	v->key = key;
 	v->type = shader->type;
 
@@ -226,8 +228,8 @@ fail:
 	return NULL;
 }
 
-struct ir3_shader_variant *
-ir3_shader_variant(struct ir3_shader *shader, struct ir3_shader_key key,
+static inline struct ir3_shader_variant *
+shader_variant(struct ir3_shader *shader, struct ir3_shader_key key,
 		struct pipe_debug_callback *debug)
 {
 	struct ir3_shader_variant *v;
@@ -238,7 +240,6 @@ ir3_shader_variant(struct ir3_shader *shader, struct ir3_shader_key key,
 	 */
 	switch (shader->type) {
 	case SHADER_FRAGMENT:
-		key.binning_pass = false;
 		if (key.has_per_samp) {
 			key.vsaturate_s = 0;
 			key.vsaturate_t = 0;
@@ -269,7 +270,7 @@ ir3_shader_variant(struct ir3_shader *shader, struct ir3_shader_key key,
 			return v;
 
 	/* compile new variant if it doesn't exist already: */
-	v = create_variant(shader, key);
+	v = create_variant(shader, key, false);
 	if (v) {
 		v->next = shader->variants;
 		shader->variants = v;
@@ -280,6 +281,22 @@ ir3_shader_variant(struct ir3_shader *shader, struct ir3_shader_key key,
 }
 
 
+struct ir3_shader_variant *
+ir3_shader_variant(struct ir3_shader *shader, struct ir3_shader_key key,
+		bool binning_pass, struct pipe_debug_callback *debug)
+{
+	struct ir3_shader_variant *v =
+			shader_variant(shader, key, debug);
+
+	if (binning_pass) {
+		if (!v->binning)
+			v->binning = create_variant(shader, key, true);
+		return v->binning;
+	}
+
+	return v;
+}
+
 void
 ir3_shader_destroy(struct ir3_shader *shader)
 {
@@ -332,7 +349,7 @@ ir3_shader_create(struct ir3_compiler *compiler,
 		 */
 		static struct ir3_shader_key key;
 		memset(&key, 0, sizeof(key));
-		ir3_shader_variant(shader, key, debug);
+		ir3_shader_variant(shader, key, false, debug);
 	}
 	return shader;
 }
@@ -755,7 +772,7 @@ max_tf_vtx(struct fd_context *ctx, const struct ir3_shader_variant *v)
 
 	if (ctx->screen->gpu_id >= 500)
 		return 0;
-	if (v->key.binning_pass)
+	if (v->binning_pass)
 		return 0;
 	if (v->shader->stream_output.num_outputs == 0)
 		return 0;
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_shader.h b/src/gallium/drivers/freedreno/ir3/ir3_shader.h
index 6bc24f47d75..1c31061af47 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_shader.h
+++ b/src/gallium/drivers/freedreno/ir3/ir3_shader.h
@@ -114,7 +114,6 @@ struct ir3_shader_key {
 			/*
 			 * Vertex shader variant parameters:
 			 */
-			unsigned binning_pass : 1;
 			unsigned vclamp_color : 1;
 
 			/*
@@ -218,6 +217,12 @@ struct ir3_shader_variant {
 
 	struct ir3_shader_key key;
 
+	/* vertex shaders can have an extra version for hwbinning pass,
+	 * which is pointed to by so->binning:
+	 */
+	bool binning_pass;
+	struct ir3_shader_variant *binning;
+
 	struct ir3_driver_const_layout const_layout;
 	struct ir3_info info;
 	struct ir3 *ir;
@@ -373,7 +378,8 @@ ir3_shader_create_compute(struct ir3_compiler *compiler,
 		struct pipe_debug_callback *debug);
 void ir3_shader_destroy(struct ir3_shader *shader);
 struct ir3_shader_variant * ir3_shader_variant(struct ir3_shader *shader,
-		struct ir3_shader_key key, struct pipe_debug_callback *debug);
+		struct ir3_shader_key key, bool binning_pass,
+		struct pipe_debug_callback *debug);
 void ir3_shader_disasm(struct ir3_shader_variant *so, uint32_t *bin, FILE *out);
 uint64_t ir3_shader_outputs(const struct ir3_shader *so);