return bitcast(bld_base, type, cval);
}
+static LLVMValueRef si_build_alloca_undef(struct gallivm_state *gallivm,
+ LLVMTypeRef type,
+ const char *name)
+{
+ LLVMValueRef ptr = lp_build_alloca(gallivm, type, name);
+ LLVMBuildStore(gallivm->builder, LLVMGetUndef(type), ptr);
+ return ptr;
+}
+
static void emit_declaration(
struct lp_build_tgsi_context * bld_base,
const struct tgsi_full_declaration *decl)
for (idx = decl->Range.First; idx <= decl->Range.Last; idx++) {
unsigned chan;
for (chan = 0; chan < TGSI_NUM_CHANNELS; chan++) {
- ctx->soa.addr[idx][chan] = lp_build_alloca(
+ ctx->soa.addr[idx][chan] = si_build_alloca_undef(
&ctx->gallivm,
ctx->soa.bld_base.uint_bld.elem_type, "");
}
for (idx = first; idx <= last; idx++) {
for (i = 0; i < TGSI_NUM_CHANNELS; i++) {
ctx->temps[idx * TGSI_NUM_CHANNELS + i] =
- lp_build_alloca(bld_base->base.gallivm, bld_base->base.vec_type,
- "temp");
+ si_build_alloca_undef(bld_base->base.gallivm,
+ bld_base->base.vec_type,
+ "temp");
}
}
break;
unsigned chan;
assert(idx < RADEON_LLVM_MAX_OUTPUTS);
for (chan = 0; chan < TGSI_NUM_CHANNELS; chan++) {
- ctx->soa.outputs[idx][chan] = lp_build_alloca(&ctx->gallivm,
+ ctx->soa.outputs[idx][chan] = si_build_alloca_undef(
+ &ctx->gallivm,
ctx->soa.bld_base.base.elem_type, "");
}
}
coords[i] = LLVMBuildExtractElement(builder, v,
lp_build_const_int32(gallivm, i), "");
- coords[2] = lp_build_intrinsic(builder, "fabs",
+ coords[2] = lp_build_intrinsic(builder, "llvm.fabs.f32",
type, &coords[2], 1, LLVMReadNoneAttribute);
coords[2] = lp_build_emit_llvm_unary(bld_base, TGSI_OPCODE_RCP, coords[2]);
memcpy(coords_arg, coords, sizeof(coords));
}
-static void txd_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;
-
- LLVMValueRef coords[4];
- unsigned chan, src;
- for (src = 0; src < 3; src++) {
- for (chan = 0; chan < 4; chan++)
- coords[chan] = lp_build_emit_fetch(bld_base, inst, src, chan);
-
- emit_data->args[src] = lp_build_gather_values(bld_base->base.gallivm,
- coords, 4);
- }
- emit_data->arg_count = 3;
- emit_data->dst_type = LLVMVectorType(bld_base->base.elem_type, 4);
-}
-
-
-static void txp_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;
- LLVMValueRef src_w;
- unsigned chan;
- LLVMValueRef coords[5];
-
- emit_data->dst_type = LLVMVectorType(bld_base->base.elem_type, 4);
- src_w = lp_build_emit_fetch(bld_base, emit_data->inst, 0, TGSI_CHAN_W);
-
- for (chan = 0; chan < 3; chan++ ) {
- LLVMValueRef arg = lp_build_emit_fetch(bld_base,
- emit_data->inst, 0, chan);
- coords[chan] = lp_build_emit_llvm_binary(bld_base,
- TGSI_OPCODE_DIV, arg, src_w);
- }
- coords[3] = bld_base->base.one;
-
- if ((inst->Texture.Texture == TGSI_TEXTURE_CUBE ||
- inst->Texture.Texture == TGSI_TEXTURE_CUBE_ARRAY ||
- inst->Texture.Texture == TGSI_TEXTURE_SHADOWCUBE ||
