LLVMValueRef linear_sample, linear_center, linear_centroid;
LLVMValueRef front_face;
LLVMValueRef ancillary;
+ LLVMValueRef sample_coverage;
LLVMValueRef frag_pos[4];
LLVMBasicBlockRef continue_block;
LLVMTypeRef v16i8;
LLVMTypeRef voidt;
+ LLVMValueRef i1true;
+ LLVMValueRef i1false;
LLVMValueRef i32zero;
LLVMValueRef i32one;
LLVMValueRef f32zero;
int num_locals;
LLVMValueRef *locals;
bool has_ddxy;
- unsigned num_clips;
- unsigned num_culls;
+ uint8_t num_output_clips;
+ uint8_t num_output_culls;
bool has_ds_bpermute;
unsigned gs_max_out_vertices;
};
-struct ac_tex_info {
- LLVMValueRef args[12];
- int arg_count;
- LLVMTypeRef dst_type;
- bool has_offset;
-};
-
static LLVMValueRef get_sampler_desc(struct nir_to_llvm_context *ctx,
nir_deref_var *deref,
enum desc_type desc_type);
return 1;
if (slot == VARYING_SLOT_CLIP_DIST0)
return 2;
- if (slot == VARYING_SLOT_CLIP_DIST1)
- return 3;
+ /* 3 is reserved for clip dist as well */
if (slot >= VARYING_SLOT_VAR0 && slot <= VARYING_SLOT_VAR31)
return 4 + (slot - VARYING_SLOT_VAR0);
unreachable("illegal slot in get unique index\n");
for (unsigned i = 0; i < sgpr_params; ++i) {
if (array_params_mask & (1 << i)) {
LLVMValueRef P = LLVMGetParam(main_function, i);
- ac_add_function_attr(main_function, i + 1, AC_FUNC_ATTR_BYVAL);
+ ac_add_function_attr(ctx, main_function, i + 1, AC_FUNC_ATTR_BYVAL);
ac_add_attr_dereferenceable(P, UINT64_MAX);
}
else {
- ac_add_function_attr(main_function, i + 1, AC_FUNC_ATTR_INREG);
+ ac_add_function_attr(ctx, main_function, i + 1, AC_FUNC_ATTR_INREG);
}
}
LLVMValueRef ptr;
int addr_space;
- offset = LLVMConstInt(ctx->i32, idx, false);
+ offset = LLVMConstInt(ctx->i32, idx * 16, false);
ptr = ctx->shared_memory;
ptr = LLVMBuildGEP(ctx->builder, ptr, &offset, 1, "");
arg_types[arg_idx++] = ctx->f32; /* pos w float */
arg_types[arg_idx++] = ctx->i32; /* front face */
arg_types[arg_idx++] = ctx->i32; /* ancillary */
- arg_types[arg_idx++] = ctx->f32; /* sample coverage */
+ arg_types[arg_idx++] = ctx->i32; /* sample coverage */
arg_types[arg_idx++] = ctx->i32; /* fixed pt */
break;
default:
set_userdata_location_shader(ctx, AC_UD_SCRATCH_RING_OFFSETS, user_sgpr_idx, 2);
user_sgpr_idx += 2;
if (ctx->options->supports_spill) {
- ctx->ring_offsets = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.amdgcn.implicit.buffer.ptr",
- LLVMPointerType(ctx->i8, CONST_ADDR_SPACE),
- NULL, 0, AC_FUNC_ATTR_READNONE);
+ ctx->ring_offsets = ac_build_intrinsic(&ctx->ac, "llvm.amdgcn.implicit.buffer.ptr",
+ LLVMPointerType(ctx->i8, CONST_ADDR_SPACE),
+ NULL, 0, AC_FUNC_ATTR_READNONE);
ctx->ring_offsets = LLVMBuildBitCast(ctx->builder, ctx->ring_offsets,
const_array(ctx->v16i8, 8), "");
} else
ctx->frag_pos[3] = LLVMGetParam(ctx->main_function, arg_idx++);
ctx->front_face = LLVMGetParam(ctx->main_function, arg_idx++);
ctx->ancillary = LLVMGetParam(ctx->main_function, arg_idx++);
+ ctx->sample_coverage = LLVMGetParam(ctx->main_function, arg_idx++);
break;
default:
unreachable("Shader stage not implemented");
ctx->v4f32 = LLVMVectorType(ctx->f32, 4);
ctx->v16i8 = LLVMVectorType(ctx->i8, 16);
+ ctx->i1false = LLVMConstInt(ctx->i1, 0, false);
+ ctx->i1true = LLVMConstInt(ctx->i1, 1, false);
ctx->i32zero = LLVMConstInt(ctx->i32, 0, false);
ctx->i32one = LLVMConstInt(ctx->i32, 1, false);
ctx->f32zero = LLVMConstReal(ctx->f32, 0.0);
};
sprintf(name, "%s.f%d", intrin, get_elem_bits(ctx, result_type));
- return ac_emit_llvm_intrinsic(&ctx->ac, name, result_type, params, 1, AC_FUNC_ATTR_READNONE);
+ return ac_build_intrinsic(&ctx->ac, name, result_type, params, 1, AC_FUNC_ATTR_READNONE);
}
static LLVMValueRef emit_intrin_2f_param(struct nir_to_llvm_context *ctx,
};
sprintf(name, "%s.f%d", intrin, get_elem_bits(ctx, result_type));
- return ac_emit_llvm_intrinsic(&ctx->ac, name, result_type, params, 2, AC_FUNC_ATTR_READNONE);
+ return ac_build_intrinsic(&ctx->ac, name, result_type, params, 2, AC_FUNC_ATTR_READNONE);
}
static LLVMValueRef emit_intrin_3f_param(struct nir_to_llvm_context *ctx,
};
sprintf(name, "%s.f%d", intrin, get_elem_bits(ctx, result_type));
- return ac_emit_llvm_intrinsic(&ctx->ac, name, result_type, params, 3, AC_FUNC_ATTR_READNONE);
+ return ac_build_intrinsic(&ctx->ac, name, result_type, params, 3, AC_FUNC_ATTR_READNONE);
}
static LLVMValueRef emit_bcsel(struct nir_to_llvm_context *ctx,
*/
LLVMConstInt(ctx->i32, 1, false),
};
- return ac_emit_llvm_intrinsic(&ctx->ac, "llvm.cttz.i32", ctx->i32, params, 2, AC_FUNC_ATTR_READNONE);
+ return ac_build_intrinsic(&ctx->ac, "llvm.cttz.i32", ctx->i32, params, 2, AC_FUNC_ATTR_READNONE);
}
static LLVMValueRef emit_ifind_msb(struct nir_to_llvm_context *ctx,
LLVMValueRef src0)
{
- LLVMValueRef msb = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.AMDGPU.flbit.i32",
- ctx->i32, &src0, 1,
- AC_FUNC_ATTR_READNONE);
-
- /* The HW returns the last bit index from MSB, but NIR wants
- * the index from LSB. Invert it by doing "31 - msb". */
- msb = LLVMBuildSub(ctx->builder, LLVMConstInt(ctx->i32, 31, false),
- msb, "");
-
- LLVMValueRef all_ones = LLVMConstInt(ctx->i32, -1, true);
- LLVMValueRef cond = LLVMBuildOr(ctx->builder,
- LLVMBuildICmp(ctx->builder, LLVMIntEQ,
- src0, ctx->i32zero, ""),
- LLVMBuildICmp(ctx->builder, LLVMIntEQ,
- src0, all_ones, ""), "");
-
- return LLVMBuildSelect(ctx->builder, cond, all_ones, msb, "");
+ return ac_build_imsb(&ctx->ac, src0, ctx->i32);
}
static LLVMValueRef emit_ufind_msb(struct nir_to_llvm_context *ctx,
LLVMValueRef src0)
{
- LLVMValueRef args[2] = {
- src0,
- ctx->i32one,
- };
- LLVMValueRef msb = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.ctlz.i32",
- ctx->i32, args, ARRAY_SIZE(args),
- AC_FUNC_ATTR_READNONE);
-
- /* The HW returns the last bit index from MSB, but NIR wants
- * the index from LSB. Invert it by doing "31 - msb". */
- msb = LLVMBuildSub(ctx->builder, LLVMConstInt(ctx->i32, 31, false),
- msb, "");
-
- return LLVMBuildSelect(ctx->builder,
- LLVMBuildICmp(ctx->builder, LLVMIntEQ, src0,
- ctx->i32zero, ""),
- LLVMConstInt(ctx->i32, -1, true), msb, "");
+ return ac_build_umsb(&ctx->ac, src0, ctx->i32);
}
static LLVMValueRef emit_minmax_int(struct nir_to_llvm_context *ctx,
LLVMValueRef params[] = {
fsrc0,
};
- LLVMValueRef floor = ac_emit_llvm_intrinsic(&ctx->ac, intr,
- ctx->f32, params, 1,
- AC_FUNC_ATTR_READNONE);
+ LLVMValueRef floor = ac_build_intrinsic(&ctx->ac, intr,
+ ctx->f32, params, 1,
+ AC_FUNC_ATTR_READNONE);
return LLVMBuildFSub(ctx->builder, fsrc0, floor, "");
}
ret_type = LLVMStructTypeInContext(ctx->context, types,
2, true);
- res = ac_emit_llvm_intrinsic(&ctx->ac, intrin, ret_type,
- params, 2, AC_FUNC_ATTR_READNONE);
+ res = ac_build_intrinsic(&ctx->ac, intrin, ret_type,
+ params, 2, AC_FUNC_ATTR_READNONE);
res = LLVMBuildExtractValue(ctx->builder, res, 1, "");
res = LLVMBuildZExt(ctx->builder, res, ctx->i32, "");
}
static LLVMValueRef emit_bitfield_extract(struct nir_to_llvm_context *ctx,
- const char *intrin,
+ bool is_signed,
LLVMValueRef srcs[3])
{
LLVMValueRef result;
LLVMValueRef icond = LLVMBuildICmp(ctx->builder, LLVMIntEQ, srcs[2], LLVMConstInt(ctx->i32, 32, false), "");
- result = ac_emit_llvm_intrinsic(&ctx->ac, intrin, ctx->i32, srcs, 3, AC_FUNC_ATTR_READNONE);
+ result = ac_build_bfe(&ctx->ac, srcs[0], srcs[1], srcs[2], is_signed);
result = LLVMBuildSelect(ctx->builder, icond, srcs[0], result, "");
return result;
}
else
idx = 2;
- result = ac_emit_ddxy(&ctx->ac, ctx->has_ds_bpermute,
+ result = ac_build_ddxy(&ctx->ac, ctx->has_ds_bpermute,
mask, idx, ctx->lds,
src0);
return result;
case nir_op_fmod:
src[0] = to_float(ctx, src[0]);
src[1] = to_float(ctx, src[1]);
- result = ac_emit_fdiv(&ctx->ac, src[0], src[1]);
+ result = ac_build_fdiv(&ctx->ac, src[0], src[1]);
result = emit_intrin_1f_param(ctx, "llvm.floor",
to_float_type(ctx, def_type), result);
result = LLVMBuildFMul(ctx->builder, src[1] , result, "");
case nir_op_fdiv:
src[0] = to_float(ctx, src[0]);
src[1] = to_float(ctx, src[1]);
- result = ac_emit_fdiv(&ctx->ac, src[0], src[1]);
+ result = ac_build_fdiv(&ctx->ac, src[0], src[1]);
break;
case nir_op_frcp:
src[0] = to_float(ctx, src[0]);
- result = ac_emit_fdiv(&ctx->ac, ctx->f32one, src[0]);
+ result = ac_build_fdiv(&ctx->ac, ctx->f32one, src[0]);
break;
case nir_op_iand:
result = LLVMBuildAnd(ctx->builder, src[0], src[1], "");
case nir_op_frsq:
result = emit_intrin_1f_param(ctx, "llvm.sqrt",
to_float_type(ctx, def_type), src[0]);
- result = ac_emit_fdiv(&ctx->ac, ctx->f32one, result);
+ result = ac_build_fdiv(&ctx->ac, ctx->f32one, result);
break;
case nir_op_fpow:
result = emit_intrin_2f_param(ctx, "llvm.pow",
to_float_type(ctx, def_type), src[0], src[1], src[2]);
break;
case nir_op_ibitfield_extract:
- result = emit_bitfield_extract(ctx, "llvm.AMDGPU.bfe.i32", src);
+ result = emit_bitfield_extract(ctx, true, src);
break;
case nir_op_ubitfield_extract:
- result = emit_bitfield_extract(ctx, "llvm.AMDGPU.bfe.u32", src);
+ result = emit_bitfield_extract(ctx, false, src);
break;
