param);
if (LLVMGetTypeKind(LLVMTypeOf(value)) == LLVMFloatTypeKind)
- value = bitcast(&ctx->soa.bld_base,
+ value = bitcast(&ctx->bld_base,
TGSI_TYPE_UNSIGNED, value);
if (rshift)
static LLVMValueRef
get_tcs_out_patch0_offset(struct si_shader_context *ctx)
{
- return lp_build_mul_imm(&ctx->soa.bld_base.uint_bld,
+ return lp_build_mul_imm(&ctx->bld_base.uint_bld,
unpack_param(ctx,
SI_PARAM_TCS_OUT_OFFSETS,
0, 16),
static LLVMValueRef
get_tcs_out_patch0_patch_data_offset(struct si_shader_context *ctx)
{
- return lp_build_mul_imm(&ctx->soa.bld_base.uint_bld,
+ return lp_build_mul_imm(&ctx->bld_base.uint_bld,
unpack_param(ctx,
SI_PARAM_TCS_OUT_OFFSETS,
16, 16),
"");
}
-static LLVMValueRef build_gep0(struct si_shader_context *ctx,
- LLVMValueRef base_ptr, LLVMValueRef index)
-{
- LLVMValueRef indices[2] = {
- LLVMConstInt(ctx->i32, 0, 0),
- index,
- };
- return LLVMBuildGEP(ctx->gallivm.builder, base_ptr,
- indices, 2, "");
-}
-
-static void build_indexed_store(struct si_shader_context *ctx,
- LLVMValueRef base_ptr, LLVMValueRef index,
- LLVMValueRef value)
-{
- struct lp_build_tgsi_context *bld_base = &ctx->soa.bld_base;
- struct gallivm_state *gallivm = bld_base->base.gallivm;
-
- LLVMBuildStore(gallivm->builder, value,
- build_gep0(ctx, base_ptr, index));
-}
-
-/**
- * Build an LLVM bytecode indexed load using LLVMBuildGEP + LLVMBuildLoad.
- * It's equivalent to doing a load from &base_ptr[index].
- *
- * \param base_ptr Where the array starts.
- * \param index The element index into the array.
- * \param uniform Whether the base_ptr and index can be assumed to be
- * dynamically uniform
- */
-static LLVMValueRef build_indexed_load(struct si_shader_context *ctx,
- LLVMValueRef base_ptr, LLVMValueRef index,
- bool uniform)
-{
- struct lp_build_tgsi_context *bld_base = &ctx->soa.bld_base;
- struct gallivm_state *gallivm = bld_base->base.gallivm;
- LLVMValueRef pointer;
-
- pointer = build_gep0(ctx, base_ptr, index);
- if (uniform)
- LLVMSetMetadata(pointer, ctx->uniform_md_kind, ctx->empty_md);
- return LLVMBuildLoad(gallivm->builder, pointer, "");
-}
-
-/**
- * Do a load from &base_ptr[index], but also add a flag that it's loading
- * a constant from a dynamically uniform index.
- */
-static LLVMValueRef build_indexed_load_const(
- struct si_shader_context *ctx,
- LLVMValueRef base_ptr, LLVMValueRef index)
-{
- LLVMValueRef result = build_indexed_load(ctx, base_ptr, index, true);
- LLVMSetMetadata(result, ctx->invariant_load_md_kind, ctx->empty_md);
- return result;
-}
-
static LLVMValueRef get_instance_index_for_fetch(
struct si_shader_context *radeon_bld,
unsigned param_start_instance, unsigned divisor)
{
struct si_shader_context *ctx =
- si_shader_context(&radeon_bld->soa.bld_base);
- struct gallivm_state *gallivm = radeon_bld->soa.bld_base.base.gallivm;
+ si_shader_context(&radeon_bld->bld_base);
+ struct gallivm_state *gallivm = radeon_bld->bld_base.base.gallivm;
LLVMValueRef result = LLVMGetParam(radeon_bld->main_fn,
ctx->param_instance_id);
const struct tgsi_full_declaration *decl,
LLVMValueRef out[4])
{
- struct lp_build_context *base = &ctx->soa.bld_base.base;
+ struct lp_build_context *base = &ctx->bld_base.base;
struct gallivm_state *gallivm = base->gallivm;
unsigned chan;
t_offset = lp_build_const_int32(gallivm, input_index);
- t_list = build_indexed_load_const(ctx, t_list_ptr, t_offset);
+ t_list = ac_build_indexed_load_const(&ctx->ac, t_list_ptr, t_offset);
/* Build the attribute offset */
attribute_offset = lp_build_const_int32(gallivm, 0);
input, llvm_chan, "");
}
- fix_fetch = (ctx->shader->key.mono.vs.fix_fetch >> (2 * input_index)) & 3;
- if (fix_fetch) {
+ fix_fetch = (ctx->shader->key.mono.vs.fix_fetch >> (4 * input_index)) & 0xf;
+
+ switch (fix_fetch) {
+ case SI_FIX_FETCH_A2_SNORM:
+ case SI_FIX_FETCH_A2_SSCALED:
+ case SI_FIX_FETCH_A2_SINT: {
/* The hardware returns an unsigned value; convert it to a
* signed one.