- inst->Texture.Texture == TGSI_TEXTURE_SHADOWCUBE_ARRAY) &&
- inst->Instruction.Opcode != TGSI_OPCODE_TXQ &&
- inst->Instruction.Opcode != TGSI_OPCODE_TXQ_LZ) {
- radeon_llvm_emit_prepare_cube_coords(bld_base, emit_data, coords, NULL);
- }
-
- emit_data->args[0] = lp_build_gather_values(bld_base->base.gallivm,
- coords, 4);
- emit_data->arg_count = 1;
-}
-
-static void tex_fetch_args(
- struct lp_build_tgsi_context * bld_base,
- struct lp_build_emit_data * emit_data)
-{
- /* XXX: lp_build_swizzle_aos() was failing with wrong arg types,
- * when we used CHAN_ALL. We should be able to get this to work,
- * but for now we will swizzle it ourselves
- emit_data->args[0] = lp_build_emit_fetch(bld_base, emit_data->inst,
- 0, CHAN_ALL);
-
- */
-
- const struct tgsi_full_instruction * inst = emit_data->inst;
-
- LLVMValueRef coords[5];
- unsigned chan;
- for (chan = 0; chan < 4; chan++) {
- coords[chan] = lp_build_emit_fetch(bld_base, inst, 0, chan);
- }
-
- if (inst->Instruction.Opcode == TGSI_OPCODE_TEX2 ||
- inst->Instruction.Opcode == TGSI_OPCODE_TXB2 ||
- inst->Instruction.Opcode == TGSI_OPCODE_TXL2) {
- /* These instructions have additional operand that should be packed
- * into the cube coord vector by radeon_llvm_emit_prepare_cube_coords.
- * That operand should be passed as a float value in the args array
- * right after the coord vector. After packing it's not used anymore,
- * that's why arg_count is not increased */
- coords[4] = lp_build_emit_fetch(bld_base, inst, 1, 0);
- }
-
- if ((inst->Texture.Texture == TGSI_TEXTURE_CUBE ||
- inst->Texture.Texture == TGSI_TEXTURE_CUBE_ARRAY ||
- inst->Texture.Texture == TGSI_TEXTURE_SHADOWCUBE ||
- inst->Texture.Texture == TGSI_TEXTURE_SHADOWCUBE_ARRAY) &&
- inst->Instruction.Opcode != TGSI_OPCODE_TXQ &&
- inst->Instruction.Opcode != TGSI_OPCODE_TXQ_LZ) {
- radeon_llvm_emit_prepare_cube_coords(bld_base, emit_data, coords, NULL);
- }
-
- emit_data->arg_count = 1;
- emit_data->args[0] = lp_build_gather_values(bld_base->base.gallivm,
- coords, 4);
- emit_data->dst_type = LLVMVectorType(bld_base->base.elem_type, 4);
-}
-
-static void txf_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 lp_build_tgsi_soa_context *bld = lp_soa_context(bld_base);
- const struct tgsi_texture_offset * off = inst->TexOffsets;
- LLVMTypeRef offset_type = bld_base->int_bld.elem_type;
-
- /* fetch tex coords */
- tex_fetch_args(bld_base, emit_data);
-
- /* fetch tex offsets */
- if (inst->Texture.NumOffsets) {
- assert(inst->Texture.NumOffsets == 1);
-
- emit_data->args[1] = LLVMConstBitCast(
- bld->immediates[off->Index][off->SwizzleX],
- offset_type);
- emit_data->args[2] = LLVMConstBitCast(
- bld->immediates[off->Index][off->SwizzleY],
- offset_type);
- emit_data->args[3] = LLVMConstBitCast(
- bld->immediates[off->Index][off->SwizzleZ],