case nir_op_bitfield_insert:
result = emit_bitfield_insert(ctx, src[0], src[1], src[2], src[3]);
break;
case nir_op_bitfield_reverse:
- result = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.bitreverse.i32", ctx->i32, src, 1, AC_FUNC_ATTR_READNONE);
+ result = ac_build_intrinsic(&ctx->ac, "llvm.bitreverse.i32", ctx->i32, src, 1, AC_FUNC_ATTR_READNONE);
break;
case nir_op_bit_count:
- result = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.ctpop.i32", ctx->i32, src, 1, AC_FUNC_ATTR_READNONE);
+ result = ac_build_intrinsic(&ctx->ac, "llvm.ctpop.i32", ctx->i32, src, 1, AC_FUNC_ATTR_READNONE);
break;
case nir_op_vec2:
case nir_op_vec3:
src[i] = to_integer(ctx, src[i]);
result = ac_build_gather_values(&ctx->ac, src, num_components);
break;
- case nir_op_d2i:
- case nir_op_f2i:
+ case nir_op_f2i32:
+ case nir_op_f2i64:
src[0] = to_float(ctx, src[0]);
result = LLVMBuildFPToSI(ctx->builder, src[0], def_type, "");
break;
- case nir_op_d2u:
- case nir_op_f2u:
+ case nir_op_f2u32:
+ case nir_op_f2u64:
src[0] = to_float(ctx, src[0]);
result = LLVMBuildFPToUI(ctx->builder, src[0], def_type, "");
break;
- case nir_op_i2d:
- case nir_op_i2f:
+ case nir_op_i2f32:
+ case nir_op_i2f64:
result = LLVMBuildSIToFP(ctx->builder, src[0], to_float_type(ctx, def_type), "");
break;
- case nir_op_u2d:
- case nir_op_u2f:
+ case nir_op_u2f32:
+ case nir_op_u2f64:
result = LLVMBuildUIToFP(ctx->builder, src[0], to_float_type(ctx, def_type), "");
break;
- case nir_op_f2d:
+ case nir_op_f2f64:
result = LLVMBuildFPExt(ctx->builder, src[0], to_float_type(ctx, def_type), "");
break;
- case nir_op_d2f:
+ case nir_op_f2f32:
result = LLVMBuildFPTrunc(ctx->builder, src[0], to_float_type(ctx, def_type), "");
break;
+ case nir_op_u2u32:
+ case nir_op_u2u64:
+ if (get_elem_bits(ctx, LLVMTypeOf(src[0])) < get_elem_bits(ctx, def_type))
+ result = LLVMBuildZExt(ctx->builder, src[0], def_type, "");
+ else
+ result = LLVMBuildTrunc(ctx->builder, src[0], def_type, "");
+ break;
+ case nir_op_i2i32:
+ case nir_op_i2i64:
+ if (get_elem_bits(ctx, LLVMTypeOf(src[0])) < get_elem_bits(ctx, def_type))
+ result = LLVMBuildSExt(ctx->builder, src[0], def_type, "");
+ else
+ result = LLVMBuildTrunc(ctx->builder, src[0], def_type, "");
+ break;
case nir_op_bcsel:
result = emit_bcsel(ctx, src[0], src[1], src[2]);
break;
}
static LLVMValueRef radv_lower_gather4_integer(struct nir_to_llvm_context *ctx,
- struct ac_tex_info *tinfo,
- nir_tex_instr *instr,
- const char *intr_name,
- unsigned coord_vgpr_index)
+ struct ac_image_args *args,
+ nir_tex_instr *instr)
{
- LLVMValueRef coord = tinfo->args[0];
+ enum glsl_base_type stype = glsl_get_sampler_result_type(instr->texture->var->type);
+ LLVMValueRef coord = args->addr;
LLVMValueRef half_texel[2];
+ LLVMValueRef compare_cube_wa;
+ LLVMValueRef result;
int c;
+ unsigned coord_vgpr_index = (unsigned)args->offset + (unsigned)args->compare;
//TODO Rect
{
- LLVMValueRef txq_args[10];
- int txq_arg_count = 0;
- LLVMValueRef size;
- bool da = instr->is_array || instr->sampler_dim == GLSL_SAMPLER_DIM_CUBE;
- txq_args[txq_arg_count++] = LLVMConstInt(ctx->i32, 0, false);
- txq_args[txq_arg_count++] = tinfo->args[1];
- txq_args[txq_arg_count++] = LLVMConstInt(ctx->i32, 0xf, 0); /* dmask */
- txq_args[txq_arg_count++] = LLVMConstInt(ctx->i32, 0, 0); /* unorm */
- txq_args[txq_arg_count++] = LLVMConstInt(ctx->i32, 0, 0); /* r128 */
- txq_args[txq_arg_count++] = LLVMConstInt(ctx->i32, da ? 1 : 0, 0);
- txq_args[txq_arg_count++] = LLVMConstInt(ctx->i32, 0, 0); /* glc */
- txq_args[txq_arg_count++] = LLVMConstInt(ctx->i32, 0, 0); /* slc */
- txq_args[txq_arg_count++] = LLVMConstInt(ctx->i32, 0, 0); /* tfe */
- txq_args[txq_arg_count++] = LLVMConstInt(ctx->i32, 0, 0); /* lwe */
- size = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.SI.getresinfo.i32", ctx->v4i32,
- txq_args, txq_arg_count,
- AC_FUNC_ATTR_READNONE);
+ struct ac_image_args txq_args = { 0 };
+
+ txq_args.da = instr->is_array || instr->sampler_dim == GLSL_SAMPLER_DIM_CUBE;
+ txq_args.opcode = ac_image_get_resinfo;
+ txq_args.dmask = 0xf;
+ txq_args.addr = ctx->i32zero;
+ txq_args.resource = args->resource;
+ LLVMValueRef size = ac_build_image_opcode(&ctx->ac, &txq_args);
for (c = 0; c < 2; c++) {
half_texel[c] = LLVMBuildExtractElement(ctx->builder, size,
LLVMConstInt(ctx->i32, c, false), "");
half_texel[c] = LLVMBuildUIToFP(ctx->builder, half_texel[c], ctx->f32, "");
- half_texel[c] = ac_emit_fdiv(&ctx->ac, ctx->f32one, half_texel[c]);
+ half_texel[c] = ac_build_fdiv(&ctx->ac, ctx->f32one, half_texel[c]);
half_texel[c] = LLVMBuildFMul(ctx->builder, half_texel[c],
LLVMConstReal(ctx->f32, -0.5), "");
}
}
+ LLVMValueRef orig_coords = args->addr;
+
for (c = 0; c < 2; c++) {
LLVMValueRef tmp;
LLVMValueRef index = LLVMConstInt(ctx->i32, coord_vgpr_index + c, 0);
coord = LLVMBuildInsertElement(ctx->builder, coord, tmp, index, "");
}
- tinfo->args[0] = coord;
- return ac_emit_llvm_intrinsic(&ctx->ac, intr_name, tinfo->dst_type, tinfo->args, tinfo->arg_count,
- AC_FUNC_ATTR_READNONE | AC_FUNC_ATTR_NOUNWIND);
+ /*
+ * Apparantly cube has issue with integer types that the workaround doesn't solve,
+ * so this tests if the format is 8_8_8_8 and an integer type do an alternate
+ * workaround by sampling using a scaled type and converting.
+ * This is taken from amdgpu-pro shaders.
+ */
+ /* NOTE this produces some ugly code compared to amdgpu-pro,
+ * LLVM ends up dumping SGPRs into VGPRs to deal with the compare/select,
+ * and then reads them back. -pro generates two selects,
+ * one s_cmp for the descriptor rewriting
+ * one v_cmp for the coordinate and result changes.
+ */
+ if (instr->sampler_dim == GLSL_SAMPLER_DIM_CUBE) {
+ LLVMValueRef tmp, tmp2;
+
+ /* workaround 8/8/8/8 uint/sint cube gather bug */
+ /* first detect it then change to a scaled read and f2i */
+ tmp = LLVMBuildExtractElement(ctx->builder, args->resource, ctx->i32one, "");
+ tmp2 = tmp;
+
+ /* extract the DATA_FORMAT */
+ tmp = ac_build_bfe(&ctx->ac, tmp, LLVMConstInt(ctx->i32, 20, false),
+ LLVMConstInt(ctx->i32, 6, false), false);
+
+ /* is the DATA_FORMAT == 8_8_8_8 */
+ compare_cube_wa = LLVMBuildICmp(ctx->builder, LLVMIntEQ, tmp, LLVMConstInt(ctx->i32, V_008F14_IMG_DATA_FORMAT_8_8_8_8, false), "");
+
+ if (stype == GLSL_TYPE_UINT)
+ /* Create a NUM FORMAT - 0x2 or 0x4 - USCALED or UINT */
+ tmp = LLVMBuildSelect(ctx->builder, compare_cube_wa, LLVMConstInt(ctx->i32, 0x8000000, false),
+ LLVMConstInt(ctx->i32, 0x10000000, false), "");
+ else
+ /* Create a NUM FORMAT - 0x3 or 0x5 - SSCALED or SINT */
+ tmp = LLVMBuildSelect(ctx->builder, compare_cube_wa, LLVMConstInt(ctx->i32, 0xc000000, false),
+ LLVMConstInt(ctx->i32, 0x14000000, false), "");
+
+ /* replace the NUM FORMAT in the descriptor */
+ tmp2 = LLVMBuildAnd(ctx->builder, tmp2, LLVMConstInt(ctx->i32, C_008F14_NUM_FORMAT_GFX6, false), "");
+ tmp2 = LLVMBuildOr(ctx->builder, tmp2, tmp, "");
+
+ args->resource = LLVMBuildInsertElement(ctx->builder, args->resource, tmp2, ctx->i32one, "");
+
+ /* don't modify the coordinates for this case */
+ coord = LLVMBuildSelect(ctx->builder, compare_cube_wa, orig_coords, coord, "");
+ }
+ args->addr = coord;
+ result = ac_build_image_opcode(&ctx->ac, args);
+
+ if (instr->sampler_dim == GLSL_SAMPLER_DIM_CUBE) {
+ LLVMValueRef tmp, tmp2;
+
+ /* if the cube workaround is in place, f2i the result. */
+ for (c = 0; c < 4; c++) {
+ tmp = LLVMBuildExtractElement(ctx->builder, result, LLVMConstInt(ctx->i32, c, false), "");
+ if (stype == GLSL_TYPE_UINT)
+ tmp2 = LLVMBuildFPToUI(ctx->builder, tmp, ctx->i32, "");
+ else
+ tmp2 = LLVMBuildFPToSI(ctx->builder, tmp, ctx->i32, "");
+ tmp = LLVMBuildBitCast(ctx->builder, tmp, ctx->i32, "");
+ tmp2 = LLVMBuildBitCast(ctx->builder, tmp2, ctx->i32, "");
+ tmp = LLVMBuildSelect(ctx->builder, compare_cube_wa, tmp2, tmp, "");
+ tmp = LLVMBuildBitCast(ctx->builder, tmp, ctx->f32, "");
+ result = LLVMBuildInsertElement(ctx->builder, result, tmp, LLVMConstInt(ctx->i32, c, false), "");
+ }
+ }
+ return result;
}
static LLVMValueRef build_tex_intrinsic(struct nir_to_llvm_context *ctx,
nir_tex_instr *instr,
- struct ac_tex_info *tinfo)
-{
- const char *name = "llvm.SI.image.sample";
- const char *infix = "";
- char intr_name[127];
- char type[64];
- bool is_shadow = instr->is_shadow;
- bool has_offset = tinfo->has_offset;
+ struct ac_image_args *args)
+{
+ if (instr->sampler_dim == GLSL_SAMPLER_DIM_BUF) {
+ return ac_build_buffer_load_format(&ctx->ac,
+ args->resource,
+ args->addr,
+ LLVMConstInt(ctx->i32, 0, false),
+ true);
+ }
+
+ args->opcode = ac_image_sample;
+ args->compare = instr->is_shadow;
+
switch (instr->op) {
case nir_texop_txf:
case nir_texop_txf_ms:
case nir_texop_samples_identical:
- name = instr->sampler_dim == GLSL_SAMPLER_DIM_MS ? "llvm.SI.image.load" :
- instr->sampler_dim == GLSL_SAMPLER_DIM_BUF ? "llvm.SI.vs.load.input" :
- "llvm.SI.image.load.mip";
- is_shadow = false;
- has_offset = false;
