radeonsi: move TCS_OUT_LAYOUT.PatchVerticesIn to lower bits

For a later patch.

Reviewed-by: Nicolai Hähnle <[email protected]>

3 files changed, 6 insertions, 5 deletions

diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index 6f8fee6c28e..efdf80b961e 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -2014,7 +2014,7 @@ static LLVMValueRef si_load_patch_vertices_in(struct ac_shader_abi *abi)
 {
 	struct si_shader_context *ctx = si_shader_context_from_abi(abi);
 	if (ctx->type == PIPE_SHADER_TESS_CTRL)
-		return unpack_param(ctx, ctx->param_tcs_out_lds_layout, 26, 6);
+		return unpack_param(ctx, ctx->param_tcs_out_lds_layout, 13, 6);
 	else if (ctx->type == PIPE_SHADER_TESS_EVAL)
 		return get_num_tcs_out_vertices(ctx);
 	else
diff --git a/src/gallium/drivers/radeonsi/si_shader_internal.h b/src/gallium/drivers/radeonsi/si_shader_internal.h
index 42a1b9f107c..40947ffc079 100644
--- a/src/gallium/drivers/radeonsi/si_shader_internal.h
+++ b/src/gallium/drivers/radeonsi/si_shader_internal.h
@@ -162,7 +162,7 @@ struct si_shader_context {
 	/* Layout of TCS outputs / TES inputs:
 	 *   [0:12] = stride between output patches in DW, num_outputs * num_vertices * 4
 	 *            max = 32*32*4 + 32*4
-	 *   [26:31] = gl_PatchVerticesIn, max = 32
+	 *   [13:18] = gl_PatchVerticesIn, max = 32
 	 */
 	int param_tcs_out_lds_layout;
 	int param_tcs_offchip_addr_base64k;
diff --git a/src/gallium/drivers/radeonsi/si_state_draw.c b/src/gallium/drivers/radeonsi/si_state_draw.c
index 06ef84d20da..b245a383987 100644
--- a/src/gallium/drivers/radeonsi/si_state_draw.c
+++ b/src/gallium/drivers/radeonsi/si_state_draw.c
@@ -233,7 +233,8 @@ static void si_emit_derived_tess_state(struct si_context *sctx,
 
 	tcs_in_layout = S_VS_STATE_LS_OUT_PATCH_SIZE(input_patch_size / 4) |
 			S_VS_STATE_LS_OUT_VERTEX_SIZE(input_vertex_size / 4);
-	tcs_out_layout = output_patch_size / 4;
+	tcs_out_layout = (output_patch_size / 4) |
+			 (num_tcs_input_cp << 13);
 	tcs_out_offsets = (output_patch0_offset / 16) |
 			  ((perpatch_output_offset / 16) << 16);
 	offchip_layout = *num_patches |
@@ -268,7 +269,7 @@ static void si_emit_derived_tess_state(struct si_context *sctx,
 				      GFX9_SGPR_TCS_OFFCHIP_LAYOUT * 4, 3);
 		radeon_emit(cs, offchip_layout);
 		radeon_emit(cs, tcs_out_offsets);
-		radeon_emit(cs, tcs_out_layout | (num_tcs_input_cp << 26));
+		radeon_emit(cs, tcs_out_layout);
 	} else {
 		unsigned ls_rsrc2 = ls_current->config.rsrc2;
 
@@ -288,7 +289,7 @@ static void si_emit_derived_tess_state(struct si_context *sctx,
 			R_00B430_SPI_SHADER_USER_DATA_HS_0 + GFX6_SGPR_TCS_OFFCHIP_LAYOUT * 4, 4);
 		radeon_emit(cs, offchip_layout);
 		radeon_emit(cs, tcs_out_offsets);
-		radeon_emit(cs, tcs_out_layout | (num_tcs_input_cp << 26));
+		radeon_emit(cs, tcs_out_layout);
 		radeon_emit(cs, tcs_in_layout);
 	}