- offset_type);
- } else {
- emit_data->args[1] = bld_base->int_bld.zero;
- emit_data->args[2] = bld_base->int_bld.zero;
- emit_data->args[3] = bld_base->int_bld.zero;
- }
-
- emit_data->arg_count = 4;
-}
-
static void emit_icmp(
const struct lp_build_tgsi_action * action,
struct lp_build_tgsi_context * bld_base,
LLVMBuildSelect(builder, v, emit_data->args[1], emit_data->args[2], "");
}
-static void emit_cmp(
+static void emit_cmp(const struct lp_build_tgsi_action *action,
+ struct lp_build_tgsi_context *bld_base,
+ struct lp_build_emit_data *emit_data)
+{
+ LLVMBuilderRef builder = bld_base->base.gallivm->builder;
+ LLVMValueRef cond, *args = emit_data->args;
+
+ cond = LLVMBuildFCmp(builder, LLVMRealOLT, args[0],
+ bld_base->base.zero, "");
+
+ emit_data->output[emit_data->chan] =
+ LLVMBuildSelect(builder, cond, args[1], args[2], "");
+}
+
+static void emit_set_cond(
const struct lp_build_tgsi_action *action,
struct lp_build_tgsi_context * bld_base,
struct lp_build_emit_data * emit_data)
emit_data->args[0], "");
}
+static void emit_frac(
+ const struct lp_build_tgsi_action * action,
+ struct lp_build_tgsi_context * bld_base,
+ struct lp_build_emit_data * emit_data)
+{
+ LLVMBuilderRef builder = bld_base->base.gallivm->builder;
+ char *intr;
+
+ if (emit_data->info->opcode == TGSI_OPCODE_FRC)
+ intr = "llvm.floor.f32";
+ else if (emit_data->info->opcode == TGSI_OPCODE_DFRAC)
+ intr = "llvm.floor.f64";
+ else {
+ assert(0);
+ return;
+ }
+
+ LLVMValueRef floor = lp_build_intrinsic(builder, intr, emit_data->dst_type,
+ &emit_data->args[0], 1,
+ LLVMReadNoneAttribute);
+ emit_data->output[emit_data->chan] = LLVMBuildFSub(builder,
+ emit_data->args[0], floor, "");
+}
+
static void emit_f2i(
const struct lp_build_tgsi_action * action,
struct lp_build_tgsi_context * bld_base,
LLVMBuildSelect(builder, cond, all_ones, msb, "");
}
+static void emit_iabs(const struct lp_build_tgsi_action *action,
+ struct lp_build_tgsi_context *bld_base,
+ struct lp_build_emit_data *emit_data)
+{
+ LLVMBuilderRef builder = bld_base->base.gallivm->builder;
+
+ emit_data->output[emit_data->chan] =
+ lp_build_emit_llvm_binary(bld_base, TGSI_OPCODE_IMAX,
+ emit_data->args[0],
+ LLVMBuildNeg(builder,
+ emit_data->args[0], ""));
+}
+
+static void emit_minmax_int(const struct lp_build_tgsi_action *action,
+ struct lp_build_tgsi_context *bld_base,
+ struct lp_build_emit_data *emit_data)
+{
+ LLVMBuilderRef builder = bld_base->base.gallivm->builder;
+ LLVMIntPredicate op;
+
+ switch (emit_data->info->opcode) {
+ default:
+ assert(0);
+ case TGSI_OPCODE_IMAX:
+ op = LLVMIntSGT;
+ break;
+ case TGSI_OPCODE_IMIN:
+ op = LLVMIntSLT;
+ break;
+ case TGSI_OPCODE_UMAX:
+ op = LLVMIntUGT;
+ break;
+ case TGSI_OPCODE_UMIN:
+ op = LLVMIntULT;
+ break;
+ }
+