*/
}
out[3] = tmp;
+ break;
+ }
+ case SI_FIX_FETCH_RGBA_32_UNORM:
+ case SI_FIX_FETCH_RGBX_32_UNORM:
+ for (chan = 0; chan < 4; chan++) {
+ out[chan] = LLVMBuildBitCast(gallivm->builder, out[chan],
+ ctx->i32, "");
+ out[chan] = LLVMBuildUIToFP(gallivm->builder,
+ out[chan], ctx->f32, "");
+ out[chan] = LLVMBuildFMul(gallivm->builder, out[chan],
+ LLVMConstReal(ctx->f32, 1.0 / UINT_MAX), "");
+ }
+ /* RGBX UINT returns 1 in alpha, which would be rounded to 0 by normalizing. */
+ if (fix_fetch == SI_FIX_FETCH_RGBX_32_UNORM)
+ out[3] = LLVMConstReal(ctx->f32, 1);
+ break;
+ case SI_FIX_FETCH_RGBA_32_SNORM:
+ case SI_FIX_FETCH_RGBX_32_SNORM:
+ case SI_FIX_FETCH_RGBA_32_FIXED:
+ case SI_FIX_FETCH_RGBX_32_FIXED: {
+ double scale;
+ if (fix_fetch >= SI_FIX_FETCH_RGBA_32_FIXED)
+ scale = 1.0 / 0x10000;
+ else
+ scale = 1.0 / INT_MAX;
+
+ for (chan = 0; chan < 4; chan++) {
+ out[chan] = LLVMBuildBitCast(gallivm->builder, out[chan],
+ ctx->i32, "");
+ out[chan] = LLVMBuildSIToFP(gallivm->builder,
+ out[chan], ctx->f32, "");
+ out[chan] = LLVMBuildFMul(gallivm->builder, out[chan],
+ LLVMConstReal(ctx->f32, scale), "");
+ }
+ /* RGBX SINT returns 1 in alpha, which would be rounded to 0 by normalizing. */
+ if (fix_fetch == SI_FIX_FETCH_RGBX_32_SNORM ||
+ fix_fetch == SI_FIX_FETCH_RGBX_32_FIXED)
+ out[3] = LLVMConstReal(ctx->f32, 1);
+ break;
+ }
+ case SI_FIX_FETCH_RGBA_32_USCALED:
+ for (chan = 0; chan < 4; chan++) {
+ out[chan] = LLVMBuildBitCast(gallivm->builder, out[chan],
+ ctx->i32, "");
+ out[chan] = LLVMBuildUIToFP(gallivm->builder,
+ out[chan], ctx->f32, "");
+ }
+ break;
+ case SI_FIX_FETCH_RGBA_32_SSCALED:
+ for (chan = 0; chan < 4; chan++) {
+ out[chan] = LLVMBuildBitCast(gallivm->builder, out[chan],
+ ctx->i32, "");
+ out[chan] = LLVMBuildSIToFP(gallivm->builder,
+ out[chan], ctx->f32, "");
+ }
+ break;
}
}
const struct tgsi_ind_register *ind,
int rel_index)
{
- struct gallivm_state *gallivm = ctx->soa.bld_base.base.gallivm;
+ struct gallivm_state *gallivm = ctx->bld_base.base.gallivm;
LLVMValueRef result;
- result = ctx->soa.addr[ind->Index][ind->Swizzle];
+ result = ctx->addrs[ind->Index][ind->Swizzle];
result = LLVMBuildLoad(gallivm->builder, result, "");
result = LLVMBuildAdd(gallivm->builder, result,
lp_build_const_int32(gallivm, rel_index), "");
LLVMValueRef vertex_dw_stride,
LLVMValueRef base_addr)
{
- struct gallivm_state *gallivm = ctx->soa.bld_base.base.gallivm;
+ struct gallivm_state *gallivm = ctx->bld_base.base.gallivm;
struct tgsi_shader_info *info = &ctx->shader->selector->info;
ubyte *name, *index, *array_first;
int first, param;
LLVMValueRef vertex_index,
LLVMValueRef param_index)
{
- struct gallivm_state *gallivm = ctx->soa.bld_base.base.gallivm;
+ struct gallivm_state *gallivm = ctx->bld_base.base.gallivm;
LLVMValueRef base_addr, vertices_per_patch, num_patches, total_vertices;
LLVMValueRef param_stride, constant16;
