X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fradeonsi%2Fsi_shader_tgsi_alu.c;h=f54d025aec05427ba101ddde4798c40ffa5da2fe;hb=986d6f12fb4bddc1e01ce6f0ee0c2d4a4dbd8b40;hp=3008c3ade1486bf08d65dd73e61fa2dc41f85a67;hpb=965904eebd1d42c00fd3d5d5fe8f3b85a731ff73;p=mesa.git diff --git a/src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c b/src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c index 3008c3ade14..f54d025aec0 100644 --- a/src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c +++ b/src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c @@ -25,42 +25,15 @@ #include "si_shader_internal.h" #include "ac_llvm_util.h" -static void kill_if_fetch_args(struct lp_build_tgsi_context *bld_base, - struct lp_build_emit_data *emit_data) -{ - const struct tgsi_full_instruction *inst = emit_data->inst; - struct si_shader_context *ctx = si_shader_context(bld_base); - LLVMBuilderRef builder = ctx->ac.builder; - unsigned i; - LLVMValueRef conds[TGSI_NUM_CHANNELS]; - - for (i = 0; i < TGSI_NUM_CHANNELS; i++) { - LLVMValueRef value = lp_build_emit_fetch(bld_base, inst, 0, i); - conds[i] = LLVMBuildFCmp(builder, LLVMRealOGE, value, - ctx->ac.f32_0, ""); - } - - /* And the conditions together */ - for (i = TGSI_NUM_CHANNELS - 1; i > 0; i--) { - conds[i - 1] = LLVMBuildAnd(builder, conds[i], conds[i - 1], ""); - } - - emit_data->dst_type = ctx->voidt; - emit_data->arg_count = 1; - emit_data->args[0] = conds[0]; -} - void si_llvm_emit_kill(struct ac_shader_abi *abi, LLVMValueRef visible) { struct si_shader_context *ctx = si_shader_context_from_abi(abi); LLVMBuilderRef builder = ctx->ac.builder; if (ctx->shader->selector->force_correct_derivs_after_kill) { - /* LLVM 6.0 can kill immediately while maintaining WQM. */ - if (HAVE_LLVM >= 0x0600) { - ac_build_kill_if_false(&ctx->ac, - ac_build_wqm_vote(&ctx->ac, visible)); - } + /* Kill immediately while maintaining WQM. */ + ac_build_kill_if_false(&ctx->ac, + ac_build_wqm_vote(&ctx->ac, visible)); LLVMValueRef mask = LLVMBuildLoad(builder, ctx->postponed_kill, ""); mask = LLVMBuildAnd(builder, mask, visible, ""); @@ -79,10 +52,26 @@ static void kil_emit(const struct lp_build_tgsi_action *action, LLVMValueRef visible; if (emit_data->inst->Instruction.Opcode == TGSI_OPCODE_KILL_IF) { - visible = emit_data->args[0]; + const struct tgsi_full_instruction *inst = emit_data->inst; + struct si_shader_context *ctx = si_shader_context(bld_base); + LLVMBuilderRef builder = ctx->ac.builder; + unsigned i; + LLVMValueRef conds[TGSI_NUM_CHANNELS]; + + for (i = 0; i < TGSI_NUM_CHANNELS; i++) { + LLVMValueRef value = lp_build_emit_fetch(bld_base, inst, 0, i); + conds[i] = LLVMBuildFCmp(builder, LLVMRealOGE, value, + ctx->ac.f32_0, ""); + } + + /* And the conditions together */ + for (i = TGSI_NUM_CHANNELS - 1; i > 0; i--) { + conds[i - 1] = LLVMBuildAnd(builder, conds[i], conds[i - 1], ""); + } + visible = conds[0]; } else { assert(emit_data->inst->Instruction.Opcode == TGSI_OPCODE_KILL); - visible = LLVMConstInt(ctx->i1, false, 0); + visible = ctx->i1false; } si_llvm_emit_kill(&ctx->abi, visible); @@ -244,7 +233,9 @@ static void emit_arl(const struct lp_build_tgsi_action *action, struct