+ emit_data->output[emit_data->chan] =
+ LLVMBuildSelect(builder,
+ LLVMBuildICmp(builder, op, emit_data->args[0],
+ emit_data->args[1], ""),
+ emit_data->args[0],
+ emit_data->args[1], "");
+}
+
void radeon_llvm_context_init(struct radeon_llvm_context * ctx)
{
struct lp_type type;
ctx->gallivm.context);
ctx->gallivm.builder = LLVMCreateBuilderInContext(ctx->gallivm.context);
- ctx->store_output_intr = "llvm.AMDGPU.store.output.";
- ctx->swizzle_intr = "llvm.AMDGPU.swizzle";
struct lp_build_tgsi_context * bld_base = &ctx->soa.bld_base;
- /* XXX: We need to revisit this.I think the correct way to do this is
- * to use length = 4 here and use the elem_bld for everything. */
type.floating = TRUE;
type.fixed = FALSE;
type.sign = TRUE;
/* Allocate outputs */
ctx->soa.outputs = ctx->outputs;
- /* XXX: Is there a better way to initialize all this ? */
-
lp_set_default_actions(bld_base);
bld_base->op_actions[TGSI_OPCODE_ABS].emit = build_tgsi_intrinsic_nomem;
- bld_base->op_actions[TGSI_OPCODE_ABS].intr_name = "fabs";
+ bld_base->op_actions[TGSI_OPCODE_ABS].intr_name = "llvm.fabs.f32";
bld_base->op_actions[TGSI_OPCODE_AND].emit = emit_and;
bld_base->op_actions[TGSI_OPCODE_ARL].emit = emit_arl;
bld_base->op_actions[TGSI_OPCODE_BFI].emit = emit_bfi;
bld_base->op_actions[TGSI_OPCODE_BGNLOOP].emit = bgnloop_emit;
bld_base->op_actions[TGSI_OPCODE_BREV].emit = build_tgsi_intrinsic_nomem;
- bld_base->op_actions[TGSI_OPCODE_BREV].intr_name = "llvm.AMDGPU.brev";
+ bld_base->op_actions[TGSI_OPCODE_BREV].intr_name =
+ HAVE_LLVM >= 0x0308 ? "llvm.bitreverse.i32" : "llvm.AMDGPU.brev";
bld_base->op_actions[TGSI_OPCODE_BRK].emit = brk_emit;
bld_base->op_actions[TGSI_OPCODE_CEIL].emit = build_tgsi_intrinsic_nomem;
- bld_base->op_actions[TGSI_OPCODE_CEIL].intr_name = "ceil";
+ bld_base->op_actions[TGSI_OPCODE_CEIL].intr_name = "llvm.ceil.f32";
bld_base->op_actions[TGSI_OPCODE_CLAMP].emit = build_tgsi_intrinsic_nomem;
- bld_base->op_actions[TGSI_OPCODE_CLAMP].intr_name = "llvm.AMDIL.clamp.";
- bld_base->op_actions[TGSI_OPCODE_CMP].emit = build_tgsi_intrinsic_nomem;
- bld_base->op_actions[TGSI_OPCODE_CMP].intr_name = "llvm.AMDGPU.cndlt";
+ bld_base->op_actions[TGSI_OPCODE_CLAMP].intr_name =
+ HAVE_LLVM >= 0x0308 ? "llvm.AMDGPU.clamp." : "llvm.AMDIL.clamp.";
+ bld_base->op_actions[TGSI_OPCODE_CMP].emit = emit_cmp;
bld_base->op_actions[TGSI_OPCODE_CONT].emit = cont_emit;
bld_base->op_actions[TGSI_OPCODE_COS].emit = build_tgsi_intrinsic_nomem;
bld_base->op_actions[TGSI_OPCODE_COS].intr_name = "llvm.cos.f32";
bld_base->op_actions[TGSI_OPCODE_DABS].emit = build_tgsi_intrinsic_nomem;
- bld_base->op_actions[TGSI_OPCODE_DABS].intr_name = "fabs";
+ bld_base->op_actions[TGSI_OPCODE_DABS].intr_name = "llvm.fabs.f64";
bld_base->op_actions[TGSI_OPCODE_DFMA].emit = build_tgsi_intrinsic_nomem;