const struct tgsi_full_dst_register *dst,
const struct tgsi_full_src_register *src)
{
- struct gallivm_state *gallivm = ctx->soa.bld_base.base.gallivm;
+ struct gallivm_state *gallivm = ctx->bld_base.base.gallivm;
struct tgsi_shader_info *info = &ctx->shader->selector->info;
ubyte *name, *index, *array_first;
struct tgsi_full_src_register reg;
dw_addr = lp_build_add(&bld_base->uint_bld, dw_addr,
lp_build_const_int32(gallivm, swizzle));
- value = build_indexed_load(ctx, ctx->lds, dw_addr, false);
+ value = ac_build_indexed_load(&ctx->ac, ctx->lds, dw_addr, false);
if (tgsi_type_is_64bit(type)) {
LLVMValueRef value2;
dw_addr = lp_build_add(&bld_base->uint_bld, dw_addr,
lp_build_const_int32(gallivm, 1));
- value2 = build_indexed_load(ctx, ctx->lds, dw_addr, false);
+ value2 = ac_build_indexed_load(&ctx->ac, ctx->lds, dw_addr, false);
return si_llvm_emit_fetch_64bit(bld_base, type, value, value2);
}
lp_build_const_int32(gallivm, swizzle));
value = LLVMBuildBitCast(gallivm->builder, value, ctx->i32, "");
- build_indexed_store(ctx, ctx->lds,
- dw_addr, value);
+ ac_build_indexed_store(&ctx->ac, ctx->lds,
+ dw_addr, value);
}
static LLVMValueRef fetch_input_tcs(
rw_buffers = LLVMGetParam(ctx->main_fn,
SI_PARAM_RW_BUFFERS);
- buffer = build_indexed_load_const(ctx, rw_buffers,
+ buffer = ac_build_indexed_load_const(&ctx->ac, rw_buffers,
lp_build_const_int32(gallivm, SI_HS_RING_TESS_OFFCHIP));
base = LLVMGetParam(ctx->main_fn, ctx->param_oc_lds);
rw_buffers = LLVMGetParam(ctx->main_fn,
SI_PARAM_RW_BUFFERS);
- buffer = build_indexed_load_const(ctx, rw_buffers,
+ buffer = ac_build_indexed_load_const(&ctx->ac, rw_buffers,
lp_build_const_int32(gallivm, SI_HS_RING_TESS_OFFCHIP));
base = LLVMGetParam(ctx->main_fn, ctx->param_oc_lds);
struct lp_build_context *base = &bld_base->base;
struct si_shader_context *ctx = si_shader_context(bld_base);
struct si_shader *shader = ctx->shader;
- struct lp_build_context *uint = &ctx->soa.bld_base.uint_bld;
+ struct lp_build_context *uint = &ctx->bld_base.uint_bld;
struct gallivm_state *gallivm = base->gallivm;
LLVMValueRef vtx_offset;
LLVMValueRef args[9];
}
}
-static LLVMValueRef build_fs_interp(
- struct lp_build_tgsi_context *bld_base,
- LLVMValueRef llvm_chan,
- LLVMValueRef attr_number,
- LLVMValueRef params,
- LLVMValueRef i,
- LLVMValueRef j) {
-
- struct si_shader_context *ctx = si_shader_context(bld_base);
- struct gallivm_state *gallivm = bld_base->base.gallivm;
- LLVMValueRef args[5];
- LLVMValueRef p1;
- if (HAVE_LLVM < 0x0400) {
- LLVMValueRef ij[2];
- ij[0] = LLVMBuildBitCast(gallivm->builder, i, ctx->i32, "");
- ij[1] = LLVMBuildBitCast(gallivm->builder, j, ctx->i32, "");
-
- args[0] = llvm_chan;
- args[1] = attr_number;
- args[2] = params;
- args[3] = lp_build_gather_values(gallivm, ij, 2);
- return lp_build_intrinsic(gallivm->builder, "llvm.SI.fs.interp",
- ctx->f32, args, 4,
- LP_FUNC_ATTR_READNONE);
- }
-
