From: Józef Kucia Subject: [PATCH vkd3d 1/6] vkd3d-shader: Fix bitfield instructions. Message-Id: <20181018125947.27607-1-joseph.kucia@gmail.com> Date: Thu, 18 Oct 2018 14:59:42 +0200 From: Józef Kucia Avoid overwriting registers before all results are computed. Signed-off-by: Józef Kucia --- libs/vkd3d-shader/spirv.c | 13 +++++++------ tests/d3d12.c | 25 +++++++++++++++++++++++++ 2 files changed, 32 insertions(+), 6 deletions(-) diff --git a/libs/vkd3d-shader/spirv.c b/libs/vkd3d-shader/spirv.c index 7cf58f8eb2f2..1d97a98c2e80 100644 --- a/libs/vkd3d-shader/spirv.c +++ b/libs/vkd3d-shader/spirv.c @@ -4678,11 +4678,11 @@ static void vkd3d_dxbc_compiler_emit_udiv(struct vkd3d_dxbc_compiler *compiler, static void vkd3d_dxbc_compiler_emit_bitfield_instruction(struct vkd3d_dxbc_compiler *compiler, const struct vkd3d_shader_instruction *instruction) { + uint32_t src_ids[4], constituents[VKD3D_VEC4_SIZE], type_id, mask_id; struct vkd3d_spirv_builder *builder = &compiler->spirv_builder; const struct vkd3d_shader_dst_param *dst = instruction->dst; const struct vkd3d_shader_src_param *src = instruction->src; - uint32_t src_ids[4], result_id, type_id, mask_id; - unsigned int i, j, src_count; + unsigned int i, j, k, src_count; DWORD write_mask; SpvOp op; @@ -4702,7 +4702,7 @@ static void vkd3d_dxbc_compiler_emit_bitfield_instruction(struct vkd3d_dxbc_comp return; } - for (i = 0; i < VKD3D_VEC4_SIZE; ++i) + for (i = 0, k = 0; i < VKD3D_VEC4_SIZE; ++i) { if (!(write_mask = dst->write_mask & (VKD3DSP_WRITEMASK_0 << i))) continue; @@ -4717,11 +4717,12 @@ static void vkd3d_dxbc_compiler_emit_bitfield_instruction(struct vkd3d_dxbc_comp src_ids[j] = vkd3d_spirv_build_op_and(builder, int_type_id, src_ids[j], mask_id); } - result_id = vkd3d_spirv_build_op_trv(builder, &builder->function_stream, + constituents[k++] = vkd3d_spirv_build_op_trv(builder, &builder->function_stream, op, type_id, src_ids, src_count); - - vkd3d_dxbc_compiler_emit_store_reg(compiler, &dst->reg, write_mask, result_id); } + + vkd3d_dxbc_compiler_emit_store_dst_components(compiler, + dst, vkd3d_component_type_from_data_type(dst->reg.data_type), constituents); } static void vkd3d_dxbc_compiler_emit_f16tof32(struct vkd3d_dxbc_compiler *compiler, diff --git a/tests/d3d12.c b/tests/d3d12.c index f40aded6ca98..34c4802f7fde 100644 --- a/tests/d3d12.c +++ b/tests/d3d12.c @@ -6285,6 +6285,29 @@ static void test_shader_instructions(void) 0x00000000, 0x00208aa6, 0x00000000, 0x00000000, 0x0100003e, }; static const D3D12_SHADER_BYTECODE ps_ibfe = {ps_ibfe_code, sizeof(ps_ibfe_code)}; + static const DWORD ps_ibfe2_code[] = + { +#if 0 + ps_5_0 + dcl_globalFlags refactoringAllowed + dcl_constantbuffer cb0[1], immediateIndexed + dcl_output o0.xyzw + dcl_temps 1 + mov r0.xyzw, cb0[0].xyzw + ibfe r0.xyzw, r0.xxxx, r0.yyyy, r0.zzzz + mov o0.xyzw, r0.xyzw + ret +#endif + 0x43425844, 0x347a9c0e, 0x3eff39a4, 0x3dd41cc5, 0xff87ec8d, 0x00000001, 0x000000fc, 0x00000003, + 0x0000002c, 0x0000003c, 0x00000070, 0x4e475349, 0x00000008, 0x00000000, 0x00000008, 0x4e47534f, + 0x0000002c, 0x00000001, 0x00000008, 0x00000020, 0x00000000, 0x00000000, 0x00000001, 0x00000000, + 0x0000000f, 0x545f5653, 0x65677261, 0xabab0074, 0x58454853, 0x00000084, 0x00000050, 0x00000021, + 0x0100086a, 0x04000059, 0x00208e46, 0x00000000, 0x00000001, 0x03000065, 0x001020f2, 0x00000000, + 0x02000068, 0x00000001, 0x06000036, 0x001000f2, 0x00000000, 0x00208e46, 0x00000000, 0x00000000, + 0x0900008b, 0x001000f2, 0x00000000, 0x00100006, 0x00000000, 0x00100556, 0x00000000, 0x00100aa6, + 0x00000000, 0x05000036, 0x001020f2, 0x00000000, 0x00100e46, 0x00000000, 0x0100003e, + }; + static const D3D12_SHADER_BYTECODE ps_ibfe2 = {ps_ibfe2_code, sizeof(ps_ibfe2_code)}; static const DWORD ps_ubfe_code[] = { #if 0 @@ -7203,6 +7226,8 @@ static void test_shader_instructions(void) {&ps_ibfe, {{{31, 31, 0x80000000}}}, {{0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff}}}, {&ps_ibfe, {{{31, 31, 0x7fffffff}}}, {{0x00000000, 0x00000000, 0x00000000, 0x00000000}}}, + {&ps_ibfe2, {{{16, 15, 0x3fffffff}}}, {{0x00007fff, 0x00007fff, 0x00007fff, 0x00007fff}}}, + {&ps_ubfe, {{{0x00000000}}}, {{0x00000000, 0x00000000, 0x00000000, 0x00000000}}}, {&ps_ubfe, {{{0xffffffff}}}, {{0x0000000f, 0x007fffff, 0x0000007f, 0x3fffffff}}}, {&ps_ubfe, {{{0xff000000}}}, {{0x00000000, 0x007f0000, 0x00000000, 0x3f800000}}}, -- 2.18.1