bld_base->op_actions[TGSI_OPCODE_DFMA].intr_name = "llvm.fma.f64";
- bld_base->op_actions[TGSI_OPCODE_DFRAC].emit = build_tgsi_intrinsic_nomem;
- bld_base->op_actions[TGSI_OPCODE_DFRAC].intr_name = "llvm.AMDIL.fraction.";
+ bld_base->op_actions[TGSI_OPCODE_DFRAC].emit = emit_frac;
bld_base->op_actions[TGSI_OPCODE_DNEG].emit = emit_dneg;
bld_base->op_actions[TGSI_OPCODE_DSEQ].emit = emit_dcmp;
bld_base->op_actions[TGSI_OPCODE_DSGE].emit = emit_dcmp;
bld_base->op_actions[TGSI_OPCODE_DSLT].emit = emit_dcmp;
bld_base->op_actions[TGSI_OPCODE_DSNE].emit = emit_dcmp;
- bld_base->op_actions[TGSI_OPCODE_DDX].intr_name = "llvm.AMDGPU.ddx";
- bld_base->op_actions[TGSI_OPCODE_DDX].fetch_args = tex_fetch_args;
- bld_base->op_actions[TGSI_OPCODE_DDY].intr_name = "llvm.AMDGPU.ddy";
- bld_base->op_actions[TGSI_OPCODE_DDY].fetch_args = tex_fetch_args;
bld_base->op_actions[TGSI_OPCODE_DRSQ].emit = build_tgsi_intrinsic_nomem;
bld_base->op_actions[TGSI_OPCODE_DRSQ].intr_name = "llvm.AMDGPU.rsq.f64";
bld_base->op_actions[TGSI_OPCODE_DSQRT].emit = build_tgsi_intrinsic_nomem;
bld_base->op_actions[TGSI_OPCODE_ENDIF].emit = endif_emit;
bld_base->op_actions[TGSI_OPCODE_ENDLOOP].emit = endloop_emit;
bld_base->op_actions[TGSI_OPCODE_EX2].emit = build_tgsi_intrinsic_nomem;
- bld_base->op_actions[TGSI_OPCODE_EX2].intr_name = "llvm.AMDIL.exp.";
+ bld_base->op_actions[TGSI_OPCODE_EX2].intr_name =
+ HAVE_LLVM >= 0x0308 ? "llvm.exp2.f32" : "llvm.AMDIL.exp.";
bld_base->op_actions[TGSI_OPCODE_FLR].emit = build_tgsi_intrinsic_nomem;
- bld_base->op_actions[TGSI_OPCODE_FLR].intr_name = "floor";
+ bld_base->op_actions[TGSI_OPCODE_FLR].intr_name = "llvm.floor.f32";
bld_base->op_actions[TGSI_OPCODE_FMA].emit = build_tgsi_intrinsic_nomem;
bld_base->op_actions[TGSI_OPCODE_FMA].intr_name = "llvm.fma.f32";
- bld_base->op_actions[TGSI_OPCODE_FRC].emit = build_tgsi_intrinsic_nomem;
- bld_base->op_actions[TGSI_OPCODE_FRC].intr_name = "llvm.AMDIL.fraction.";
+ bld_base->op_actions[TGSI_OPCODE_FRC].emit = emit_frac;
bld_base->op_actions[TGSI_OPCODE_F2I].emit = emit_f2i;
bld_base->op_actions[TGSI_OPCODE_F2U].emit = emit_f2u;
bld_base->op_actions[TGSI_OPCODE_FSEQ].emit = emit_fcmp;
bld_base->op_actions[TGSI_OPCODE_FSGE].emit = emit_fcmp;
bld_base->op_actions[TGSI_OPCODE_FSLT].emit = emit_fcmp;
bld_base->op_actions[TGSI_OPCODE_FSNE].emit = emit_fcmp;
- bld_base->op_actions[TGSI_OPCODE_IABS].emit = build_tgsi_intrinsic_nomem;
- bld_base->op_actions[TGSI_OPCODE_IABS].intr_name = "llvm.AMDIL.abs.";
+ bld_base->op_actions[TGSI_OPCODE_IABS].emit = emit_iabs;
bld_base->op_actions[TGSI_OPCODE_IBFE].emit = build_tgsi_intrinsic_nomem;
bld_base->op_actions[TGSI_OPCODE_IBFE].intr_name = "llvm.AMDGPU.bfe.i32";
bld_base->op_actions[TGSI_OPCODE_IDIV].emit = emit_idiv;