+ args->opcode = instr->sampler_dim == GLSL_SAMPLER_DIM_MS ? ac_image_load : ac_image_load_mip;
+ args->compare = false;
+ args->offset = false;
break;
case nir_texop_txb:
- infix = ".b";
+ args->bias = true;
break;
case nir_texop_txl:
- infix = ".l";
+ args->lod = true;
break;
case nir_texop_txs:
- name = "llvm.SI.getresinfo";
- break;
case nir_texop_query_levels:
- name = "llvm.SI.getresinfo";
+ args->opcode = ac_image_get_resinfo;
break;
case nir_texop_tex:
if (ctx->stage != MESA_SHADER_FRAGMENT)
- infix = ".lz";
+ args->level_zero = true;
break;
case nir_texop_txd:
- infix = ".d";
+ args->deriv = true;
break;
case nir_texop_tg4:
- name = "llvm.SI.gather4";
- infix = ".lz";
+ args->opcode = ac_image_gather4;
+ args->level_zero = true;
break;
case nir_texop_lod:
- name = "llvm.SI.getlod";
- is_shadow = false;
- has_offset = false;
+ args->opcode = ac_image_get_lod;
+ args->compare = false;
+ args->offset = false;
break;
default:
break;
}
- build_int_type_name(LLVMTypeOf(tinfo->args[0]), type, sizeof(type));
- sprintf(intr_name, "%s%s%s%s.%s", name, is_shadow ? ".c" : "", infix,
- has_offset ? ".o" : "", type);
-
if (instr->op == nir_texop_tg4) {
enum glsl_base_type stype = glsl_get_sampler_result_type(instr->texture->var->type);
if (stype == GLSL_TYPE_UINT || stype == GLSL_TYPE_INT) {
- return radv_lower_gather4_integer(ctx, tinfo, instr, intr_name,
- (int)has_offset + (int)is_shadow);
+ return radv_lower_gather4_integer(ctx, args, instr);
}
}
- return ac_emit_llvm_intrinsic(&ctx->ac, intr_name, tinfo->dst_type, tinfo->args, tinfo->arg_count,
- AC_FUNC_ATTR_READNONE | AC_FUNC_ATTR_NOUNWIND);
-
+ return ac_build_image_opcode(&ctx->ac, args);
}
static LLVMValueRef visit_vulkan_resource_index(struct nir_to_llvm_context *ctx,
unsigned desc_set = nir_intrinsic_desc_set(instr);
unsigned binding = nir_intrinsic_binding(instr);
LLVMValueRef desc_ptr = ctx->descriptor_sets[desc_set];
- struct radv_descriptor_set_layout *layout = ctx->options->layout->set[desc_set].layout;
+ struct radv_pipeline_layout *pipeline_layout = ctx->options->layout;
+ struct radv_descriptor_set_layout *layout = pipeline_layout->set[desc_set].layout;
unsigned base_offset = layout->binding[binding].offset;
LLVMValueRef offset, stride;
if (layout->binding[binding].type == VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC ||
layout->binding[binding].type == VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC) {
+ unsigned idx = pipeline_layout->set[desc_set].dynamic_offset_start +
+ layout->binding[binding].dynamic_offset_offset;
desc_ptr = ctx->push_constants;
- base_offset = ctx->options->layout->push_constant_size;
- base_offset += 16 * layout->binding[binding].dynamic_offset_offset;
+ base_offset = pipeline_layout->push_constant_size + 16 * idx;
stride = LLVMConstInt(ctx->i32, 16, false);
} else
stride = LLVMConstInt(ctx->i32, layout->binding[binding].size, false);
params[1] = get_src(ctx, instr->src[1]);
params[2] = LLVMConstInt(ctx->i32, 0, false); /* vindex */
- params[4] = LLVMConstInt(ctx->i1, 0, false); /* glc */
- params[5] = LLVMConstInt(ctx->i1, 0, false); /* slc */
+ params[4] = ctx->i1false; /* glc */
+ params[5] = ctx->i1false; /* slc */
if (components_32bit > 1)
data_type = LLVMVectorType(ctx->f32, components_32bit);
}
params[0] = data;
params[3] = offset;
- ac_emit_llvm_intrinsic(&ctx->ac, store_name,
- ctx->voidt, params, 6, 0);
+ ac_build_intrinsic(&ctx->ac, store_name,
+ ctx->voidt, params, 6, 0);
}
}
params[arg_count++] = get_src(ctx, instr->src[0]);
params[arg_count++] = LLVMConstInt(ctx->i32, 0, false); /* vindex */
params[arg_count++] = get_src(ctx, instr->src[1]); /* voffset */
- params[arg_count++] = LLVMConstInt(ctx->i1, 0, false); /* slc */
+ params[arg_count++] = ctx->i1false; /* slc */
switch (instr->intrinsic) {
case nir_intrinsic_ssbo_atomic_add:
abort();
}
- return ac_emit_llvm_intrinsic(&ctx->ac, name, ctx->i32, params, arg_count, 0);
+ return ac_build_intrinsic(&ctx->ac, name, ctx->i32, params, arg_count, 0);
}
static LLVMValueRef visit_load_buffer(struct nir_to_llvm_context *ctx,
get_src(ctx, instr->src[0]),
LLVMConstInt(ctx->i32, 0, false),
offset,
- LLVMConstInt(ctx->i1, 0, false),
- LLVMConstInt(ctx->i1, 0, false),
+ ctx->i1false,
+ ctx->i1false,
};
- results[i] = ac_emit_llvm_intrinsic(&ctx->ac, load_name, data_type, params, 5, 0);
+ results[i] = ac_build_intrinsic(&ctx->ac, load_name, data_type, params, 5, 0);
}
LLVMBuildAdd(ctx->builder, LLVMConstInt(ctx->i32, 4 * i, 0),
offset, "")
};
- results[i] = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.SI.load.const", ctx->f32,
- params, 2, AC_FUNC_ATTR_READNONE);
+ results[i] = ac_build_intrinsic(&ctx->ac, "llvm.SI.load.const", ctx->f32,
+ params, 2,
+ AC_FUNC_ATTR_READNONE |
+ AC_FUNC_ATTR_LEGACY);
}
vtx_offset = LLVMBuildMul(ctx->builder, ctx->gs_vtx_offset[vtx_offset_param],
LLVMConstInt(ctx->i32, 4, false), "");
+ param = shader_io_get_unique_index(instr->variables[0]->var->data.location);
for (unsigned i = 0; i < instr->num_components; i++) {
- param = shader_io_get_unique_index(instr->variables[0]->var->data.location);
+
args[0] = ctx->esgs_ring;
args[1] = vtx_offset;
args[2] = LLVMConstInt(ctx->i32, (param * 4 + i + const_index) * 256, false);
args[7] = ctx->i32zero; /* SLC */
args[8] = ctx->i32zero; /* TFE */
- value[i] = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.SI.buffer.load.dword.i32.i32",
- ctx->i32, args, 9, AC_FUNC_ATTR_READONLY);
+ value[i] = ac_build_intrinsic(&ctx->ac, "llvm.SI.buffer.load.dword.i32.i32",
+ ctx->i32, args, 9,
+ AC_FUNC_ATTR_READONLY |
+ AC_FUNC_ATTR_LEGACY);
}
result = ac_build_gather_values(&ctx->ac, value, instr->num_components);
value = llvm_extract_elem(ctx, src, chan);
- if (instr->variables[0]->var->data.location == VARYING_SLOT_CLIP_DIST0 ||
- instr->variables[0]->var->data.location == VARYING_SLOT_CULL_DIST0)
+ if (instr->variables[0]->var->data.compact)
stride = 1;
if (indir_index) {
unsigned count = glsl_count_attribute_slots(
return 0;
}
+
+
+/* Adjust the sample index according to FMASK.
+ *
+ * For uncompressed MSAA surfaces, FMASK should return 0x76543210,
+ * which is the identity mapping. Each nibble says which physical sample
+ * should be fetched to get that sample.
+ *
+ * For example, 0x11111100 means there are only 2 samples stored and
+ * the second sample covers 3/4 of the pixel. When reading samples 0
+ * and 1, return physical sample 0 (determined by the first two 0s
+ * in FMASK), otherwise return physical sample 1.
+ *
+ * The sample index should be adjusted as follows:
+ * sample_index = (fmask >> (sample_index * 4)) & 0xF;
+ */
+static LLVMValueRef adjust_sample_index_using_fmask(struct nir_to_llvm_context *ctx,
+ LLVMValueRef coord_x, LLVMValueRef coord_y,
+ LLVMValueRef coord_z,
+ LLVMValueRef sample_index,
+ LLVMValueRef fmask_desc_ptr)
+{
+ LLVMValueRef fmask_load_address[4];
+ LLVMValueRef res;
+
+ fmask_load_address[0] = coord_x;
+ fmask_load_address[1] = coord_y;
+ if (coord_z) {
+ fmask_load_address[2] = coord_z;
+ fmask_load_address[3] = LLVMGetUndef(ctx->i32);
+ }
+
+ struct ac_image_args args = {0};
+
+ args.opcode = ac_image_load;
+ args.da = coord_z ? true : false;
+ args.resource = fmask_desc_ptr;
+ args.dmask = 0xf;
+ args.addr = ac_build_gather_values(&ctx->ac, fmask_load_address, coord_z ? 4 : 2);
+
+ res = ac_build_image_opcode(&ctx->ac, &args);
+
+ res = to_integer(ctx, res);
+ LLVMValueRef four = LLVMConstInt(ctx->i32, 4, false);
+ LLVMValueRef F = LLVMConstInt(ctx->i32, 0xf, false);
+
+ LLVMValueRef fmask = LLVMBuildExtractElement(ctx->builder,
+ res,
+ ctx->i32zero, "");
+
+ LLVMValueRef sample_index4 =
+ LLVMBuildMul(ctx->builder, sample_index, four, "");
+ LLVMValueRef shifted_fmask =
+ LLVMBuildLShr(ctx->builder, fmask, sample_index4, "");
+ LLVMValueRef final_sample =
+ LLVMBuildAnd(ctx->builder, shifted_fmask, F, "");
+
+ /* Don't rewrite the sample index if WORD1.DATA_FORMAT of the FMASK
+ * resource descriptor is 0 (invalid),
+ */
+ LLVMValueRef fmask_desc =
+ LLVMBuildBitCast(ctx->builder, fmask_desc_ptr,
+ ctx->v8i32, "");
+
+ LLVMValueRef fmask_word1 =
+ LLVMBuildExtractElement(ctx->builder, fmask_desc,
+ ctx->i32one, "");
+
+ LLVMValueRef word1_is_nonzero =
+ LLVMBuildICmp(ctx->builder, LLVMIntNE,
+ fmask_word1, ctx->i32zero, "");
+
+ /* Replace the MSAA sample index. */
+ sample_index =
+ LLVMBuildSelect(ctx->builder, word1_is_nonzero,
+ final_sample, sample_index, "");
+ return sample_index;
+}
+
static LLVMValueRef get_image_coords(struct nir_to_llvm_context *ctx,
nir_intrinsic_instr *instr)
{
LLVMConstInt(ctx->i32, 2, false), LLVMConstInt(ctx->i32, 3, false),
};
LLVMValueRef res;
+ LLVMValueRef sample_index = llvm_extract_elem(ctx, get_src(ctx, instr->src[1]), 0);
+
int count;
enum glsl_sampler_dim dim = glsl_get_sampler_dim(type);
bool add_frag_pos = (dim == GLSL_SAMPLER_DIM_SUBPASS ||
count = image_type_to_components_count(dim,
glsl_sampler_type_is_array(type));
+ if (is_ms) {
+ LLVMValueRef fmask_load_address[3];
+ int chan;
+