- args[0] = i;
- args[1] = llvm_chan;
- args[2] = attr_number;
- args[3] = params;
-
- p1 = lp_build_intrinsic(gallivm->builder, "llvm.amdgcn.interp.p1",
- ctx->f32, args, 4, LP_FUNC_ATTR_READNONE);
-
- args[0] = p1;
- args[1] = j;
- args[2] = llvm_chan;
- args[3] = attr_number;
- args[4] = params;
-
- return lp_build_intrinsic(gallivm->builder, "llvm.amdgcn.interp.p2",
- ctx->f32, args, 5, LP_FUNC_ATTR_READNONE);
-}
-
-static LLVMValueRef build_fs_interp_mov(
- struct lp_build_tgsi_context *bld_base,
- LLVMValueRef parameter,
- LLVMValueRef llvm_chan,
- LLVMValueRef attr_number,
- LLVMValueRef params) {
-
- struct si_shader_context *ctx = si_shader_context(bld_base);
- struct gallivm_state *gallivm = bld_base->base.gallivm;
- LLVMValueRef args[4];
- if (HAVE_LLVM < 0x0400) {
- args[0] = llvm_chan;
- args[1] = attr_number;
- args[2] = params;
-
- return lp_build_intrinsic(gallivm->builder,
- "llvm.SI.fs.constant",
- ctx->f32, args, 3,
- LP_FUNC_ATTR_READNONE);
- }
-
- args[0] = parameter;
- args[1] = llvm_chan;
- args[2] = attr_number;
- args[3] = params;
-
- return lp_build_intrinsic(gallivm->builder, "llvm.amdgcn.interp.mov",
- ctx->f32, args, 4, LP_FUNC_ATTR_READNONE);
-}
-
/**
* Interpolate a fragment shader input.
*
LLVMValueRef face,
LLVMValueRef result[4])
{
- struct lp_build_tgsi_context *bld_base = &ctx->soa.bld_base;
+ struct lp_build_tgsi_context *bld_base = &ctx->bld_base;
struct lp_build_context *base = &bld_base->base;
struct lp_build_context *uint = &bld_base->uint_bld;
struct gallivm_state *gallivm = base->gallivm;
LLVMValueRef front, back;
if (interp) {
- front = build_fs_interp(bld_base, llvm_chan,
+ front = ac_build_fs_interp(&ctx->ac, llvm_chan,
attr_number, prim_mask,
i, j);
- back = build_fs_interp(bld_base, llvm_chan,
+ back = ac_build_fs_interp(&ctx->ac, llvm_chan,
back_attr_number, prim_mask,
i, j);
} else {
- front = build_fs_interp_mov(bld_base,
+ front = ac_build_fs_interp_mov(&ctx->ac,
lp_build_const_int32(gallivm, 2), /* P0 */
llvm_chan, attr_number, prim_mask);
- back = build_fs_interp_mov(bld_base,
+ back = ac_build_fs_interp_mov(&ctx->ac,
lp_build_const_int32(gallivm, 2), /* P0 */
llvm_chan, back_attr_number, prim_mask);
}
}
} else if (semantic_name == TGSI_SEMANTIC_FOG) {
if (interp) {
- result[0] = build_fs_interp(bld_base, uint->zero,
- attr_number, prim_mask, i, j);
+ result[0] = ac_build_fs_interp(&ctx->ac, uint->zero,
+ attr_number, prim_mask, i, j);
} else {
- result[0] = build_fs_interp_mov(bld_base, uint->zero,
- lp_build_const_int32(gallivm, 2), /* P0 */
- attr_number, prim_mask);
+ result[0] = ac_build_fs_interp_mov(&ctx->ac, uint->zero,
+ lp_build_const_int32(gallivm, 2), /* P0 */
+ attr_number, prim_mask);
}
result[1] =
result[2] = lp_build_const_float(gallivm, 0.0f);
LLVMValueRef llvm_chan = lp_build_const_int32(gallivm, chan);
if (interp) {
- result[chan] = build_fs_interp(bld_base,
