diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi.c index e450092a82c..66f752989ab 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi.c @@ -313,7 +313,7 @@ lp_build_tgsi_inst_llvm( } if (info->num_dst > 0 && info->opcode != TGSI_OPCODE_STORE) { - bld_base->emit_store(bld_base, inst, info, emit_data.output); + bld_base->emit_store(bld_base, inst, info, 0, emit_data.output); } return TRUE; } diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h b/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h index eb632b700ab..eeeea507810 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h @@ -370,6 +370,7 @@ struct lp_build_tgsi_context void (*emit_store)(struct lp_build_tgsi_context *, const struct tgsi_full_instruction *, const struct tgsi_opcode_info *, + unsigned index, LLVMValueRef dst[4]); void (*emit_declaration)(struct lp_build_tgsi_context *, diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c index 3e47372b5cf..435d1075aaa 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c @@ -1913,19 +1913,18 @@ emit_store( struct lp_build_tgsi_context * bld_base, const struct tgsi_full_instruction * inst, const struct tgsi_opcode_info * info, + unsigned index, LLVMValueRef dst[4]) { - unsigned chan_index; enum tgsi_opcode_type dtype = tgsi_opcode_infer_dst_type(inst->Instruction.Opcode); - if(info->num_dst) { - TGSI_FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { - - if (tgsi_type_is_64bit(dtype) && (chan_index == 1 || chan_index == 3)) - continue; - emit_store_chan(bld_base, inst, 0, chan_index, dst[chan_index]); - } + unsigned writemask = inst->Dst[index].Register.WriteMask; + while (writemask) { + unsigned chan_index = u_bit_scan(&writemask); + if (tgsi_type_is_64bit(dtype) && (chan_index == 1 || chan_index == 3)) + continue; + emit_store_chan(bld_base, inst, index, chan_index, dst[chan_index]); } } diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c index fb18753a108..d012c19a58a 100644 --- a/src/gallium/drivers/radeonsi/si_shader.c +++ b/src/gallium/drivers/radeonsi/si_shader.c @@ -1139,11 +1139,12 @@ static LLVMValueRef fetch_input_tes( static void store_output_tcs(struct lp_build_tgsi_context *bld_base, const struct tgsi_full_instruction *inst, const struct tgsi_opcode_info *info, + unsigned index, LLVMValueRef dst[4]) { struct si_shader_context *ctx = si_shader_context(bld_base); struct gallivm_state *gallivm = &ctx->gallivm; - const struct tgsi_full_dst_register *reg = &inst->Dst[0]; + const struct tgsi_full_dst_register *reg = &inst->Dst[index]; const struct tgsi_shader_info *sh_info = &ctx->shader->selector->info; unsigned chan_index; LLVMValueRef dw_addr, stride; @@ -1157,7 +1158,7 @@ static void store_output_tcs(struct lp_build_tgsi_context *bld_base, */ if (reg->Register.File != TGSI_FILE_OUTPUT || (dst[0] && LLVMGetTypeKind(LLVMTypeOf(dst[0])) == LLVMVectorTypeKind)) { - si_llvm_emit_store(bld_base, inst, info, dst); + si_llvm_emit_store(bld_base, inst, info, index, dst); return; } @@ -1191,8 +1192,9 @@ static void store_output_tcs(struct lp_build_tgsi_context *bld_base, base = LLVMGetParam(ctx->main_fn, ctx->param_tcs_offchip_offset); buf_addr = get_tcs_tes_buffer_address_from_reg(ctx, reg, NULL); - - TGSI_FOR_EACH_DST0_ENABLED_CHANNEL(inst, chan_index) { + uint32_t writemask = reg->Register.WriteMask; + while (writemask) { + chan_index = u_bit_scan(&writemask); LLVMValueRef value = dst[chan_index]; if (inst->Instruction.Saturate) @@ -1205,7 +1207,7 @@ static void store_output_tcs(struct lp_build_tgsi_context *bld_base, value = LLVMBuildBitCast(gallivm->builder, value, ctx->i32, ""); values[chan_index] = value; - if (inst->Dst[0].Register.WriteMask != 0xF && !is_tess_factor) { + if (reg->Register.WriteMask != 0xF && !is_tess_factor) { ac_build_buffer_store_dword(&ctx->ac, buffer, value, 1, buf_addr, base, 4 * chan_index, 1, 0, true, false); @@ -1224,7 +1226,7 @@ static void store_output_tcs(struct lp_build_tgsi_context *bld_base, } } - if (inst->Dst[0].Register.WriteMask == 0xF && !is_tess_factor) { + if (reg->Register.WriteMask == 0xF && !is_tess_factor) { LLVMValueRef value = lp_build_gather_values(gallivm, values, 4); ac_build_buffer_store_dword(&ctx->ac, buffer, value, 4, buf_addr, diff --git a/src/gallium/drivers/radeonsi/si_shader_internal.h b/src/gallium/drivers/radeonsi/si_shader_internal.h index 023f9a6a093..141dd34be3d 100644 --- a/src/gallium/drivers/radeonsi/si_shader_internal.h +++ b/src/gallium/drivers/radeonsi/si_shader_internal.h @@ -289,6 +289,7 @@ LLVMValueRef si_llvm_emit_fetch(struct lp_build_tgsi_context *bld_base, void si_llvm_emit_store(struct lp_build_tgsi_context *bld_base, const struct tgsi_full_instruction *inst, const struct tgsi_opcode_info *info, + unsigned index, LLVMValueRef dst[4]); /* Combine these with & instead of |. */ diff --git a/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c b/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c index be7a0b97489..f0c7803c997 100644 --- a/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c +++ b/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c @@ -881,14 +881,14 @@ static void emit_declaration(struct lp_build_tgsi_context *bld_base, void si_llvm_emit_store(struct lp_build_tgsi_context *bld_base, const struct tgsi_full_instruction *inst, const struct tgsi_opcode_info *info, + unsigned index, LLVMValueRef dst[4]) { struct si_shader_context *ctx = si_shader_context(bld_base); struct gallivm_state *gallivm = &ctx->gallivm; - const struct tgsi_full_dst_register *reg = &inst->Dst[0]; + const struct tgsi_full_dst_register *reg = &inst->Dst[index]; LLVMBuilderRef builder = ctx->gallivm.builder; LLVMValueRef temp_ptr, temp_ptr2 = NULL; - unsigned chan, chan_index; bool is_vec_store = false; enum tgsi_opcode_type dtype = tgsi_opcode_infer_dst_type(inst->Instruction.Opcode); @@ -899,16 +899,20 @@ void si_llvm_emit_store(struct lp_build_tgsi_context *bld_base, if (is_vec_store) { LLVMValueRef values[4] = {}; - TGSI_FOR_EACH_DST0_ENABLED_CHANNEL(inst, chan) { + uint32_t writemask = reg->Register.WriteMask; + while (writemask) { + unsigned chan = u_bit_scan(&writemask); LLVMValueRef index = LLVMConstInt(ctx->i32, chan, 0); values[chan] = LLVMBuildExtractElement(gallivm->builder, dst[0], index, ""); } - bld_base->emit_store(bld_base, inst, info, values); + bld_base->emit_store(bld_base, inst, info, index, values); return; } - TGSI_FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) { + uint32_t writemask = reg->Register.WriteMask; + while (writemask) { + unsigned chan_index = u_bit_scan(&writemask); LLVMValueRef value = dst[chan_index]; if (tgsi_type_is_64bit(dtype) && (chan_index == 1 || chan_index == 3))