+ fmask_load_address[0] = LLVMBuildExtractElement(ctx->builder, src0, masks[0], "");
+ fmask_load_address[1] = LLVMBuildExtractElement(ctx->builder, src0, masks[1], "");
+ if (glsl_sampler_type_is_array(type))
+ fmask_load_address[2] = LLVMBuildExtractElement(ctx->builder, src0, masks[2], "");
+ else
+ fmask_load_address[2] = NULL;
+ if (add_frag_pos) {
+ for (chan = 0; chan < 2; ++chan)
+ fmask_load_address[chan] = LLVMBuildAdd(ctx->builder, fmask_load_address[chan], LLVMBuildFPToUI(ctx->builder, ctx->frag_pos[chan], ctx->i32, ""), "");
+ }
+ sample_index = adjust_sample_index_using_fmask(ctx,
+ fmask_load_address[0],
+ fmask_load_address[1],
+ fmask_load_address[2],
+ sample_index,
+ get_sampler_desc(ctx, instr->variables[0], DESC_FMASK));
+ }
if (count == 1) {
if (instr->src[0].ssa->num_components)
res = LLVMBuildExtractElement(ctx->builder, src0, masks[0], "");
coords[chan] = LLVMBuildAdd(ctx->builder, coords[chan], LLVMBuildFPToUI(ctx->builder, ctx->frag_pos[chan], ctx->i32, ""), "");
}
if (is_ms) {
- coords[count] = llvm_extract_elem(ctx, get_src(ctx, instr->src[1]), 0);
+ coords[count] = sample_index;
count++;
}
return res;
}
-static void build_type_name_for_intr(
- LLVMTypeRef type,
- char *buf, unsigned bufsize)
-{
- LLVMTypeRef elem_type = type;
-
- assert(bufsize >= 8);
-
- if (LLVMGetTypeKind(type) == LLVMVectorTypeKind) {
- int ret = snprintf(buf, bufsize, "v%u",
- LLVMGetVectorSize(type));
- if (ret < 0) {
- char *type_name = LLVMPrintTypeToString(type);
- fprintf(stderr, "Error building type name for: %s\n",
- type_name);
- return;
- }
- elem_type = LLVMGetElementType(type);
- buf += ret;
- bufsize -= ret;
- }
- switch (LLVMGetTypeKind(elem_type)) {
- default: break;
- case LLVMIntegerTypeKind:
- snprintf(buf, bufsize, "i%d", LLVMGetIntTypeWidth(elem_type));
- break;
- case LLVMFloatTypeKind:
- snprintf(buf, bufsize, "f32");
- break;
- case LLVMDoubleTypeKind:
- snprintf(buf, bufsize, "f64");
- break;
- }
-}
-
-static void get_image_intr_name(const char *base_name,
- LLVMTypeRef data_type,
- LLVMTypeRef coords_type,
- LLVMTypeRef rsrc_type,
- char *out_name, unsigned out_len)
-{
- char coords_type_name[8];
-
- build_type_name_for_intr(coords_type, coords_type_name,
- sizeof(coords_type_name));
-
- if (HAVE_LLVM <= 0x0309) {
- snprintf(out_name, out_len, "%s.%s", base_name, coords_type_name);
- } else {
- char data_type_name[8];
- char rsrc_type_name[8];
-
- build_type_name_for_intr(data_type, data_type_name,
- sizeof(data_type_name));
- build_type_name_for_intr(rsrc_type, rsrc_type_name,
- sizeof(rsrc_type_name));
- snprintf(out_name, out_len, "%s.%s.%s.%s", base_name,
- data_type_name, coords_type_name, rsrc_type_name);
- }
-}
-
static LLVMValueRef visit_image_load(struct nir_to_llvm_context *ctx,
nir_intrinsic_instr *instr)
{
params[1] = LLVMBuildExtractElement(ctx->builder, get_src(ctx, instr->src[0]),
LLVMConstInt(ctx->i32, 0, false), ""); /* vindex */
params[2] = LLVMConstInt(ctx->i32, 0, false); /* voffset */
- params[3] = LLVMConstInt(ctx->i1, 0, false); /* glc */
- params[4] = LLVMConstInt(ctx->i1, 0, false); /* slc */
- res = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.amdgcn.buffer.load.format.v4f32", ctx->v4f32,
- params, 5, 0);
+ params[3] = ctx->i1false; /* glc */
+ params[4] = ctx->i1false; /* slc */
+ res = ac_build_intrinsic(&ctx->ac, "llvm.amdgcn.buffer.load.format.v4f32", ctx->v4f32,
+ params, 5, 0);
res = trim_vector(ctx, res, instr->dest.ssa.num_components);
res = to_integer(ctx, res);
} else {
bool is_da = glsl_sampler_type_is_array(type) ||
glsl_get_sampler_dim(type) == GLSL_SAMPLER_DIM_CUBE;
- LLVMValueRef da = is_da ? ctx->i32one : ctx->i32zero;
- LLVMValueRef glc = LLVMConstInt(ctx->i1, 0, false);
- LLVMValueRef slc = LLVMConstInt(ctx->i1, 0, false);
+ LLVMValueRef da = is_da ? ctx->i1true : ctx->i1false;
+ LLVMValueRef glc = ctx->i1false;
+ LLVMValueRef slc = ctx->i1false;
params[0] = get_image_coords(ctx, instr);
params[1] = get_sampler_desc(ctx, instr->variables[0], DESC_IMAGE);
params[2] = LLVMConstInt(ctx->i32, 15, false); /* dmask */
if (HAVE_LLVM <= 0x0309) {
- params[3] = LLVMConstInt(ctx->i1, 0, false); /* r128 */
+ params[3] = ctx->i1false; /* r128 */
params[4] = da;
params[5] = glc;
params[6] = slc;
} else {
- LLVMValueRef lwe = LLVMConstInt(ctx->i1, 0, false);
+ LLVMValueRef lwe = ctx->i1false;
params[3] = glc;
params[4] = slc;
params[5] = lwe;
params[6] = da;
}
- get_image_intr_name("llvm.amdgcn.image.load",
- ctx->v4f32, /* vdata */
- LLVMTypeOf(params[0]), /* coords */
- LLVMTypeOf(params[1]), /* rsrc */
- intrinsic_name, sizeof(intrinsic_name));
+ ac_get_image_intr_name("llvm.amdgcn.image.load",
+ ctx->v4f32, /* vdata */
+ LLVMTypeOf(params[0]), /* coords */
+ LLVMTypeOf(params[1]), /* rsrc */
+ intrinsic_name, sizeof(intrinsic_name));
- res = ac_emit_llvm_intrinsic(&ctx->ac, intrinsic_name, ctx->v4f32,
- params, 7, AC_FUNC_ATTR_READONLY);
+ res = ac_build_intrinsic(&ctx->ac, intrinsic_name, ctx->v4f32,
+ params, 7, AC_FUNC_ATTR_READONLY);
}
return to_integer(ctx, res);
}
LLVMValueRef params[8];
char intrinsic_name[64];
const nir_variable *var = instr->variables[0]->var;
- LLVMValueRef i1false = LLVMConstInt(ctx->i1, 0, 0);
- LLVMValueRef i1true = LLVMConstInt(ctx->i1, 1, 0);
const struct glsl_type *type = glsl_without_array(var->type);
if (ctx->stage == MESA_SHADER_FRAGMENT)
params[2] = LLVMBuildExtractElement(ctx->builder, get_src(ctx, instr->src[0]),
LLVMConstInt(ctx->i32, 0, false), ""); /* vindex */
params[3] = LLVMConstInt(ctx->i32, 0, false); /* voffset */
- params[4] = i1false; /* glc */
- params[5] = i1false; /* slc */
- ac_emit_llvm_intrinsic(&ctx->ac, "llvm.amdgcn.buffer.store.format.v4f32", ctx->voidt,
- params, 6, 0);
+ params[4] = ctx->i1false; /* glc */
+ params[5] = ctx->i1false; /* slc */
+ ac_build_intrinsic(&ctx->ac, "llvm.amdgcn.buffer.store.format.v4f32", ctx->voidt,
+ params, 6, 0);
} else {
bool is_da = glsl_sampler_type_is_array(type) ||
glsl_get_sampler_dim(type) == GLSL_SAMPLER_DIM_CUBE;
- LLVMValueRef da = is_da ? i1true : i1false;
- LLVMValueRef glc = i1false;
- LLVMValueRef slc = i1false;
+ LLVMValueRef da = is_da ? ctx->i1true : ctx->i1false;
+ LLVMValueRef glc = ctx->i1false;
+ LLVMValueRef slc = ctx->i1false;
params[0] = to_float(ctx, get_src(ctx, instr->src[2]));
params[1] = get_image_coords(ctx, instr); /* coords */
params[2] = get_sampler_desc(ctx, instr->variables[0], DESC_IMAGE);
params[3] = LLVMConstInt(ctx->i32, 15, false); /* dmask */
if (HAVE_LLVM <= 0x0309) {
- params[4] = i1false; /* r128 */
+ params[4] = ctx->i1false; /* r128 */
params[5] = da;
params[6] = glc;
params[7] = slc;
} else {
- LLVMValueRef lwe = i1false;
+ LLVMValueRef lwe = ctx->i1false;
params[4] = glc;
params[5] = slc;
params[6] = lwe;
params[7] = da;
}
- get_image_intr_name("llvm.amdgcn.image.store",
- LLVMTypeOf(params[0]), /* vdata */
- LLVMTypeOf(params[1]), /* coords */
- LLVMTypeOf(params[2]), /* rsrc */
- intrinsic_name, sizeof(intrinsic_name));
+ ac_get_image_intr_name("llvm.amdgcn.image.store",
+ LLVMTypeOf(params[0]), /* vdata */
+ LLVMTypeOf(params[1]), /* coords */
+ LLVMTypeOf(params[2]), /* rsrc */
+ intrinsic_name, sizeof(intrinsic_name));
- ac_emit_llvm_intrinsic(&ctx->ac, intrinsic_name, ctx->voidt,
- params, 8, 0);
+ ac_build_intrinsic(&ctx->ac, intrinsic_name, ctx->voidt,
+ params, 8, 0);
}
}
LLVMValueRef params[6];
int param_count = 0;
const nir_variable *var = instr->variables[0]->var;
- LLVMValueRef i1false = LLVMConstInt(ctx->i1, 0, 0);
- LLVMValueRef i1true = LLVMConstInt(ctx->i1, 1, 0);
+
const char *base_name = "llvm.amdgcn.image.atomic";
const char *atomic_name;
LLVMValueRef coords;
coords = params[param_count++] = LLVMBuildExtractElement(ctx->builder, get_src(ctx, instr->src[0]),
LLVMConstInt(ctx->i32, 0, false), ""); /* vindex */
params[param_count++] = ctx->i32zero; /* voffset */
- params[param_count++] = i1false; /* glc */
- params[param_count++] = i1false; /* slc */
+ params[param_count++] = ctx->i1false; /* glc */
+ params[param_count++] = ctx->i1false; /* slc */
} else {
bool da = glsl_sampler_type_is_array(type) ||
glsl_get_sampler_dim(type) == GLSL_SAMPLER_DIM_CUBE;
coords = params[param_count++] = get_image_coords(ctx, instr);
params[param_count++] = get_sampler_desc(ctx, instr->variables[0], DESC_IMAGE);
- params[param_count++] = i1false; /* r128 */
- params[param_count++] = da ? i1true : i1false; /* da */
- params[param_count++] = i1false; /* slc */
+ params[param_count++] = ctx->i1false; /* r128 */
+ params[param_count++] = da ? ctx->i1true : ctx->i1false; /* da */
+ params[param_count++] = ctx->i1false; /* slc */
}
switch (instr->intrinsic) {
snprintf(intrinsic_name, sizeof(intrinsic_name),
"%s.%s.%s", base_name, atomic_name, coords_type);
- return ac_emit_llvm_intrinsic(&ctx->ac, intrinsic_name, ctx->i32, params, param_count, 0);
+ return ac_build_intrinsic(&ctx->ac, intrinsic_name, ctx->i32, params, param_count, 0);
}
static LLVMValueRef visit_image_size(struct nir_to_llvm_context *ctx,
