[Mesa-dev] [PATCH 04/20] radeonsi: implement shared memory load/store

[Bas Nieuwenhuizen]

Signed-off-by: Bas Nieuwenhuizen <bas at basnieuwenhuizen.nl>
---
 src/gallium/drivers/radeonsi/si_shader.c | 75 +++++++++++++++++++++++++++++++-
 1 file changed, 73 insertions(+), 2 deletions(-)

diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index 2ce37ca..97d4404 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -3078,7 +3078,7 @@ static void load_fetch_args(
 
 		buffer_append_args(ctx, emit_data, rsrc, bld_base->uint_bld.zero,
 				   offset, false);
-	} else {
+	} else if (inst->Src[0].Register.File == TGSI_FILE_IMAGE) {
 		LLVMValueRef coords;
 
 		image_fetch_rsrc(bld_base, &inst->Src[0], false, &rsrc);
@@ -3131,6 +3131,38 @@ static void load_emit_buffer(struct si_shader_context *ctx,
 			LLVMReadOnlyAttribute | LLVMNoUnwindAttribute);
 }
 
+static void load_emit_memory(
+		struct si_shader_context *ctx,
+		struct lp_build_emit_data *emit_data)
+{
+	const struct tgsi_full_instruction *inst = emit_data->inst;
+	struct lp_build_context *base = &ctx->radeon_bld.soa.bld_base.base;
+	struct gallivm_state *gallivm = &ctx->radeon_bld.gallivm;
+	LLVMBuilderRef builder = gallivm->builder;
+	unsigned writemask = inst->Dst[0].Register.WriteMask;
+	LLVMValueRef channels[4], offset, ptr, derived_ptr, index;
+	int chan, addr_space;
+
+	offset = lp_build_emit_fetch(&ctx->radeon_bld.soa.bld_base, inst, 1, 0);
+	offset = LLVMBuildBitCast(builder, offset, ctx->i32, "");
+
+	ptr = ctx->memory_regions[inst->Src[0].Register.Index];
+	ptr = LLVMBuildGEP(builder, ptr, &offset, 1, "");
+	addr_space = LLVMGetPointerAddressSpace(LLVMTypeOf(ptr));
+	ptr = LLVMBuildBitCast(builder, ptr, LLVMPointerType(base->elem_type, addr_space), "");
+
+	for (chan = 0; chan < 4; ++chan) {
+		if (!(writemask & (1 << chan))) {
+			channels[chan] = LLVMGetUndef(base->elem_type);
+			continue;
+		}
+		index = lp_build_const_int32(gallivm, chan);
+		derived_ptr = LLVMBuildGEP(builder, ptr, &index, 1, "");
+		channels[chan] = LLVMBuildLoad(builder, derived_ptr, "");
+	}
+	emit_data->output[emit_data->chan] = lp_build_gather_values(gallivm, channels, 4);
+}
+
 static void load_emit(
 		const struct lp_build_tgsi_action *action,
 		struct lp_build_tgsi_context *bld_base,
@@ -3143,6 +3175,11 @@ static void load_emit(
 	char intrinsic_name[32];
 	char coords_type[8];
 
+	if (inst->Src[0].Register.File == TGSI_FILE_MEMORY) {
+		load_emit_memory(ctx, emit_data);
+		return;
+	}
+
 	if (inst->Memory.Qualifier & TGSI_MEMORY_VOLATILE)
 		emit_optimization_barrier(ctx);
 
@@ -3208,7 +3245,7 @@ static void store_fetch_args(
 
 		buffer_append_args(ctx, emit_data, rsrc, bld_base->uint_bld.zero,
 				   offset, false);
-	} else {
+	} else if (inst->Dst[0].Register.File == TGSI_FILE_IMAGE) {
 		unsigned target = inst->Memory.Texture;
 		LLVMValueRef coords;
 
@@ -3304,6 +3341,37 @@ static void store_emit_buffer(
 	}
 }
 
+static void store_emit_memory(
+		struct si_shader_context *ctx,
+		struct lp_build_emit_data *emit_data)
+{
+	const struct tgsi_full_instruction *inst = emit_data->inst;
+	struct gallivm_state *gallivm = &ctx->radeon_bld.gallivm;
+	struct lp_build_context *base = &ctx->radeon_bld.soa.bld_base.base;
+	LLVMBuilderRef builder = gallivm->builder;
+	unsigned writemask = inst->Dst[0].Register.WriteMask;
+	LLVMValueRef  offset, ptr, derived_ptr, data, index;
+	int chan, addr_space;
+
+	offset = lp_build_emit_fetch(&ctx->radeon_bld.soa.bld_base, inst, 0, 0);
+	offset = LLVMBuildBitCast(builder, offset, ctx->i32, "");
+
+	ptr = ctx->memory_regions[inst->Dst[0].Register.Index];
+	ptr = LLVMBuildGEP(builder, ptr, &offset, 1, "");
+	addr_space = LLVMGetPointerAddressSpace(LLVMTypeOf(ptr));
+	ptr = LLVMBuildBitCast(builder, ptr, LLVMPointerType(base->elem_type, addr_space), "");
+
+	for (chan = 0; chan < 4; ++chan) {
+		if (!(writemask & (1 << chan))) {
+			continue;
+		}
+		data = lp_build_emit_fetch(&ctx->radeon_bld.soa.bld_base, inst, 1, chan);
+		index = lp_build_const_int32(gallivm, chan);
+		derived_ptr = LLVMBuildGEP(builder, ptr, &index, 1, "");
+		LLVMBuildStore(builder, data, derived_ptr);
+	}
+}
+
 static void store_emit(
 		const struct lp_build_tgsi_action *action,
 		struct lp_build_tgsi_context *bld_base,
@@ -3319,6 +3387,9 @@ static void store_emit(
 	if (inst->Dst[0].Register.File == TGSI_FILE_BUFFER) {
 		store_emit_buffer(si_shader_context(bld_base), emit_data);
 		return;
+	} else if (inst->Dst[0].Register.File == TGSI_FILE_MEMORY) {
+		store_emit_memory(si_shader_context(bld_base), emit_data);
+		return;
 	}
 
 	if (target == TGSI_TEXTURE_BUFFER) {
-- 
2.7.4