+ result[chan] = ac_build_fs_interp(&ctx->ac,
llvm_chan, attr_number, prim_mask, i, j);
} else {
- result[chan] = build_fs_interp_mov(bld_base,
+ result[chan] = ac_build_fs_interp_mov(&ctx->ac,
lp_build_const_int32(gallivm, 2), /* P0 */
llvm_chan, attr_number, prim_mask);
}
const struct tgsi_full_declaration *decl,
LLVMValueRef out[4])
{
- struct lp_build_context *base = &radeon_bld->soa.bld_base.base;
+ struct lp_build_context *base = &radeon_bld->bld_base.base;
struct si_shader_context *ctx =
- si_shader_context(&radeon_bld->soa.bld_base);
+ si_shader_context(&radeon_bld->bld_base);
struct si_shader *shader = ctx->shader;
LLVMValueRef main_fn = radeon_bld->main_fn;
LLVMValueRef interp_param = NULL;
static LLVMValueRef get_sample_id(struct si_shader_context *radeon_bld)
{
- return unpack_param(si_shader_context(&radeon_bld->soa.bld_base),
+ return unpack_param(si_shader_context(&radeon_bld->bld_base),
SI_PARAM_ANCILLARY, 8, 4);
}
static LLVMValueRef load_sample_position(struct si_shader_context *radeon_bld, LLVMValueRef sample_id)
{
struct si_shader_context *ctx =
- si_shader_context(&radeon_bld->soa.bld_base);
- struct lp_build_context *uint_bld = &radeon_bld->soa.bld_base.uint_bld;
+ si_shader_context(&radeon_bld->bld_base);
+ struct lp_build_context *uint_bld = &radeon_bld->bld_base.uint_bld;
struct gallivm_state *gallivm = &radeon_bld->gallivm;
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef desc = LLVMGetParam(ctx->main_fn, SI_PARAM_RW_BUFFERS);
LLVMValueRef buf_index = lp_build_const_int32(gallivm, SI_PS_CONST_SAMPLE_POSITIONS);
- LLVMValueRef resource = build_indexed_load_const(ctx, desc, buf_index);
+ LLVMValueRef resource = ac_build_indexed_load_const(&ctx->ac, desc, buf_index);
/* offset = sample_id * 8 (8 = 2 floats containing samplepos.xy) */
LLVMValueRef offset0 = lp_build_mul_imm(uint_bld, sample_id, 8);
const struct tgsi_full_declaration *decl)
{
struct si_shader_context *ctx =
- si_shader_context(&radeon_bld->soa.bld_base);
- struct lp_build_context *bld = &radeon_bld->soa.bld_base.base;
+ si_shader_context(&radeon_bld->bld_base);
+ struct lp_build_context *bld = &radeon_bld->bld_base.base;
struct gallivm_state *gallivm = &radeon_bld->gallivm;
LLVMValueRef value = 0;
LLVMGetParam(radeon_bld->main_fn, SI_PARAM_POS_X_FLOAT),
LLVMGetParam(radeon_bld->main_fn, SI_PARAM_POS_Y_FLOAT),
LLVMGetParam(radeon_bld->main_fn, SI_PARAM_POS_Z_FLOAT),
- lp_build_emit_llvm_unary(&radeon_bld->soa.bld_base, TGSI_OPCODE_RCP,
+ lp_build_emit_llvm_unary(&radeon_bld->bld_base, TGSI_OPCODE_RCP,
LLVMGetParam(radeon_bld->main_fn,
SI_PARAM_POS_W_FLOAT)),
};
lp_build_const_float(gallivm, 0),
lp_build_const_float(gallivm, 0)
};
- pos[0] = lp_build_emit_llvm_unary(&radeon_bld->soa.bld_base,
+ pos[0] = lp_build_emit_llvm_unary(&radeon_bld->bld_base,
TGSI_OPCODE_FRC, pos[0]);
- pos[1] = lp_build_emit_llvm_unary(&radeon_bld->soa.bld_base,
+ pos[1] = lp_build_emit_llvm_unary(&radeon_bld->bld_base,