nir_intrinsic_instr *instr)
{
LLVMValueRef res;
- LLVMValueRef params[10];
const nir_variable *var = instr->variables[0]->var;
const struct glsl_type *type = instr->variables[0]->var->type;
bool da = glsl_sampler_type_is_array(var->type) ||
if (glsl_get_sampler_dim(type) == GLSL_SAMPLER_DIM_BUF)
return get_buffer_size(ctx, get_sampler_desc(ctx, instr->variables[0], DESC_BUFFER), true);
- params[0] = ctx->i32zero;
- params[1] = get_sampler_desc(ctx, instr->variables[0], DESC_IMAGE);
- params[2] = LLVMConstInt(ctx->i32, 15, false);
- params[3] = ctx->i32zero;
- params[4] = ctx->i32zero;
- params[5] = da ? ctx->i32one : ctx->i32zero;
- params[6] = ctx->i32zero;
- params[7] = ctx->i32zero;
- params[8] = ctx->i32zero;
- params[9] = ctx->i32zero;
-
- res = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.SI.getresinfo.i32", ctx->v4i32,
- params, 10, AC_FUNC_ATTR_READNONE);
+
+ struct ac_image_args args = { 0 };
+
+ args.da = da;
+ args.dmask = 0xf;
+ args.resource = get_sampler_desc(ctx, instr->variables[0], DESC_IMAGE);
+ args.opcode = ac_image_get_resinfo;
+ args.addr = ctx->i32zero;
+
+ res = ac_build_image_opcode(&ctx->ac, &args);
if (glsl_get_sampler_dim(type) == GLSL_SAMPLER_DIM_CUBE &&
glsl_sampler_type_is_array(type)) {
return res;
}
-static void emit_waitcnt(struct nir_to_llvm_context *ctx)
+#define NOOP_WAITCNT 0xf7f
+#define LGKM_CNT 0x07f
+#define VM_CNT 0xf70
+
+static void emit_waitcnt(struct nir_to_llvm_context *ctx,
+ unsigned simm16)
{
LLVMValueRef args[1] = {
- LLVMConstInt(ctx->i32, 0xf70, false),
+ LLVMConstInt(ctx->i32, simm16, false),
};
- ac_emit_llvm_intrinsic(&ctx->ac, "llvm.amdgcn.s.waitcnt",
- ctx->voidt, args, 1, 0);
+ ac_build_intrinsic(&ctx->ac, "llvm.amdgcn.s.waitcnt",
+ ctx->voidt, args, 1, 0);
}
static void emit_barrier(struct nir_to_llvm_context *ctx)
{
// TODO tess
- ac_emit_llvm_intrinsic(&ctx->ac, "llvm.amdgcn.s.barrier",
- ctx->voidt, NULL, 0, 0);
+ ac_build_intrinsic(&ctx->ac, "llvm.amdgcn.s.barrier",
+ ctx->voidt, NULL, 0, 0);
}
static void emit_discard_if(struct nir_to_llvm_context *ctx,
cond = LLVMBuildSelect(ctx->builder, cond,
LLVMConstReal(ctx->f32, -1.0f),
ctx->f32zero, "");
- ac_emit_llvm_intrinsic(&ctx->ac, "llvm.AMDGPU.kill",
- ctx->voidt,
- &cond, 1, 0);
+ ac_build_kill(&ctx->ac, cond);
}
static LLVMValueRef
location = INTERP_CENTROID;
break;
case nir_intrinsic_interp_var_at_sample:
- case nir_intrinsic_interp_var_at_offset:
location = INTERP_SAMPLE;
src0 = get_src(ctx, instr->src[0]);
break;
+ case nir_intrinsic_interp_var_at_offset:
+ location = INTERP_CENTER;
+ src0 = get_src(ctx, instr->src[0]);
default:
break;
}
interp_param = lookup_interp_param(ctx, instr->variables[0]->var->data.interpolation, location);
attr_number = LLVMConstInt(ctx->i32, input_index, false);
- if (location == INTERP_SAMPLE) {
+ if (location == INTERP_SAMPLE || location == INTERP_CENTER) {
LLVMValueRef ij_out[2];
LLVMValueRef ddxy_out = emit_ddxy_interp(ctx, interp_param);
{
LLVMValueRef gs_next_vertex;
LLVMValueRef can_emit, kill;
- LLVMValueRef args[2];
int idx;
assert(instr->const_index[0] == 0);
kill = LLVMBuildSelect(ctx->builder, can_emit,
LLVMConstReal(ctx->f32, 1.0f),
LLVMConstReal(ctx->f32, -1.0f), "");
- ac_emit_llvm_intrinsic(&ctx->ac, "llvm.AMDGPU.kill",
- ctx->voidt, &kill, 1, 0);
+ ac_build_kill(&ctx->ac, kill);
/* loop num outputs */
idx = 0;
for (unsigned i = 0; i < RADEON_LLVM_MAX_OUTPUTS; ++i) {
LLVMValueRef *out_ptr = &ctx->outputs[i * 4];
+ int length = 4;
+ int slot = idx;
+ int slot_inc = 1;
+
if (!(ctx->output_mask & (1ull << i)))
continue;
- for (unsigned j = 0; j < 4; j++) {
+ if (i == VARYING_SLOT_CLIP_DIST0) {
+ /* pack clip and cull into a single set of slots */
+ length = ctx->num_output_clips + ctx->num_output_culls;
+ if (length > 4)
+ slot_inc = 2;
+ }
+ for (unsigned j = 0; j < length; j++) {
LLVMValueRef out_val = LLVMBuildLoad(ctx->builder,
out_ptr[j], "");
- LLVMValueRef voffset = LLVMConstInt(ctx->i32, (idx * 4 + j) * ctx->gs_max_out_vertices, false);
+ LLVMValueRef voffset = LLVMConstInt(ctx->i32, (slot * 4 + j) * ctx->gs_max_out_vertices, false);
voffset = LLVMBuildAdd(ctx->builder, voffset, gs_next_vertex, "");
voffset = LLVMBuildMul(ctx->builder, voffset, LLVMConstInt(ctx->i32, 4, false), "");
out_val = LLVMBuildBitCast(ctx->builder, out_val, ctx->i32, "");
- ac_build_tbuffer_store(&ctx->ac, ctx->gsvs_ring,
- out_val, 1,
- voffset, ctx->gs2vs_offset, 0,
- V_008F0C_BUF_DATA_FORMAT_32,
- V_008F0C_BUF_NUM_FORMAT_UINT,
- 1, 0, 1, 1, 0);
+ ac_build_buffer_store_dword(&ctx->ac, ctx->gsvs_ring,
+ out_val, 1,
+ voffset, ctx->gs2vs_offset, 0,
+ 1, 1, true, true);
}
- idx++;
+ idx += slot_inc;
}
gs_next_vertex = LLVMBuildAdd(ctx->builder, gs_next_vertex,
ctx->i32one, "");
LLVMBuildStore(ctx->builder, gs_next_vertex, ctx->gs_next_vertex);
- ac_emit_sendmsg(&ctx->ac, AC_SENDMSG_GS_OP_EMIT | AC_SENDMSG_GS | (0 << 8), ctx->gs_wave_id);
+ ac_build_sendmsg(&ctx->ac, AC_SENDMSG_GS_OP_EMIT | AC_SENDMSG_GS | (0 << 8), ctx->gs_wave_id);
}
static void
visit_end_primitive(struct nir_to_llvm_context *ctx,
nir_intrinsic_instr *instr)
{
- ac_emit_sendmsg(&ctx->ac, AC_SENDMSG_GS_OP_CUT | AC_SENDMSG_GS | (0 << 8), ctx->gs_wave_id);
+ ac_build_sendmsg(&ctx->ac, AC_SENDMSG_GS_OP_CUT | AC_SENDMSG_GS | (0 << 8), ctx->gs_wave_id);
}
static void visit_intrinsic(struct nir_to_llvm_context *ctx,
ctx->shader_info->fs.force_persample = true;
result = load_sample_pos(ctx);
break;
+ case nir_intrinsic_load_sample_mask_in:
+ result = ctx->sample_coverage;
+ break;
case nir_intrinsic_load_front_face:
result = ctx->front_face;
break;
break;
case nir_intrinsic_discard:
ctx->shader_info->fs.can_discard = true;
- ac_emit_llvm_intrinsic(&ctx->ac, "llvm.AMDGPU.kilp",
- ctx->voidt,
- NULL, 0, 0);
+ ac_build_intrinsic(&ctx->ac, "llvm.AMDGPU.kilp",
+ ctx->voidt,
+ NULL, 0, AC_FUNC_ATTR_LEGACY);
break;
case nir_intrinsic_discard_if:
emit_discard_if(ctx, instr);
break;
case nir_intrinsic_memory_barrier:
- emit_waitcnt(ctx);
+ emit_waitcnt(ctx, VM_CNT);
break;
case nir_intrinsic_barrier:
emit_barrier(ctx);
LLVMBuilderRef builder = ctx->builder;
LLVMTypeRef type;
LLVMValueRef index = NULL;
+ unsigned constant_index = 0;
assert(deref->var->data.binding < layout->binding_count);
if (child->deref_array_type == nir_deref_array_type_indirect) {
index = get_src(ctx, child->indirect);
}
+
+ constant_index = child->base_offset;
+ }
+ if (desc_type == DESC_SAMPLER && binding->immutable_samplers &&
+ (!index || binding->immutable_samplers_equal)) {
+ if (binding->immutable_samplers_equal)
+ constant_index = 0;
+
+ LLVMValueRef constants[] = {
+ LLVMConstInt(ctx->i32, binding->immutable_samplers[constant_index * 4 + 0], 0),
+ LLVMConstInt(ctx->i32, binding->immutable_samplers[constant_index * 4 + 1], 0),
+ LLVMConstInt(ctx->i32, binding->immutable_samplers[constant_index * 4 + 2], 0),
+ LLVMConstInt(ctx->i32, binding->immutable_samplers[constant_index * 4 + 3], 0),
+ };
+ return ac_build_gather_values(&ctx->ac, constants, 4);
}
assert(stride % type_size == 0);
}
static void set_tex_fetch_args(struct nir_to_llvm_context *ctx,
- struct ac_tex_info *tinfo,
+ struct ac_image_args *args,
nir_tex_instr *instr,
nir_texop op,
LLVMValueRef res_ptr, LLVMValueRef samp_ptr,
LLVMValueRef *param, unsigned count,
unsigned dmask)
{
- int num_args;
unsigned is_rect = 0;
bool da = instr->is_array || instr->sampler_dim == GLSL_SAMPLER_DIM_CUBE;
param[count++] = LLVMGetUndef(ctx->i32);
if (count > 1)
- tinfo->args[0] = ac_build_gather_values(&ctx->ac, param, count);
+ args->addr = ac_build_gather_values(&ctx->ac, param, count);
else
- tinfo->args[0] = param[0];
-
- tinfo->args[1] = res_ptr;
- num_args = 2;
+ args->addr = param[0];
- if (op == nir_texop_txf ||
- op == nir_texop_txf_ms ||
- op == nir_texop_query_levels ||
- op == nir_texop_texture_samples ||
- op == nir_texop_txs)
- tinfo->dst_type = ctx->v4i32;
- else {
- tinfo->dst_type = ctx->v4f32;
- tinfo->args[num_args++] = samp_ptr;
- }
+ args->resource = res_ptr;
+ args->sampler = samp_ptr;
if (instr->sampler_dim == GLSL_SAMPLER_DIM_BUF && op == nir_texop_txf) {
- tinfo->args[0] = res_ptr;
- tinfo->args[1] = LLVMConstInt(ctx->i32, 0, false);
- tinfo->args[2] = param[0];
- tinfo->arg_count = 3;
+ args->addr = param[0];
return;
}
- tinfo->args[num_args++] = LLVMConstInt(ctx->i32, dmask, 0);
- tinfo->args[num_args++] = LLVMConstInt(ctx->i32, is_rect, 0); /* unorm */
- tinfo->args[num_args++] = LLVMConstInt(ctx->i32, 0, 0); /* r128 */
- tinfo->args[num_args++] = LLVMConstInt(ctx->i32, da ? 1 : 0, 0);
- tinfo->args[num_args++] = LLVMConstInt(ctx->i32, 0, 0); /* glc */
- tinfo->args[num_args++] = LLVMConstInt(ctx->i32, 0, 0); /* slc */
- tinfo->args[num_args++] = LLVMConstInt(ctx->i32, 0, 0); /* tfe */
- tinfo->args[num_args++] = LLVMConstInt(ctx->i32, 0, 0); /* lwe */
-
- tinfo->arg_count = num_args;
+ args->dmask = dmask;
+ args->unorm = is_rect;
+ args->da = da;
}
/* Disable anisotropic filtering if BASE_LEVEL == LAST_LEVEL.