bld_base->op_actions[TGSI_OPCODE_IF].emit = if_emit;
bld_base->op_actions[TGSI_OPCODE_UIF].emit = uif_emit;
- bld_base->op_actions[TGSI_OPCODE_IMAX].emit = build_tgsi_intrinsic_nomem;
- bld_base->op_actions[TGSI_OPCODE_IMAX].intr_name = "llvm.AMDGPU.imax";
- bld_base->op_actions[TGSI_OPCODE_IMIN].emit = build_tgsi_intrinsic_nomem;
- bld_base->op_actions[TGSI_OPCODE_IMIN].intr_name = "llvm.AMDGPU.imin";
+ bld_base->op_actions[TGSI_OPCODE_IMAX].emit = emit_minmax_int;
+ bld_base->op_actions[TGSI_OPCODE_IMIN].emit = emit_minmax_int;
bld_base->op_actions[TGSI_OPCODE_IMSB].emit = emit_imsb;
bld_base->op_actions[TGSI_OPCODE_INEG].emit = emit_ineg;
bld_base->op_actions[TGSI_OPCODE_ISHR].emit = emit_ishr;
bld_base->op_actions[TGSI_OPCODE_LSB].emit = emit_lsb;
bld_base->op_actions[TGSI_OPCODE_LG2].emit = build_tgsi_intrinsic_nomem;
bld_base->op_actions[TGSI_OPCODE_LG2].intr_name = "llvm.log2.f32";
- bld_base->op_actions[TGSI_OPCODE_LRP].emit = build_tgsi_intrinsic_nomem;
- bld_base->op_actions[TGSI_OPCODE_LRP].intr_name = "llvm.AMDGPU.lrp";
bld_base->op_actions[TGSI_OPCODE_MOD].emit = emit_mod;
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_POW].emit = build_tgsi_intrinsic_nomem;
bld_base->op_actions[TGSI_OPCODE_POW].intr_name = "llvm.pow.f32";
bld_base->op_actions[TGSI_OPCODE_ROUND].emit = build_tgsi_intrinsic_nomem;
- bld_base->op_actions[TGSI_OPCODE_ROUND].intr_name = "llvm.AMDIL.round.nearest.";
- bld_base->op_actions[TGSI_OPCODE_SGE].emit = emit_cmp;
- bld_base->op_actions[TGSI_OPCODE_SEQ].emit = emit_cmp;
+ bld_base->op_actions[TGSI_OPCODE_ROUND].intr_name = "llvm.rint.f32";
+ bld_base->op_actions[TGSI_OPCODE_RSQ].intr_name = "llvm.AMDGPU.rsq.clamped.f32";
+ bld_base->op_actions[TGSI_OPCODE_RSQ].emit = build_tgsi_intrinsic_nomem;
+ bld_base->op_actions[TGSI_OPCODE_SGE].emit = emit_set_cond;
+ bld_base->op_actions[TGSI_OPCODE_SEQ].emit = emit_set_cond;
bld_base->op_actions[TGSI_OPCODE_SHL].emit = emit_shl;
- bld_base->op_actions[TGSI_OPCODE_SLE].emit = emit_cmp;
- bld_base->op_actions[TGSI_OPCODE_SLT].emit = emit_cmp;
- bld_base->op_actions[TGSI_OPCODE_SNE].emit = emit_cmp;
- bld_base->op_actions[TGSI_OPCODE_SGT].emit = emit_cmp;
+ bld_base->op_actions[TGSI_OPCODE_SLE].emit = emit_set_cond;
+ bld_base->op_actions[TGSI_OPCODE_SLT].emit = emit_set_cond;
+ bld_base->op_actions[TGSI_OPCODE_SNE].emit = emit_set_cond;
+ bld_base->op_actions[TGSI_OPCODE_SGT].emit = emit_set_cond;
bld_base->op_actions[TGSI_OPCODE_SIN].emit = build_tgsi_intrinsic_nomem;
bld_base->op_actions[TGSI_OPCODE_SIN].intr_name = "llvm.sin.f32";
bld_base->op_actions[TGSI_OPCODE_SQRT].emit = build_tgsi_intrinsic_nomem;
bld_base->op_actions[TGSI_OPCODE_SQRT].intr_name = "llvm.sqrt.f32";
bld_base->op_actions[TGSI_OPCODE_SSG].emit = emit_ssg;
- bld_base->op_actions[TGSI_OPCODE_TEX].fetch_args = tex_fetch_args;
