[Mesa-dev] [PATCH 16/19] radeonsi: don't call build_gep0 in si_llvm_emit_ddxy on VI

[Marek Olšák]

From: Marek Olšák <marek.olsak at amd.com>

---
 src/gallium/drivers/radeonsi/si_shader.c | 14 +++++++++-----
 1 file changed, 9 insertions(+), 5 deletions(-)

diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index 67ab16b..414810e 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -5007,62 +5007,66 @@ static void si_llvm_emit_ddxy(
 	struct gallivm_state *gallivm = bld_base->base.gallivm;
 	const struct tgsi_full_instruction *inst = emit_data->inst;
 	unsigned opcode = inst->Instruction.Opcode;
 	LLVMValueRef store_ptr, load_ptr0, load_ptr1, thread_id;
 	LLVMValueRef tl, trbl, result[4];
 	LLVMValueRef tl_tid, trbl_tid;
 	unsigned swizzle[4];
 	unsigned c;
 	int idx;
 	unsigned mask;
+	bool has_ds_bpermute = HAVE_LLVM >= 0x0309 &&
+			       ctx->screen->b.chip_class >= VI;
 
 	thread_id = get_thread_id(ctx);;
-	store_ptr = build_gep0(ctx, ctx->lds, thread_id);
 
 	if (opcode == TGSI_OPCODE_DDX_FINE)
 		mask = TID_MASK_LEFT;
 	else if (opcode == TGSI_OPCODE_DDY_FINE)
 		mask = TID_MASK_TOP;
 	else
 		mask = TID_MASK_TOP_LEFT;
 
 	tl_tid = LLVMBuildAnd(gallivm->builder, thread_id,
 				lp_build_const_int32(gallivm, mask), "");
-	load_ptr0 = build_gep0(ctx, ctx->lds, tl_tid);
 
 	/* for DDX we want to next X pixel, DDY next Y pixel. */
 	idx = (opcode == TGSI_OPCODE_DDX || opcode == TGSI_OPCODE_DDX_FINE) ? 1 : 2;
 	trbl_tid = LLVMBuildAdd(gallivm->builder, tl_tid,
 				  lp_build_const_int32(gallivm, idx), "");
-	load_ptr1 = build_gep0(ctx, ctx->lds, trbl_tid);
+
+	if (!has_ds_bpermute) {
+		store_ptr = build_gep0(ctx, ctx->lds, thread_id);
+		load_ptr0 = build_gep0(ctx, ctx->lds, tl_tid);
+		load_ptr1 = build_gep0(ctx, ctx->lds, trbl_tid);
+	}
 
 	for (c = 0; c < 4; ++c) {
 		unsigned i;
 		LLVMValueRef val;
 		LLVMValueRef args[2];
 
 		swizzle[c] = tgsi_util_get_full_src_register_swizzle(&inst->Src[0], c);
 		for (i = 0; i < c; ++i) {
 			if (swizzle[i] == swizzle[c]) {
 				result[c] = result[i];
 				break;
 			}
 		}
 		if (i != c)
 			continue;
 
 		val = LLVMBuildBitCast(gallivm->builder,
 				lp_build_emit_fetch(bld_base, inst, 0, c),
 						ctx->i32, "");
 
-		if ((HAVE_LLVM >= 0x0309) && ctx->screen->b.family >= CHIP_TONGA) {
-
+		if (has_ds_bpermute) {
 	                args[0] = LLVMBuildMul(gallivm->builder, tl_tid,
                                         lp_build_const_int32(gallivm, 4), "");
 			args[1] = val;
 			tl = lp_build_intrinsic(gallivm->builder,
 					"llvm.amdgcn.ds.bpermute", ctx->i32,
 					args, 2, LLVMReadNoneAttribute);
 
 	                args[0] = LLVMBuildMul(gallivm->builder, trbl_tid,
                                         lp_build_const_int32(gallivm, 4), "");
 			trbl = lp_build_intrinsic(gallivm->builder,
-- 
2.7.4