LLVMValueRef coord)
{
coord = to_float(ctx, coord);
- coord = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.rint.f32", ctx->f32, &coord, 1, 0);
+ coord = ac_build_intrinsic(&ctx->ac, "llvm.rint.f32", ctx->f32, &coord, 1, 0);
coord = to_integer(ctx, coord);
return coord;
}
static void visit_tex(struct nir_to_llvm_context *ctx, nir_tex_instr *instr)
{
LLVMValueRef result = NULL;
- struct ac_tex_info tinfo = { 0 };
+ struct ac_image_args args = { 0 };
unsigned dmask = 0xf;
LLVMValueRef address[16];
LLVMValueRef coords[5];
}
if (instr->op == nir_texop_txs && instr->sampler_dim == GLSL_SAMPLER_DIM_BUF) {
- result = get_buffer_size(ctx, res_ptr, false);
+ result = get_buffer_size(ctx, res_ptr, true);
goto write_result;
}
for (chan = 0; chan < 3; ++chan)
offset[chan] = ctx->i32zero;
- tinfo.has_offset = true;
+ args.offset = true;
for (chan = 0; chan < get_llvm_num_components(offsets); chan++) {
offset[chan] = llvm_extract_elem(ctx, offsets, chan);
offset[chan] = LLVMBuildAnd(ctx->builder, offset[chan],
if (instr->op == nir_texop_samples_identical) {
LLVMValueRef txf_address[4];
- struct ac_tex_info txf_info = { 0 };
+ struct ac_image_args txf_args = { 0 };
unsigned txf_count = count;
memcpy(txf_address, address, sizeof(txf_address));
txf_address[2] = ctx->i32zero;
txf_address[3] = ctx->i32zero;
- set_tex_fetch_args(ctx, &txf_info, instr, nir_texop_txf,
+ set_tex_fetch_args(ctx, &txf_args, instr, nir_texop_txf,
fmask_ptr, NULL,
txf_address, txf_count, 0xf);
- result = build_tex_intrinsic(ctx, instr, &txf_info);
+ result = build_tex_intrinsic(ctx, instr, &txf_args);
result = LLVMBuildExtractElement(ctx->builder, result, ctx->i32zero, "");
result = emit_int_cmp(ctx, LLVMIntEQ, result, ctx->i32zero);
goto write_result;
}
- /* Adjust the sample index according to FMASK.
- *
- * For uncompressed MSAA surfaces, FMASK should return 0x76543210,
- * which is the identity mapping. Each nibble says which physical sample
- * should be fetched to get that sample.
- *
- * For example, 0x11111100 means there are only 2 samples stored and
- * the second sample covers 3/4 of the pixel. When reading samples 0
- * and 1, return physical sample 0 (determined by the first two 0s
- * in FMASK), otherwise return physical sample 1.
- *
- * The sample index should be adjusted as follows:
- * sample_index = (fmask >> (sample_index * 4)) & 0xF;
- */
if (instr->sampler_dim == GLSL_SAMPLER_DIM_MS &&
instr->op != nir_texop_txs) {
- LLVMValueRef txf_address[4];
- struct ac_tex_info txf_info = { 0 };
- unsigned txf_count = count;
- memcpy(txf_address, address, sizeof(txf_address));
-
- if (!instr->is_array)
- txf_address[2] = ctx->i32zero;
- txf_address[3] = ctx->i32zero;
-
- set_tex_fetch_args(ctx, &txf_info, instr, nir_texop_txf,
- fmask_ptr, NULL,
- txf_address, txf_count, 0xf);
-
- result = build_tex_intrinsic(ctx, instr, &txf_info);
- LLVMValueRef four = LLVMConstInt(ctx->i32, 4, false);
- LLVMValueRef F = LLVMConstInt(ctx->i32, 0xf, false);
-
- LLVMValueRef fmask = LLVMBuildExtractElement(ctx->builder,
- result,
- ctx->i32zero, "");
-
unsigned sample_chan = instr->is_array ? 3 : 2;
-
- LLVMValueRef sample_index4 =
- LLVMBuildMul(ctx->builder, address[sample_chan], four, "");
- LLVMValueRef shifted_fmask =
- LLVMBuildLShr(ctx->builder, fmask, sample_index4, "");
- LLVMValueRef final_sample =
- LLVMBuildAnd(ctx->builder, shifted_fmask, F, "");
-
- /* Don't rewrite the sample index if WORD1.DATA_FORMAT of the FMASK
- * resource descriptor is 0 (invalid),
- */
- LLVMValueRef fmask_desc =
- LLVMBuildBitCast(ctx->builder, fmask_ptr,
- ctx->v8i32, "");
-
- LLVMValueRef fmask_word1 =
- LLVMBuildExtractElement(ctx->builder, fmask_desc,
- ctx->i32one, "");
-
- LLVMValueRef word1_is_nonzero =
- LLVMBuildICmp(ctx->builder, LLVMIntNE,
- fmask_word1, ctx->i32zero, "");
-
- /* Replace the MSAA sample index. */
- address[sample_chan] =
- LLVMBuildSelect(ctx->builder, word1_is_nonzero,
- final_sample, address[sample_chan], "");
+ address[sample_chan] = adjust_sample_index_using_fmask(ctx,
+ address[0],
+ address[1],
+ instr->is_array ? address[2] : NULL,
+ address[sample_chan],
+ fmask_ptr);
}
if (offsets && instr->op == nir_texop_txf) {
else
dmask = 1 << instr->component;
}
- set_tex_fetch_args(ctx, &tinfo, instr, instr->op,
+ set_tex_fetch_args(ctx, &args, instr, instr->op,
res_ptr, samp_ptr, address, count, dmask);
- result = build_tex_intrinsic(ctx, instr, &tinfo);
+ result = build_tex_intrinsic(ctx, instr, &args);
if (instr->op == nir_texop_query_levels)
result = LLVMBuildExtractElement(ctx->builder, result, LLVMConstInt(ctx->i32, 3, false), "");
args[0] = t_list;
args[1] = LLVMConstInt(ctx->i32, 0, false);
args[2] = buffer_index;
- input = ac_emit_llvm_intrinsic(&ctx->ac,
+ input = ac_build_intrinsic(&ctx->ac,
"llvm.SI.vs.load.input", ctx->v4f32, args, 3,
- AC_FUNC_ATTR_READNONE | AC_FUNC_ATTR_NOUNWIND);
+ AC_FUNC_ATTR_READNONE | AC_FUNC_ATTR_NOUNWIND |
+ AC_FUNC_ATTR_LEGACY);
for (unsigned chan = 0; chan < 4; chan++) {
LLVMValueRef llvm_chan = LLVMConstInt(ctx->i32, chan, false);
}
}
-
static void interp_fs_input(struct nir_to_llvm_context *ctx,
unsigned attr,
LLVMValueRef interp_param,
for(int i = 0; i < 3; ++i)
inputs[i] = ctx->frag_pos[i];
- inputs[3] = ac_emit_fdiv(&ctx->ac, ctx->f32one, ctx->frag_pos[3]);
+ inputs[3] = ac_build_fdiv(&ctx->ac, ctx->f32one, ctx->frag_pos[3]);
}
}
ctx->shader_info->fs.num_interp = index;
{
int idx = variable->data.location + variable->data.index;
unsigned attrib_count = glsl_count_attribute_slots(variable->type, false);
-
+ uint64_t mask_attribs;
variable->data.driver_location = idx * 4;
+ mask_attribs = ((1ull << attrib_count) - 1) << idx;
if (ctx->stage == MESA_SHADER_VERTEX ||
ctx->stage == MESA_SHADER_GEOMETRY) {
- if (idx == VARYING_SLOT_CLIP_DIST0 ||
- idx == VARYING_SLOT_CULL_DIST0) {
- int length = glsl_get_length(variable->type);
+ if (idx == VARYING_SLOT_CLIP_DIST0) {
+ int length = ctx->num_output_clips + ctx->num_output_culls;
if (ctx->stage == MESA_SHADER_VERTEX) {
- if (idx == VARYING_SLOT_CLIP_DIST0) {
- ctx->shader_info->vs.clip_dist_mask = (1 << length) - 1;
- ctx->num_clips = length;
- } else if (idx == VARYING_SLOT_CULL_DIST0) {
- ctx->shader_info->vs.cull_dist_mask = (1 << length) - 1;
- ctx->num_culls = length;
- }
+ ctx->shader_info->vs.outinfo.clip_dist_mask = (1 << ctx->num_output_clips) - 1;
+ ctx->shader_info->vs.outinfo.cull_dist_mask = (1 << ctx->num_output_culls) - 1;
}
+
if (length > 4)
attrib_count = 2;
else
attrib_count = 1;
+ mask_attribs = 1ull << idx;
}
}
si_build_alloca_undef(ctx, ctx->f32, "");
}
}
- ctx->output_mask |= ((1ull << attrib_count) - 1) << idx;
+ ctx->output_mask |= mask_attribs;
}
static void
si_llvm_init_export_args(struct nir_to_llvm_context *ctx,
LLVMValueRef *values,
unsigned target,
- LLVMValueRef *args)
+ struct ac_export_args *args)
{
/* Default is 0xf. Adjusted below depending on the format. */
- args[0] = LLVMConstInt(ctx->i32, target != V_008DFC_SQ_EXP_NULL ? 0xf : 0, false);
+ args->enabled_channels = 0xf;
+
/* Specify whether the EXEC mask represents the valid mask */
- args[1] = LLVMConstInt(ctx->i32, 0, false);
+ args->valid_mask = 0;
/* Specify whether this is the last export */
- args[2] = LLVMConstInt(ctx->i32, 0, false);
+ args->done = 0;
+
/* Specify the target we are exporting */
- args[3] = LLVMConstInt(ctx->i32, target, false);
+ args->target = target;
- args[4] = LLVMConstInt(ctx->i32, 0, false); /* COMPR flag */
- args[5] = LLVMGetUndef(ctx->f32);
- args[6] = LLVMGetUndef(ctx->f32);
- args[7] = LLVMGetUndef(ctx->f32);
- args[8] = LLVMGetUndef(ctx->f32);
+ args->compr = false;
+ args->out[0] = LLVMGetUndef(ctx->f32);
+ args->out[1] = LLVMGetUndef(ctx->f32);
+ args->out[2] = LLVMGetUndef(ctx->f32);
+ args->out[3] = LLVMGetUndef(ctx->f32);
if (!values)
return;
switch(col_format) {
case V_028714_SPI_SHADER_ZERO:
- args[0] = LLVMConstInt(ctx->i32, 0x0, 0);
- args[3] = LLVMConstInt(ctx->i32, V_008DFC_SQ_EXP_NULL, 0);
+ args->enabled_channels = 0; /* writemask */
+ args->target = V_008DFC_SQ_EXP_NULL;
break;
case V_028714_SPI_SHADER_32_R:
- args[0] = LLVMConstInt(ctx->i32, 0x1, 0);
- args[5] = values[0];
+ args->enabled_channels = 1;
+ args->out[0] = values[0];
break;
case V_028714_SPI_SHADER_32_GR:
- args[0] = LLVMConstInt(ctx->i32, 0x3, 0);
- args[5] = values[0];
- args[6] = values[1];
+ args->enabled_channels = 0x3;
+ args->out[0] = values[0];
+ args->out[1] = values[1];
break;
case V_028714_SPI_SHADER_32_AR:
- args[0] = LLVMConstInt(ctx->i32, 0x9, 0);
- args[5] = values[0];
- args[8] = values[3];
+ args->enabled_channels = 0x9;
+ args->out[0] = values[0];
+ args->out[3] = values[3];
break;
case V_028714_SPI_SHADER_FP16_ABGR:
- args[4] = ctx->i32one;
+ args->compr = 1;
for (unsigned chan = 0; chan < 2; chan++) {
LLVMValueRef pack_args[2] = {
};
LLVMValueRef packed;
- packed = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.SI.packf16",
- ctx->i32, pack_args, 2,
- AC_FUNC_ATTR_READNONE);
