[Mesa-dev] [PATCH 13/14] ac/radv: get rid of geometry user sgpr for num entries.

[Dave Airlie]

From: Dave Airlie <airlied at redhat.com>

This drops one of the geometry specific user sgprs,
we can work this out at compile time.

Signed-off-by: Dave Airlie <airlied at redhat.com>
---
 src/amd/common/ac_nir_to_llvm.c | 15 +++++++--------
 src/amd/vulkan/radv_pipeline.c  |  9 +--------
 2 files changed, 8 insertions(+), 16 deletions(-)

diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
index 1cf181ddeba..9e4069b535a 100644
--- a/src/amd/common/ac_nir_to_llvm.c
+++ b/src/amd/common/ac_nir_to_llvm.c
@@ -89,7 +89,6 @@ struct radv_shader_context {
 	LLVMValueRef tes_v;
 
 	LLVMValueRef gsvs_ring_stride;
-	LLVMValueRef gsvs_num_entries;
 	LLVMValueRef gs2vs_offset;
 	LLVMValueRef gs_wave_id;
 	LLVMValueRef gs_vtx_offset[6];
@@ -637,7 +636,7 @@ static void allocate_user_sgprs(struct radv_shader_context *ctx,
 				user_sgpr_info->sgpr_count += count_vs_user_sgprs(ctx);
 			}
 		}
-		user_sgpr_info->sgpr_count += 2;
+		user_sgpr_info->sgpr_count += 1;
 		break;
 	default:
 		break;
@@ -969,8 +968,6 @@ static void create_function(struct radv_shader_context *ctx,
 
 			add_arg(&args, ARG_SGPR, ctx->ac.i32,
 				&ctx->gsvs_ring_stride);
-			add_arg(&args, ARG_SGPR, ctx->ac.i32,
-				&ctx->gsvs_num_entries);
 			if (needs_view_index)
 				add_arg(&args, ARG_SGPR, ctx->ac.i32,
 					&ctx->abi.view_index);
@@ -1000,8 +997,6 @@ static void create_function(struct radv_shader_context *ctx,
 
 			add_arg(&args, ARG_SGPR, ctx->ac.i32,
 				&ctx->gsvs_ring_stride);
-			add_arg(&args, ARG_SGPR, ctx->ac.i32,
-				&ctx->gsvs_num_entries);
 			if (needs_view_index)
 				add_arg(&args, ARG_SGPR, ctx->ac.i32,
 					&ctx->abi.view_index);
@@ -1128,7 +1123,7 @@ static void create_function(struct radv_shader_context *ctx,
 							   &user_sgpr_idx);
 		}
 		set_loc_shader(ctx, AC_UD_GS_VS_RING_STRIDE_ENTRIES,
-			       &user_sgpr_idx, 2);
+			       &user_sgpr_idx, 1);
 		if (ctx->abi.view_index)
 			set_loc_shader(ctx, AC_UD_VIEW_INDEX, &user_sgpr_idx, 1);
 		break;
@@ -6744,12 +6739,16 @@ ac_setup_rings(struct radv_shader_context *ctx)
 	}
 	if (ctx->stage == MESA_SHADER_GEOMETRY) {
 		LLVMValueRef tmp;
+		uint32_t num_entries = 64;
 		ctx->esgs_ring = ac_build_load_to_sgpr(&ctx->ac, ctx->ring_offsets, LLVMConstInt(ctx->ac.i32, RING_ESGS_GS, false));
 		ctx->gsvs_ring = ac_build_load_to_sgpr(&ctx->ac, ctx->ring_offsets, LLVMConstInt(ctx->ac.i32, RING_GSVS_GS, false));
 
 		ctx->gsvs_ring = LLVMBuildBitCast(ctx->ac.builder, ctx->gsvs_ring, ctx->ac.v4i32, "");
 
-		ctx->gsvs_ring = LLVMBuildInsertElement(ctx->ac.builder, ctx->gsvs_ring, ctx->gsvs_num_entries, LLVMConstInt(ctx->ac.i32, 2, false), "");
+		tmp = LLVMConstInt(ctx->ac.i32, num_entries, false);
+		if (ctx->options->chip_class >= VI)
+			tmp = LLVMBuildMul(ctx->ac.builder, LLVMBuildLShr(ctx->ac.builder, ctx->gsvs_ring_stride, LLVMConstInt(ctx->ac.i32, 16, false), ""), tmp, "");
+		ctx->gsvs_ring = LLVMBuildInsertElement(ctx->ac.builder, ctx->gsvs_ring, tmp, LLVMConstInt(ctx->ac.i32, 2, false), "");
 		tmp = LLVMBuildExtractElement(ctx->ac.builder, ctx->gsvs_ring, ctx->ac.i32_1, "");
 		tmp = LLVMBuildOr(ctx->ac.builder, tmp, ctx->gsvs_ring_stride, "");
 		ctx->gsvs_ring = LLVMBuildInsertElement(ctx->ac.builder, ctx->gsvs_ring, tmp, ctx->ac.i32_1, "");
diff --git a/src/amd/vulkan/radv_pipeline.c b/src/amd/vulkan/radv_pipeline.c
index e9a9ae975b1..5d1b5f6e352 100644
--- a/src/amd/vulkan/radv_pipeline.c
+++ b/src/amd/vulkan/radv_pipeline.c
@@ -2622,16 +2622,9 @@ radv_pipeline_generate_geometry_shader(struct radeon_winsys_cs *cs,
 							     AC_UD_GS_VS_RING_STRIDE_ENTRIES);
 	if (loc->sgpr_idx != -1) {
 		uint32_t stride = gs->info.gs.max_gsvs_emit_size;
-		uint32_t num_entries = 64;
-		bool is_vi = pipeline->device->physical_device->rad_info.chip_class >= VI;
-
-		if (is_vi)
-			num_entries *= stride;
-
 		stride = S_008F04_STRIDE(stride);
-		radeon_set_sh_reg_seq(cs, R_00B230_SPI_SHADER_USER_DATA_GS_0 + loc->sgpr_idx * 4, 2);
+		radeon_set_sh_reg_seq(cs, R_00B230_SPI_SHADER_USER_DATA_GS_0 + loc->sgpr_idx * 4, 1);
 		radeon_emit(cs, stride);
-		radeon_emit(cs, num_entries);
 	}
 }
 
-- 
2.14.3