radeonsi: fix GPU hangs with bindless textures and LLVM 7.0

Tested-by: Dieter Nützel <[email protected]>

3 files changed, 33 insertions, 5 deletions

diff --git a/src/gallium/drivers/radeonsi/si_shader_internal.h b/src/gallium/drivers/radeonsi/si_shader_internal.h
index a638dbf8f1a..235c46ecf92 100644
--- a/src/gallium/drivers/radeonsi/si_shader_internal.h
+++ b/src/gallium/drivers/radeonsi/si_shader_internal.h
@@ -305,7 +305,8 @@ LLVMValueRef si_load_sampler_desc(struct si_shader_context *ctx,
 				  enum ac_descriptor_type type);
 LLVMValueRef si_load_image_desc(struct si_shader_context *ctx,
 				LLVMValueRef list, LLVMValueRef index,
-				enum ac_descriptor_type desc_type, bool dcc_off);
+				enum ac_descriptor_type desc_type, bool dcc_off,
+				bool bindless);
 
 void si_load_system_value(struct si_shader_context *ctx,
 			  unsigned index,
diff --git a/src/gallium/drivers/radeonsi/si_shader_nir.c b/src/gallium/drivers/radeonsi/si_shader_nir.c
index 0aefca22385..5d6280b80f7 100644
--- a/src/gallium/drivers/radeonsi/si_shader_nir.c
+++ b/src/gallium/drivers/radeonsi/si_shader_nir.c
@@ -920,7 +920,7 @@ si_nir_load_sampler_desc(struct ac_shader_abi *abi,
 				     index, "");
 
 		/* TODO: be smarter about when we use dcc_off */
-		return si_load_image_desc(ctx, list, index, desc_type, write);
+		return si_load_image_desc(ctx, list, index, desc_type, write, bindless);
 	}
 
 	assert(base_index + constant_index < ctx->num_samplers);
@@ -931,6 +931,16 @@ si_nir_load_sampler_desc(struct ac_shader_abi *abi,
 	index = LLVMBuildAdd(ctx->ac.builder, index,
 			     LLVMConstInt(ctx->i32, SI_NUM_IMAGES / 2, 0), "");
 
+	if (bindless) {
+		/* Since bindless handle arithmetic can contain an unsigned integer
+		 * wraparound and si_load_sampler_desc assumes there isn't any,
+		 * use GEP without "inbounds" (inside ac_build_pointer_add)
+		 * to prevent incorrect code generation and hangs.
+		 */
+		index = LLVMBuildMul(ctx->ac.builder, index, LLVMConstInt(ctx->i32, 2, 0), "");
+		list = ac_build_pointer_add(&ctx->ac, list, index);
+		index = ctx->i32_0;
+	}
 	return si_load_sampler_desc(ctx, list, index, desc_type);
 }
 
diff --git a/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c b/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c
index eaa200a95d6..cabc448a082 100644
--- a/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c
+++ b/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c
@@ -176,7 +176,8 @@ static LLVMValueRef force_dcc_off(struct si_shader_context *ctx,
 
 LLVMValueRef si_load_image_desc(struct si_shader_context *ctx,
 				LLVMValueRef list, LLVMValueRef index,
-				enum ac_descriptor_type desc_type, bool dcc_off)
+				enum ac_descriptor_type desc_type, bool dcc_off,
+				bool bindless)
 {
 	LLVMBuilderRef builder = ctx->ac.builder;
 	LLVMValueRef rsrc;
@@ -190,7 +191,11 @@ LLVMValueRef si_load_image_desc(struct si_shader_context *ctx,
 		assert(desc_type == AC_DESC_IMAGE);
 	}
 
-	rsrc = ac_build_load_to_sgpr(&ctx->ac, list, index);
+	if (bindless)
+		rsrc = ac_build_load_to_sgpr_uint_wraparound(&ctx->ac, list, index);
+	else
+		rsrc = ac_build_load_to_sgpr(&ctx->ac, list, index);
+
 	if (desc_type == AC_DESC_IMAGE && dcc_off)
 		rsrc = force_dcc_off(ctx, rsrc);
 	return rsrc;
@@ -240,6 +245,8 @@ image_fetch_rsrc(
 				     index, "");
 	}
 
+	bool bindless = false;
+
 	if (image->Register.File != TGSI_FILE_IMAGE) {
 		/* Bindless descriptors are accessible from a different pair of
 		 * user SGPR indices.
@@ -254,11 +261,12 @@ image_fetch_rsrc(
 		 */
 		index = LLVMBuildMul(ctx->ac.builder, index,
 				     LLVMConstInt(ctx->i32, 2, 0), "");
+		bindless = true;
 	}
 
 	*rsrc = si_load_image_desc(ctx, rsrc_ptr, index,
 				   target == TGSI_TEXTURE_BUFFER ? AC_DESC_BUFFER : AC_DESC_IMAGE,
-				   dcc_off);
+				   dcc_off, bindless);
 }
 
 static void image_fetch_coords(
@@ -1068,6 +1076,15 @@ static void tex_fetch_ptrs(struct lp_build_tgsi_context *bld_base,
 				    ctx->param_bindless_samplers_and_images);
 		index = lp_build_emit_fetch_src(bld_base, reg,
 						TGSI_TYPE_UNSIGNED, 0);
+
+		/* Since bindless handle arithmetic can contain an unsigned integer
+		 * wraparound and si_load_sampler_desc assumes there isn't any,
+		 * use GEP without "inbounds" (inside ac_build_pointer_add)
+		 * to prevent incorrect code generation and hangs.
+		 */
+		index = LLVMBuildMul(ctx->ac.builder, index, LLVMConstInt(ctx->i32, 2, 0), "");
+		list = ac_build_pointer_add(&ctx->ac, list, index);
+		index = ctx->i32_0;
 	}
 
 	if (target == TGSI_TEXTURE_BUFFER)