lp_build_emit_data *emit_data) { struct si_shader_context *ctx = si_shader_context(bld_base); - LLVMValueRef floor_index = lp_build_emit_llvm_unary(bld_base, TGSI_OPCODE_FLR, emit_data->args[0]); + LLVMValueRef floor_index = + ac_build_intrinsic(&ctx->ac, "llvm.floor.f32", ctx->f32, + &emit_data->args[0], 1, AC_FUNC_ATTR_READNONE); emit_data->output[emit_data->chan] = LLVMBuildFPToSI(ctx->ac.builder, floor_index, ctx->i32, ""); } @@ -556,10 +547,8 @@ static void emit_iabs(const struct lp_build_tgsi_action *action, struct si_shader_context *ctx = si_shader_context(bld_base); emit_data->output[emit_data->chan] = - lp_build_emit_llvm_binary(bld_base, TGSI_OPCODE_IMAX, - emit_data->args[0], - LLVMBuildNeg(ctx->ac.builder, - emit_data->args[0], "")); + ac_build_imax(&ctx->ac, emit_data->args[0], + LLVMBuildNeg(ctx->ac.builder, emit_data->args[0], "")); } static void emit_minmax_int(const struct lp_build_tgsi_action *action, @@ -598,34 +587,25 @@ static void emit_minmax_int(const struct lp_build_tgsi_action *action, emit_data->args[1], ""); } -static void pk2h_fetch_args(struct lp_build_tgsi_context *bld_base, - struct lp_build_emit_data *emit_data) -{ - emit_data->args[0] = lp_build_emit_fetch(bld_base, emit_data->inst, - 0, TGSI_CHAN_X); - emit_data->args[1] = lp_build_emit_fetch(bld_base, emit_data->inst, - 0, TGSI_CHAN_Y); -} - static void emit_pk2h(const struct lp_build_tgsi_action *action, struct lp_build_tgsi_context *bld_base, struct lp_build_emit_data *emit_data) { + struct si_shader_context *ctx = si_shader_context(bld_base); + LLVMValueRef v[] = { + lp_build_emit_fetch(bld_base, emit_data->inst, 0, TGSI_CHAN_X), + lp_build_emit_fetch(bld_base, emit_data->inst, 0, TGSI_CHAN_Y), + }; + + /* From the GLSL 4.50 spec: * "The rounding mode cannot be set and is undefined." * * v_cvt_pkrtz_f16 rounds to zero, but it's fastest. */ emit_data->output[emit_data->chan] = - ac_build_cvt_pkrtz_f16(&si_shader_context(bld_base)->ac, - emit_data->args); -} - -static void up2h_fetch_args(struct lp_build_tgsi_context *bld_base, - struct lp_build_emit_data *emit_data) -{ - emit_data->args[0] = lp_build_emit_fetch(bld_base, emit_data->inst, - 0, TGSI_CHAN_X); + LLVMBuildBitCast(ctx->ac.builder, ac_build_cvt_pkrtz_f16(&ctx->ac, v), + ctx->i32, ""); } static void emit_up2h(const struct lp_build_tgsi_action *action, @@ -639,7 +619,7 @@ static void emit_up2h(const struct lp_build_tgsi_action *action, i16 = LLVMInt16TypeInContext(ctx->ac.context); const16 = LLVMConstInt(ctx->i32, 16, 0); - input = emit_data->args[0]; + input = lp_build_emit_fetch(bld_base, emit_data->inst, 0, TGSI_CHAN_X); for (i = 0; i < 2; i++) { val = i == 1 ? LLVMBuildLShr(ctx->ac.builder, input, const16, "") : input; @@ -668,19 +648,11 @@ static void emit_rsq(const struct lp_build_tgsi_action *action, struct si_shader_context *ctx = si_shader_context(bld_base); LLVMValueRef sqrt = - lp_build_emit_llvm_unary(bld_base, TGSI_OPCODE_SQRT, - emit_data->args[0]); + ac_build_intrinsic(&ctx->ac, "llvm.sqrt.f32", ctx->f32, + &emit_data->args[0], 1, AC_FUNC_ATTR_READNONE); emit_data->output[emit_data->chan] = - lp_build_emit_llvm_binary(bld_base, TGSI_OPCODE_DIV, - ctx->ac.f32_1, sqrt); -} - -static