- bld_base->op_actions[TGSI_OPCODE_TEX].intr_name = "llvm.AMDGPU.tex";
- bld_base->op_actions[TGSI_OPCODE_TEX2].fetch_args = tex_fetch_args;
- bld_base->op_actions[TGSI_OPCODE_TEX2].intr_name = "llvm.AMDGPU.tex";
- bld_base->op_actions[TGSI_OPCODE_TXB].fetch_args = tex_fetch_args;
- bld_base->op_actions[TGSI_OPCODE_TXB].intr_name = "llvm.AMDGPU.txb";
- bld_base->op_actions[TGSI_OPCODE_TXB2].fetch_args = tex_fetch_args;
- bld_base->op_actions[TGSI_OPCODE_TXB2].intr_name = "llvm.AMDGPU.txb";
- bld_base->op_actions[TGSI_OPCODE_TXD].fetch_args = txd_fetch_args;
- bld_base->op_actions[TGSI_OPCODE_TXD].intr_name = "llvm.AMDGPU.txd";
- bld_base->op_actions[TGSI_OPCODE_TXF].fetch_args = txf_fetch_args;
- bld_base->op_actions[TGSI_OPCODE_TXF].intr_name = "llvm.AMDGPU.txf";
- bld_base->op_actions[TGSI_OPCODE_TXL].fetch_args = tex_fetch_args;
- bld_base->op_actions[TGSI_OPCODE_TXL].intr_name = "llvm.AMDGPU.txl";
- bld_base->op_actions[TGSI_OPCODE_TXL2].fetch_args = tex_fetch_args;
- bld_base->op_actions[TGSI_OPCODE_TXL2].intr_name = "llvm.AMDGPU.txl";
- bld_base->op_actions[TGSI_OPCODE_TXP].fetch_args = txp_fetch_args;
- bld_base->op_actions[TGSI_OPCODE_TXP].intr_name = "llvm.AMDGPU.tex";
- bld_base->op_actions[TGSI_OPCODE_TXQ].fetch_args = tex_fetch_args;
- bld_base->op_actions[TGSI_OPCODE_TXQ].intr_name = "llvm.AMDGPU.txq";
bld_base->op_actions[TGSI_OPCODE_TRUNC].emit = build_tgsi_intrinsic_nomem;
- bld_base->op_actions[TGSI_OPCODE_TRUNC].intr_name = "llvm.AMDGPU.trunc";
+ bld_base->op_actions[TGSI_OPCODE_TRUNC].intr_name = "llvm.trunc.f32";
bld_base->op_actions[TGSI_OPCODE_UADD].emit = emit_uadd;
bld_base->op_actions[TGSI_OPCODE_UBFE].emit = build_tgsi_intrinsic_nomem;
bld_base->op_actions[TGSI_OPCODE_UBFE].intr_name = "llvm.AMDGPU.bfe.u32";
bld_base->op_actions[TGSI_OPCODE_UDIV].emit = emit_udiv;
- bld_base->op_actions[TGSI_OPCODE_UMAX].emit = build_tgsi_intrinsic_nomem;
- bld_base->op_actions[TGSI_OPCODE_UMAX].intr_name = "llvm.AMDGPU.umax";
- bld_base->op_actions[TGSI_OPCODE_UMIN].emit = build_tgsi_intrinsic_nomem;
- bld_base->op_actions[TGSI_OPCODE_UMIN].intr_name = "llvm.AMDGPU.umin";
+ bld_base->op_actions[TGSI_OPCODE_UMAX].emit = emit_minmax_int;
+ bld_base->op_actions[TGSI_OPCODE_UMIN].emit = emit_minmax_int;
bld_base->op_actions[TGSI_OPCODE_UMOD].emit = emit_umod;
bld_base->op_actions[TGSI_OPCODE_USEQ].emit = emit_icmp;
bld_base->op_actions[TGSI_OPCODE_USGE].emit = emit_icmp;
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->rsq_action.emit = build_tgsi_intrinsic_nomem;
-#if HAVE_LLVM >= 0x0305
- bld_base->rsq_action.intr_name = "llvm.AMDGPU.rsq.clamped.f32";
-#else
- bld_base->rsq_action.intr_name = "llvm.AMDGPU.rsq";
-#endif
}
void radeon_llvm_create_func(struct radeon_llvm_context * ctx,
projects / mesa.git / blobdiff