- args[chan + 5] = packed;
+ packed = ac_build_cvt_pkrtz_f16(&ctx->ac, pack_args);
+ args->out[chan] = packed;
}
break;
case V_028714_SPI_SHADER_UNORM16_ABGR:
for (unsigned chan = 0; chan < 4; chan++) {
- val[chan] = emit_float_saturate(ctx, values[chan], 0, 1);
+ val[chan] = ac_build_clamp(&ctx->ac, values[chan]);
val[chan] = LLVMBuildFMul(ctx->builder, val[chan],
LLVMConstReal(ctx->f32, 65535), "");
val[chan] = LLVMBuildFAdd(ctx->builder, val[chan],
ctx->i32, "");
}
- args[4] = ctx->i32one;
- args[5] = emit_pack_int16(ctx, val[0], val[1]);
- args[6] = emit_pack_int16(ctx, val[2], val[3]);
+ args->compr = 1;
+ args->out[0] = emit_pack_int16(ctx, val[0], val[1]);
+ args->out[1] = emit_pack_int16(ctx, val[2], val[3]);
break;
case V_028714_SPI_SHADER_SNORM16_ABGR:
val[chan] = LLVMBuildFPToSI(ctx->builder, val[chan], ctx->i32, "");
}
- args[4] = ctx->i32one;
- args[5] = emit_pack_int16(ctx, val[0], val[1]);
- args[6] = emit_pack_int16(ctx, val[2], val[3]);
+ args->compr = 1;
+ args->out[0] = emit_pack_int16(ctx, val[0], val[1]);
+ args->out[1] = emit_pack_int16(ctx, val[2], val[3]);
break;
case V_028714_SPI_SHADER_UINT16_ABGR: {
val[chan] = emit_minmax_int(ctx, LLVMIntULT, val[chan], max);
}
- args[4] = ctx->i32one;
- args[5] = emit_pack_int16(ctx, val[0], val[1]);
- args[6] = emit_pack_int16(ctx, val[2], val[3]);
+ args->compr = 1;
+ args->out[0] = emit_pack_int16(ctx, val[0], val[1]);
+ args->out[1] = emit_pack_int16(ctx, val[2], val[3]);
break;
}
val[chan] = emit_minmax_int(ctx, LLVMIntSGT, val[chan], min);
}
- args[4] = ctx->i32one;
- args[5] = emit_pack_int16(ctx, val[0], val[1]);
- args[6] = emit_pack_int16(ctx, val[2], val[3]);
+ args->compr = 1;
+ args->out[0] = emit_pack_int16(ctx, val[0], val[1]);
+ args->out[1] = emit_pack_int16(ctx, val[2], val[3]);
break;
}
default:
case V_028714_SPI_SHADER_32_ABGR:
- memcpy(&args[5], values, sizeof(values[0]) * 4);
+ memcpy(&args->out[0], values, sizeof(values[0]) * 4);
break;
}
} else
- memcpy(&args[5], values, sizeof(values[0]) * 4);
+ memcpy(&args->out[0], values, sizeof(values[0]) * 4);
- for (unsigned i = 5; i < 9; ++i)
- args[i] = to_float(ctx, args[i]);
+ for (unsigned i = 0; i < 4; ++i)
+ args->out[i] = to_float(ctx, args->out[i]);
}
static void
-handle_vs_outputs_post(struct nir_to_llvm_context *ctx)
+handle_vs_outputs_post(struct nir_to_llvm_context *ctx,
+ struct ac_vs_output_info *outinfo)
{
uint32_t param_count = 0;
unsigned target;
unsigned pos_idx, num_pos_exports = 0;
- LLVMValueRef args[9];
- LLVMValueRef pos_args[4][9] = { { 0 } };
+ struct ac_export_args args, pos_args[4] = {};
LLVMValueRef psize_value = NULL, layer_value = NULL, viewport_index_value = NULL;
int i;
- const uint64_t clip_mask = ctx->output_mask & ((1ull << VARYING_SLOT_CLIP_DIST0) |
- (1ull << VARYING_SLOT_CLIP_DIST1) |
- (1ull << VARYING_SLOT_CULL_DIST0) |
- (1ull << VARYING_SLOT_CULL_DIST1));
-
- ctx->shader_info->vs.prim_id_output = 0xffffffff;
- ctx->shader_info->vs.layer_output = 0xffffffff;
- if (clip_mask) {
+
+ outinfo->prim_id_output = 0xffffffff;
+ outinfo->layer_output = 0xffffffff;
+ if (ctx->output_mask & (1ull << VARYING_SLOT_CLIP_DIST0)) {
LLVMValueRef slots[8];
unsigned j;
- if (ctx->shader_info->vs.cull_dist_mask)
- ctx->shader_info->vs.cull_dist_mask <<= ctx->num_clips;
+ if (outinfo->cull_dist_mask)
+ outinfo->cull_dist_mask <<= ctx->num_output_clips;
i = VARYING_SLOT_CLIP_DIST0;
- for (j = 0; j < ctx->num_clips; j++)
+ for (j = 0; j < ctx->num_output_clips + ctx->num_output_culls; j++)
slots[j] = to_float(ctx, LLVMBuildLoad(ctx->builder,
ctx->outputs[radeon_llvm_reg_index_soa(i, j)], ""));
- i = VARYING_SLOT_CULL_DIST0;
- for (j = 0; j < ctx->num_culls; j++)
- slots[ctx->num_clips + j] = to_float(ctx, LLVMBuildLoad(ctx->builder,
- ctx->outputs[radeon_llvm_reg_index_soa(i, j)], ""));
- for (i = ctx->num_clips + ctx->num_culls; i < 8; i++)
+ for (i = ctx->num_output_clips + ctx->num_output_culls; i < 8; i++)
slots[i] = LLVMGetUndef(ctx->f32);
- if (ctx->num_clips + ctx->num_culls > 4) {
+ if (ctx->num_output_clips + ctx->num_output_culls > 4) {
target = V_008DFC_SQ_EXP_POS + 3;
- si_llvm_init_export_args(ctx, &slots[4], target, args);
- memcpy(pos_args[target - V_008DFC_SQ_EXP_POS],
- args, sizeof(args));
+ si_llvm_init_export_args(ctx, &slots[4], target, &args);
+ memcpy(&pos_args[target - V_008DFC_SQ_EXP_POS],
+ &args, sizeof(args));
}
target = V_008DFC_SQ_EXP_POS + 2;
- si_llvm_init_export_args(ctx, &slots[0], target, args);
- memcpy(pos_args[target - V_008DFC_SQ_EXP_POS],
- args, sizeof(args));
+ si_llvm_init_export_args(ctx, &slots[0], target, &args);
+ memcpy(&pos_args[target - V_008DFC_SQ_EXP_POS],
+ &args, sizeof(args));
}
if (i == VARYING_SLOT_POS) {
target = V_008DFC_SQ_EXP_POS;
- } else if (i == VARYING_SLOT_CLIP_DIST0 ||
- i == VARYING_SLOT_CLIP_DIST1 ||
- i == VARYING_SLOT_CULL_DIST0 ||
- i == VARYING_SLOT_CULL_DIST1) {
+ } else if (i == VARYING_SLOT_CLIP_DIST0) {
continue;
} else if (i == VARYING_SLOT_PSIZ) {
- ctx->shader_info->vs.writes_pointsize = true;
+ outinfo->writes_pointsize = true;
psize_value = values[0];
continue;
} else if (i == VARYING_SLOT_LAYER) {
- ctx->shader_info->vs.writes_layer = true;
+ outinfo->writes_layer = true;
layer_value = values[0];
- ctx->shader_info->vs.layer_output = param_count;
+ outinfo->layer_output = param_count;
target = V_008DFC_SQ_EXP_PARAM + param_count;
param_count++;
} else if (i == VARYING_SLOT_VIEWPORT) {
- ctx->shader_info->vs.writes_viewport_index = true;
+ outinfo->writes_viewport_index = true;
viewport_index_value = values[0];
continue;
} else if (i == VARYING_SLOT_PRIMITIVE_ID) {
- ctx->shader_info->vs.prim_id_output = param_count;
+ outinfo->prim_id_output = param_count;
target = V_008DFC_SQ_EXP_PARAM + param_count;
param_count++;
} else if (i >= VARYING_SLOT_VAR0) {
- ctx->shader_info->vs.export_mask |= 1u << (i - VARYING_SLOT_VAR0);
+ outinfo->export_mask |= 1u << (i - VARYING_SLOT_VAR0);
target = V_008DFC_SQ_EXP_PARAM + param_count;
param_count++;
}
- si_llvm_init_export_args(ctx, values, target, args);
+ si_llvm_init_export_args(ctx, values, target, &args);
if (target >= V_008DFC_SQ_EXP_POS &&
target <= (V_008DFC_SQ_EXP_POS + 3)) {
- memcpy(pos_args[target - V_008DFC_SQ_EXP_POS],
- args, sizeof(args));
+ memcpy(&pos_args[target - V_008DFC_SQ_EXP_POS],
+ &args, sizeof(args));
} else {
- ac_emit_llvm_intrinsic(&ctx->ac,
- "llvm.SI.export",
- ctx->voidt,
- args, 9, 0);
+ ac_build_export(&ctx->ac, &args);
}
}
/* We need to add the position output manually if it's missing. */
- if (!pos_args[0][0]) {
- pos_args[0][0] = LLVMConstInt(ctx->i32, 0xf, false);
- pos_args[0][1] = ctx->i32zero; /* EXEC mask */
- pos_args[0][2] = ctx->i32zero; /* last export? */
- pos_args[0][3] = LLVMConstInt(ctx->i32, V_008DFC_SQ_EXP_POS, false);
- pos_args[0][4] = ctx->i32zero; /* COMPR flag */
- pos_args[0][5] = ctx->f32zero; /* X */
- pos_args[0][6] = ctx->f32zero; /* Y */
- pos_args[0][7] = ctx->f32zero; /* Z */
- pos_args[0][8] = ctx->f32one; /* W */
- }
-
- uint32_t mask = ((ctx->shader_info->vs.writes_pointsize == true ? 1 : 0) |
- (ctx->shader_info->vs.writes_layer == true ? 4 : 0) |
- (ctx->shader_info->vs.writes_viewport_index == true ? 8 : 0));
+ if (!pos_args[0].out[0]) {
+ pos_args[0].enabled_channels = 0xf;
+ pos_args[0].valid_mask = 0;
+ pos_args[0].done = 0;
+ pos_args[0].target = V_008DFC_SQ_EXP_POS;
+ pos_args[0].compr = 0;
+ pos_args[0].out[0] = ctx->f32zero; /* X */
+ pos_args[0].out[1] = ctx->f32zero; /* Y */
+ pos_args[0].out[2] = ctx->f32zero; /* Z */
+ pos_args[0].out[3] = ctx->f32one; /* W */
+ }
+
+ uint32_t mask = ((outinfo->writes_pointsize == true ? 1 : 0) |
+ (outinfo->writes_layer == true ? 4 : 0) |
+ (outinfo->writes_viewport_index == true ? 8 : 0));
if (mask) {
- pos_args[1][0] = LLVMConstInt(ctx->i32, mask, false); /* writemask */
- pos_args[1][1] = ctx->i32zero; /* EXEC mask */
- pos_args[1][2] = ctx->i32zero; /* last export? */
- pos_args[1][3] = LLVMConstInt(ctx->i32, V_008DFC_SQ_EXP_POS + 1, false);
- pos_args[1][4] = ctx->i32zero; /* COMPR flag */
- pos_args[1][5] = ctx->f32zero; /* X */
- pos_args[1][6] = ctx->f32zero; /* Y */
- pos_args[1][7] = ctx->f32zero; /* Z */
- pos_args[1][8] = ctx->f32zero; /* W */
-
- if (ctx->shader_info->vs.writes_pointsize == true)
- pos_args[1][5] = psize_value;
- if (ctx->shader_info->vs.writes_layer == true)
- pos_args[1][7] = layer_value;
- if (ctx->shader_info->vs.writes_viewport_index == true)
- pos_args[1][8] = viewport_index_value;
+ pos_args[1].enabled_channels = mask;
+ pos_args[1].valid_mask = 0;
+ pos_args[1].done = 0;
+ pos_args[1].target = V_008DFC_SQ_EXP_POS + 1;
+ pos_args[1].compr = 0;
+ pos_args[1].out[0] = ctx->f32zero; /* X */
+ pos_args[1].out[1] = ctx->f32zero; /* Y */
+ pos_args[1].out[2] = ctx->f32zero; /* Z */
+ pos_args[1].out[3] = ctx->f32zero; /* W */
+
+ if (outinfo->writes_pointsize == true)
+ pos_args[1].out[0] = psize_value;
+ if (outinfo->writes_layer == true)
+ pos_args[1].out[2] = layer_value;