TGSI_OPCODE_FRC, pos[1]);
value = lp_build_gather_values(gallivm, pos, 4);
break;
rw_buffers = LLVMGetParam(ctx->main_fn,
SI_PARAM_RW_BUFFERS);
- buffer = build_indexed_load_const(ctx, rw_buffers,
+ buffer = ac_build_indexed_load_const(&ctx->ac, rw_buffers,
lp_build_const_int32(gallivm, SI_HS_RING_TESS_OFFCHIP));
base = LLVMGetParam(ctx->main_fn, ctx->param_oc_lds);
addr = get_tcs_tes_buffer_address(ctx, NULL,
lp_build_const_int32(gallivm, param));
- value = buffer_load(&radeon_bld->soa.bld_base, TGSI_TYPE_FLOAT,
+ value = buffer_load(&radeon_bld->bld_base, TGSI_TYPE_FLOAT,
~0, buffer, base, addr);
break;
slot = lp_build_const_int32(gallivm, SI_HS_CONST_DEFAULT_TESS_LEVELS);
buf = LLVMGetParam(ctx->main_fn, SI_PARAM_RW_BUFFERS);
- buf = build_indexed_load_const(ctx, buf, slot);
+ buf = ac_build_indexed_load_const(&ctx->ac, buf, slot);
offset = decl->Semantic.Name == TGSI_SEMANTIC_DEFAULT_TESSINNER_SI ? 4 : 0;
for (i = 0; i < 4; i++)
}
case TGSI_SEMANTIC_PRIMID:
- value = get_primitive_id(&radeon_bld->soa.bld_base, 0);
+ value = get_primitive_id(&radeon_bld->bld_base, 0);
break;
case TGSI_SEMANTIC_GRID_SIZE:
value = LLVMGetParam(radeon_bld->main_fn, SI_PARAM_THREAD_ID);
break;
-#if HAVE_LLVM >= 0x0309
case TGSI_SEMANTIC_HELPER_INVOCATION:
- value = lp_build_intrinsic(gallivm->builder,
- "llvm.amdgcn.ps.live",
- ctx->i1, NULL, 0,
- LP_FUNC_ATTR_READNONE);
- value = LLVMBuildNot(gallivm->builder, value, "");
- value = LLVMBuildSExt(gallivm->builder, value, ctx->i32, "");
+ if (HAVE_LLVM >= 0x0309) {
+ value = lp_build_intrinsic(gallivm->builder,
+ "llvm.amdgcn.ps.live",
+ ctx->i1, NULL, 0,
+ LP_FUNC_ATTR_READNONE);
+ value = LLVMBuildNot(gallivm->builder, value, "");
+ value = LLVMBuildSExt(gallivm->builder, value, ctx->i32, "");
+ } else {
+ assert(!"TGSI_SEMANTIC_HELPER_INVOCATION unsupported");
+ return;
+ }
break;
-#endif
default:
assert(!"unknown system value");
const struct tgsi_full_declaration *decl)
{
struct si_shader_context *ctx =
- si_shader_context(&radeon_bld->soa.bld_base);
+ si_shader_context(&radeon_bld->bld_base);
struct si_shader_selector *sel = ctx->shader->selector;
struct gallivm_state *gallivm = &radeon_bld->gallivm;
LLVMValueRef list_ptr = LLVMGetParam(ctx->main_fn,
SI_PARAM_CONST_BUFFERS);
- return build_indexed_load_const(ctx, list_ptr,
+ return ac_build_indexed_load_const(&ctx->ac, list_ptr,
LLVMConstInt(ctx->i32, i, 0));
}
index = get_bounded_indirect_index(ctx, ®->DimIndirect,
reg->Dimension.Index,
SI_NUM_CONST_BUFFERS);
- bufp = build_indexed_load_const(ctx, ptr, index);
+ bufp = ac_build_indexed_load_const(&ctx->ac, ptr, index);
} else
bufp = load_const_buffer_desc(ctx, buf);
if (reg->Register.Indirect) {
- addr = ctx->soa.addr[ireg->Index][ireg->Swizzle];
+ addr = ctx->addrs[ireg->Index][ireg->Swizzle];
addr = LLVMBuildLoad(base->gallivm->builder, addr, "load addr reg");