void dfracexp_fetch_args(struct lp_build_tgsi_context *bld_base, - struct lp_build_emit_data *emit_data) -{ - emit_data->args[0] = lp_build_emit_fetch(bld_base, emit_data->inst, 0, TGSI_CHAN_X); - emit_data->arg_count = 1; + ac_build_fdiv(&ctx->ac, ctx->ac.f32_1, sqrt); } static void dfracexp_emit(const struct lp_build_tgsi_action *action, @@ -688,13 +660,14 @@ static void dfracexp_emit(const struct lp_build_tgsi_action *action, struct lp_build_emit_data *emit_data) { struct si_shader_context *ctx = si_shader_context(bld_base); + LLVMValueRef in = lp_build_emit_fetch(bld_base, emit_data->inst, 0, TGSI_CHAN_X); emit_data->output[emit_data->chan] = ac_build_intrinsic(&ctx->ac, "llvm.amdgcn.frexp.mant.f64", - ctx->ac.f64, &emit_data->args[0], 1, 0); + ctx->ac.f64, &in, 1, 0); emit_data->output1[emit_data->chan] = ac_build_intrinsic(&ctx->ac, "llvm.amdgcn.frexp.exp.i32.f64", - ctx->ac.i32, &emit_data->args[0], 1, 0); + ctx->ac.i32, &in, 1, 0); } void si_shader_context_init_alu(struct lp_build_tgsi_context *bld_base) @@ -735,7 +708,6 @@ void si_shader_context_init_alu(struct lp_build_tgsi_context *bld_base) bld_base->op_actions[TGSI_OPCODE_DSQRT].intr_name = "llvm.sqrt.f64"; bld_base->op_actions[TGSI_OPCODE_DTRUNC].emit = build_tgsi_intrinsic_nomem; bld_base->op_actions[TGSI_OPCODE_DTRUNC].intr_name = "llvm.trunc.f64"; - bld_base->op_actions[TGSI_OPCODE_DFRACEXP].fetch_args = dfracexp_fetch_args; bld_base->op_actions[TGSI_OPCODE_DFRACEXP].emit = dfracexp_emit; bld_base->op_actions[TGSI_OPCODE_DLDEXP].emit = build_tgsi_intrinsic_nomem; bld_base->op_actions[TGSI_OPCODE_DLDEXP].intr_name = "llvm.amdgcn.ldexp.f64"; @@ -764,7 +736,6 @@ void si_shader_context_init_alu(struct lp_build_tgsi_context *bld_base) bld_base->op_actions[TGSI_OPCODE_ISLT].emit = emit_icmp; bld_base->op_actions[TGSI_OPCODE_ISSG].emit = emit_ssg; bld_base->op_actions[TGSI_OPCODE_I2F].emit = emit_i2f; - bld_base->op_actions[TGSI_OPCODE_KILL_IF].fetch_args = kill_if_fetch_args; bld_base->op_actions[TGSI_OPCODE_KILL_IF].emit = kil_emit; bld_base->op_actions[TGSI_OPCODE_KILL].emit = kil_emit; bld_base->op_actions[TGSI_OPCODE_LDEXP].emit = build_tgsi_intrinsic_nomem; @@ -780,7 +751,6 @@ void si_shader_context_init_alu(struct lp_build_tgsi_context *bld_base) bld_base->op_actions[TGSI_OPCODE_UMSB].emit = emit_umsb; bld_base->op_actions[TGSI_OPCODE_NOT].emit = emit_not; bld_base->op_actions[TGSI_OPCODE_OR].emit = emit_or; - bld_base->op_actions[TGSI_OPCODE_PK2H].fetch_args = pk2h_fetch_args; bld_base->op_actions[TGSI_OPCODE_PK2H].emit = emit_pk2h; bld_base->op_actions[TGSI_OPCODE_POPC].emit = build_tgsi_intrinsic_nomem; bld_base->op_actions[TGSI_OPCODE_POPC].intr_name = "llvm.ctpop.i32"; @@ -817,7 +787,6 @@ void si_shader_context_init_alu(struct lp_build_tgsi_context *bld_base) bld_base->op_actions[TGSI_OPCODE_U2F].emit = emit_u2f; bld_base->op_actions[TGSI_OPCODE_XOR].emit = emit_xor; bld_base->op_actions[TGSI_OPCODE_UCMP].emit = emit_ucmp; - bld_base->op_actions[TGSI_OPCODE_UP2H].fetch_args = up2h_fetch_args; bld_base->op_actions[TGSI_OPCODE_UP2H].emit = emit_up2h; bld_base->op_actions[TGSI_OPCODE_I64MAX].emit = emit_minmax_int;