+ if (outinfo->writes_viewport_index == true)
+ pos_args[1].out[3] = viewport_index_value;
}
for (i = 0; i < 4; i++) {
- if (pos_args[i][0])
+ if (pos_args[i].out[0])
num_pos_exports++;
}
pos_idx = 0;
for (i = 0; i < 4; i++) {
- if (!pos_args[i][0])
+ if (!pos_args[i].out[0])
continue;
/* Specify the target we are exporting */
- pos_args[i][3] = LLVMConstInt(ctx->i32, V_008DFC_SQ_EXP_POS + pos_idx++, false);
+ pos_args[i].target = V_008DFC_SQ_EXP_POS + pos_idx++;
if (pos_idx == num_pos_exports)
- pos_args[i][2] = ctx->i32one;
- ac_emit_llvm_intrinsic(&ctx->ac,
- "llvm.SI.export",
- ctx->voidt,
- pos_args[i], 9, 0);
+ pos_args[i].done = 1;
+ ac_build_export(&ctx->ac, &pos_args[i]);
}
- ctx->shader_info->vs.pos_exports = num_pos_exports;
- ctx->shader_info->vs.param_exports = param_count;
+ outinfo->pos_exports = num_pos_exports;
+ outinfo->param_exports = param_count;
}
static void
-handle_es_outputs_post(struct nir_to_llvm_context *ctx)
+handle_es_outputs_post(struct nir_to_llvm_context *ctx,
+ struct ac_es_output_info *outinfo)
{
int j;
uint64_t max_output_written = 0;
for (unsigned i = 0; i < RADEON_LLVM_MAX_OUTPUTS; ++i) {
LLVMValueRef *out_ptr = &ctx->outputs[i * 4];
int param_index;
+ int length = 4;
+
if (!(ctx->output_mask & (1ull << i)))
continue;
+ if (i == VARYING_SLOT_CLIP_DIST0)
+ length = ctx->num_output_clips + ctx->num_output_culls;
+
param_index = shader_io_get_unique_index(i);
if (param_index > max_output_written)
max_output_written = param_index;
- for (j = 0; j < 4; j++) {
+ for (j = 0; j < length; j++) {
LLVMValueRef out_val = LLVMBuildLoad(ctx->builder, out_ptr[j], "");
out_val = LLVMBuildBitCast(ctx->builder, out_val, ctx->i32, "");
- ac_build_tbuffer_store(&ctx->ac,
+ ac_build_buffer_store_dword(&ctx->ac,
ctx->esgs_ring,
out_val, 1,
- LLVMGetUndef(ctx->i32), ctx->es2gs_offset,
+ NULL, ctx->es2gs_offset,
(4 * param_index + j) * 4,
- V_008F0C_BUF_DATA_FORMAT_32,
- V_008F0C_BUF_NUM_FORMAT_UINT,
- 0, 0, 1, 1, 0);
+ 1, 1, true, true);
}
}
- ctx->shader_info->vs.esgs_itemsize = (max_output_written + 1) * 16;
+ outinfo->esgs_itemsize = (max_output_written + 1) * 16;
}
static void
si_export_mrt_color(struct nir_to_llvm_context *ctx,
LLVMValueRef *color, unsigned param, bool is_last)
{
- LLVMValueRef args[9];
+
+ struct ac_export_args args;
+
/* Export */
si_llvm_init_export_args(ctx, color, param,
- args);
+ &args);
if (is_last) {
- args[1] = ctx->i32one; /* whether the EXEC mask is valid */
- args[2] = ctx->i32one; /* DONE bit */
- } else if (args[0] == ctx->i32zero)
+ args.valid_mask = 1; /* whether the EXEC mask is valid */
+ args.done = 1; /* DONE bit */
+ } else if (!args.enabled_channels)
return; /* unnecessary NULL export */
- ac_emit_llvm_intrinsic(&ctx->ac, "llvm.SI.export",
- ctx->voidt, args, 9, 0);
+ ac_build_export(&ctx->ac, &args);
}
static void
LLVMValueRef depth, LLVMValueRef stencil,
LLVMValueRef samplemask)
{
- LLVMValueRef args[9];
- unsigned mask = 0;
- args[1] = ctx->i32one; /* whether the EXEC mask is valid */
- args[2] = ctx->i32one; /* DONE bit */
- /* Specify the target we are exporting */
- args[3] = LLVMConstInt(ctx->i32, V_008DFC_SQ_EXP_MRTZ, false);
+ struct ac_export_args args;
- args[4] = ctx->i32zero; /* COMP flag */
- args[5] = LLVMGetUndef(ctx->f32); /* R, depth */
- args[6] = LLVMGetUndef(ctx->f32); /* G, stencil test val[0:7], stencil op val[8:15] */
- args[7] = LLVMGetUndef(ctx->f32); /* B, sample mask */
- args[8] = LLVMGetUndef(ctx->f32); /* A, alpha to mask */
+ args.enabled_channels = 0;
+ args.valid_mask = 1;
+ args.done = 1;
+ args.target = V_008DFC_SQ_EXP_MRTZ;
+ args.compr = false;
+
+ args.out[0] = LLVMGetUndef(ctx->f32); /* R, depth */
+ args.out[1] = LLVMGetUndef(ctx->f32); /* G, stencil test val[0:7], stencil op val[8:15] */
+ args.out[2] = LLVMGetUndef(ctx->f32); /* B, sample mask */
+ args.out[3] = LLVMGetUndef(ctx->f32); /* A, alpha to mask */
if (depth) {
- args[5] = depth;
- mask |= 0x1;
+ args.out[0] = depth;
+ args.enabled_channels |= 0x1;
}
if (stencil) {
- args[6] = stencil;
- mask |= 0x2;
+ args.out[1] = stencil;
+ args.enabled_channels |= 0x2;
}
if (samplemask) {
- args[7] = samplemask;
- mask |= 0x04;
+ args.out[2] = samplemask;
+ args.enabled_channels |= 0x4;
}
/* SI (except OLAND) has a bug that it only looks
* at the X writemask component. */
if (ctx->options->chip_class == SI &&
ctx->options->family != CHIP_OLAND)
- mask |= 0x01;
+ args.enabled_channels |= 0x1;
- args[0] = LLVMConstInt(ctx->i32, mask, false);
- ac_emit_llvm_intrinsic(&ctx->ac, "llvm.SI.export",
- ctx->voidt, args, 9, 0);
+ ac_build_export(&ctx->ac, &args);
}
static void
ctx->shader_info->fs.writes_stencil = true;
stencil = to_float(ctx, LLVMBuildLoad(ctx->builder,
ctx->outputs[radeon_llvm_reg_index_soa(i, 0)], ""));
+ } else if (i == FRAG_RESULT_SAMPLE_MASK) {
+ ctx->shader_info->fs.writes_sample_mask = true;
+ samplemask = to_float(ctx, LLVMBuildLoad(ctx->builder,
+ ctx->outputs[radeon_llvm_reg_index_soa(i, 0)], ""));
} else {
bool last = false;
for (unsigned j = 0; j < 4; j++)
values[j] = to_float(ctx, LLVMBuildLoad(ctx->builder,
ctx->outputs[radeon_llvm_reg_index_soa(i, j)], ""));
- if (!ctx->shader_info->fs.writes_z && !ctx->shader_info->fs.writes_stencil)
+ if (!ctx->shader_info->fs.writes_z && !ctx->shader_info->fs.writes_stencil && !ctx->shader_info->fs.writes_sample_mask)
last = ctx->output_mask <= ((1ull << (i + 1)) - 1);
si_export_mrt_color(ctx, values, V_008DFC_SQ_EXP_MRT + index, last);
}
}
- if (depth || stencil)
+ if (depth || stencil || samplemask)
si_export_mrt_z(ctx, depth, stencil, samplemask);
else if (!index)
si_export_mrt_color(ctx, NULL, V_008DFC_SQ_EXP_NULL, true);
static void
emit_gs_epilogue(struct nir_to_llvm_context *ctx)
{
- ac_emit_sendmsg(&ctx->ac, AC_SENDMSG_GS_OP_NOP | AC_SENDMSG_GS_DONE, ctx->gs_wave_id);
+ ac_build_sendmsg(&ctx->ac, AC_SENDMSG_GS_OP_NOP | AC_SENDMSG_GS_DONE, ctx->gs_wave_id);
}
static void
switch (ctx->stage) {
case MESA_SHADER_VERTEX:
if (ctx->options->key.vs.as_es)
- handle_es_outputs_post(ctx);
+ handle_es_outputs_post(ctx, &ctx->shader_info->vs.es_info);
else
- handle_vs_outputs_post(ctx);
+ handle_vs_outputs_post(ctx, &ctx->shader_info->vs.outinfo);
break;
case MESA_SHADER_FRAGMENT:
handle_fs_outputs_post(ctx);
memset(shader_info, 0, sizeof(*shader_info));
LLVMSetTarget(ctx.module, options->supports_spill ? "amdgcn-mesa-mesa3d" : "amdgcn--");
+
+ LLVMTargetDataRef data_layout = LLVMCreateTargetDataLayout(tm);
+ char *data_layout_str = LLVMCopyStringRepOfTargetData(data_layout);
+ LLVMSetDataLayout(ctx.module, data_layout_str);
+ LLVMDisposeTargetData(data_layout);
+ LLVMDisposeMessage(data_layout_str);
+
setup_types(&ctx);
ctx.builder = LLVMCreateBuilderInContext(ctx.context);
if (nir->stage == MESA_SHADER_FRAGMENT)
handle_fs_inputs_pre(&ctx, nir);
+ ctx.num_output_clips = nir->info->clip_distance_array_size;
+ ctx.num_output_culls = nir->info->cull_distance_array_size;
+
nir_foreach_variable(variable, &nir->outputs)
handle_shader_output_decl(&ctx, variable);
ralloc_free(ctx.phis);
if (nir->stage == MESA_SHADER_GEOMETRY) {
- shader_info->gs.gsvs_vertex_size = util_bitcount64(ctx.output_mask) * 16;
+ unsigned addclip = ctx.num_output_clips + ctx.num_output_culls > 4;
+ shader_info->gs.gsvs_vertex_size = (util_bitcount64(ctx.output_mask) + addclip) * 16;
shader_info->gs.max_gsvs_emit_size = shader_info->gs.gsvs_vertex_size *
nir->info->gs.vertices_out;
}
args[8] = ctx->i32zero; /* TFE */
int idx = 0;
+
for (unsigned i = 0; i < RADEON_LLVM_MAX_OUTPUTS; ++i) {
+ int length = 4;
+ int slot = idx;
+ int slot_inc = 1;
if (!(ctx->output_mask & (1ull << i)))
continue;
- for (unsigned j = 0; j < 4; j++) {
+ if (i == VARYING_SLOT_CLIP_DIST0) {
+ /* unpack clip and cull from a single set of slots */
+ length = ctx->num_output_clips + ctx->num_output_culls;
+ if (length > 4)
+ slot_inc = 2;
+ }
+
+ for (unsigned j = 0; j < length; j++) {
LLVMValueRef value;
args[2] = LLVMConstInt(ctx->i32,
- (idx * 4 + j) *
+ (slot * 4 + j) *
ctx->gs_max_out_vertices * 16 * 4, false);
- value = ac_emit_llvm_intrinsic(&ctx->ac,
- "llvm.SI.buffer.load.dword.i32.i32",
- ctx->i32, args, 9,
- AC_FUNC_ATTR_READONLY);
+ value = ac_build_intrinsic(&ctx->ac,
+ "llvm.SI.buffer.load.dword.i32.i32",
+ ctx->i32, args, 9,
+ AC_FUNC_ATTR_READONLY |
+ AC_FUNC_ATTR_LEGACY);
LLVMBuildStore(ctx->builder,
to_float(ctx, value), ctx->outputs[radeon_llvm_reg_index_soa(i, j)]);
}
- idx++;
+ idx += slot_inc;
}
- handle_vs_outputs_post(ctx);
+ handle_vs_outputs_post(ctx, &ctx->shader_info->vs.outinfo);
}
void ac_create_gs_copy_shader(LLVMTargetMachineRef tm,
ctx.gs_max_out_vertices = geom_shader->info->gs.vertices_out;
ac_setup_rings(&ctx);
+ ctx.num_output_clips = geom_shader->info->clip_distance_array_size;
+ ctx.num_output_culls = geom_shader->info->cull_distance_array_size;
+
nir_foreach_variable(variable, &geom_shader->outputs)
handle_shader_output_decl(&ctx, variable);
projects / mesa.git / blobdiff