addr = lp_build_mul_imm(&bld_base->uint_bld, addr, 16);
addr = lp_build_add(&bld_base->uint_bld, addr,
LLVMValueRef *args)
{
struct si_shader_context *ctx = si_shader_context(bld_base);
- struct lp_build_context *uint =
- &ctx->soa.bld_base.uint_bld;
+ struct lp_build_context *uint = &ctx->bld_base.uint_bld;
struct lp_build_context *base = &bld_base->base;
struct gallivm_state *gallivm = base->gallivm;
LLVMBuilderRef builder = base->gallivm->builder;
{
struct si_shader_context *ctx = si_shader_context(bld_base);
struct lp_build_context *base = &bld_base->base;
- struct lp_build_context *uint = &ctx->soa.bld_base.uint_bld;
+ struct lp_build_context *uint = &ctx->bld_base.uint_bld;
unsigned reg_index;
unsigned chan;
unsigned const_chan;
LLVMValueRef ptr = LLVMGetParam(ctx->main_fn, SI_PARAM_RW_BUFFERS);
LLVMValueRef constbuf_index = lp_build_const_int32(base->gallivm,
SI_VS_CONST_CLIP_PLANES);
- LLVMValueRef const_resource = build_indexed_load_const(ctx, ptr, constbuf_index);
+ LLVMValueRef const_resource = ac_build_indexed_load_const(&ctx->ac, ptr, constbuf_index);
for (reg_index = 0; reg_index < 2; reg_index ++) {
LLVMValueRef *args = pos[2 + reg_index];
LLVMValueRef offset = lp_build_const_int32(gallivm,
SI_VS_STREAMOUT_BUF0 + i);
- so_buffers[i] = build_indexed_load_const(ctx, buf_ptr, offset);
+ so_buffers[i] = ac_build_indexed_load_const(&ctx->ac, buf_ptr, offset);
LLVMValueRef so_offset = LLVMGetParam(ctx->main_fn,
ctx->param_streamout_offset[i]);
struct si_shader_context *ctx = si_shader_context(bld_base);
struct si_shader *shader = ctx->shader;
struct lp_build_context *base = &bld_base->base;
- struct lp_build_context *uint =
- &ctx->soa.bld_base.uint_bld;
+ struct lp_build_context *uint = &ctx->bld_base.uint_bld;
LLVMValueRef args[9];
LLVMValueRef pos_args[4][9] = { { 0 } };
LLVMValueRef psize_value = NULL, edgeflag_value = NULL, layer_value = NULL, viewport_index_value = NULL;
case TGSI_SEMANTIC_POSITION:
target = V_008DFC_SQ_EXP_POS;
break;
- case TGSI_SEMANTIC_COLOR:
- case TGSI_SEMANTIC_BCOLOR:
- if (!export_param)
- continue;
- target = V_008DFC_SQ_EXP_PARAM + param_count;
- assert(i < ARRAY_SIZE(shader->info.vs_output_param_offset));
- shader->info.vs_output_param_offset[i] = param_count;
- param_count++;
- break;
case TGSI_SEMANTIC_CLIPDIST:
if (shader->key.opt.hw_vs.clip_disable) {
semantic_name = TGSI_SEMANTIC_GENERIC;
continue;
si_llvm_emit_clipvertex(bld_base, pos_args, outputs[i].values);
continue;
+ case TGSI_SEMANTIC_COLOR:
+ case TGSI_SEMANTIC_BCOLOR:
case TGSI_SEMANTIC_PRIMID:
case TGSI_SEMANTIC_FOG:
case TGSI_SEMANTIC_TEXCOORD:
invocation_id = unpack_param(ctx, SI_PARAM_REL_IDS, 8, 5);
rw_buffers = LLVMGetParam(ctx->main_fn, SI_PARAM_RW_BUFFERS);
- buffer = build_indexed_load_const(ctx, rw_buffers,
+ buffer = ac_build_indexed_load_const(&ctx->ac, rw_buffers,
lp_build_const_int32(gallivm, SI_HS_RING_TESS_OFFCHIP));
buffer_offset = LLVMGetParam(ctx->main_fn, ctx->param_oc_lds);
/* Get the buffer. */
rw_buffers = LLVMGetParam(ctx->main_fn,
SI_PARAM_RW_BUFFERS);
- buffer = build_indexed_load_const(ctx, rw_buffers,
+ buffer = ac_build_indexed_load_const(&ctx->ac, rw_buffers,
lp_build_const_int32(gallivm, SI_HS_RING_TESS_FACTOR));
/* Get the offset. */
/* Write outputs to LDS. The next shader (TCS aka HS) will read
* its inputs from it. */
for (i = 0; i < info->num_outputs; i++) {
- LLVMValueRef *out_ptr = ctx->soa.outputs[i];
+ LLVMValueRef *out_ptr = ctx->outputs[i];
unsigned name = info->output_semantic_name[i];
unsigned index = info->output_semantic_index[i];
int param = si_shader_io_get_unique_index(name, index);
int i;
for (i = 0; i < info->num_outputs; i++) {
- LLVMValueRef *out_ptr =
- ctx->soa.outputs[i];
+ LLVMValueRef *out_ptr = ctx->outputs[i];
int param_index;
if (info->output_semantic_name[i] == TGSI_SEMANTIC_VIEWPORT_INDEX ||
}
for (j = 0; j < 4; j++) {
- addr = ctx->soa.outputs[i][j];
+ addr = ctx->outputs[i][j];
val = LLVMBuildLoad(gallivm->builder, addr, "");
val = si_llvm_saturate(bld_base, val);
LLVMBuildStore(gallivm->builder, val, addr);
for (j = 0; j < 4; j++) {
outputs[i].values[j] =
LLVMBuildLoad(gallivm->builder,
- ctx->soa.outputs[i][j],
+ ctx->outputs[i][j],
"");
outputs[i].vertex_stream[j] =
(info->output_streams[i] >> (2 * j)) & 3;
case TGSI_SEMANTIC_COLOR:
assert(semantic_index < 8);
for (j = 0; j < 4; j++) {
- LLVMValueRef ptr = ctx->soa.outputs[i][j];
+ LLVMValueRef ptr = ctx->outputs[i][j];
LLVMValueRef result = LLVMBuildLoad(builder, ptr, "");
color[semantic_index][j] = result;
}
break;
case TGSI_SEMANTIC_POSITION:
depth = LLVMBuildLoad(builder,
- ctx->soa.outputs[i][2], "");
+ ctx->outputs[i][2], "");
break;
case TGSI_SEMANTIC_STENCIL:
stencil = LLVMBuildLoad(builder,
- ctx->soa.outputs[i][1], "");
+ ctx->outputs[i][1], "");
break;
case TGSI_SEMANTIC_SAMPLEMASK:
samplemask = LLVMBuildLoad(builder,
- ctx->soa.outputs[i][0], "");
+ ctx->outputs[i][0], "");
break;
default:
fprintf(stderr, "Warning: SI unhandled fs output type:%d\n",
reg->Register.Index,
SI_NUM_SHADER_BUFFERS);
- return build_indexed_load_const(ctx, rsrc_ptr, index);
+ return ac_build_indexed_load_const(&ctx->ac, rsrc_ptr, index);
}
static bool tgsi_is_array_sampler(unsigned target)
LLVMConstInt(ctx->i32, 2, 0), "");
index = LLVMBuildAdd(builder, index,
LLVMConstInt(ctx->i32, 1, 0), "");
- *rsrc = build_indexed_load_const(ctx, rsrc_ptr, index);
+ *rsrc = ac_build_indexed_load_const(&ctx->ac, rsrc_ptr, index);
return;
}
- tmp = build_indexed_load_const(ctx, rsrc_ptr, index);
+ tmp = ac_build_indexed_load_const(&ctx->ac, rsrc_ptr, index);
if (dcc_off)
tmp = force_dcc_off(ctx, tmp);
*rsrc = tmp;
LLVMValueRef offset, ptr;
int addr_space;
- offset = lp_build_emit_fetch(&ctx->soa.bld_base, inst, arg, 0);
+ offset = lp_build_emit_fetch(&ctx->bld_base, inst, arg, 0);
offset = LLVMBuildBitCast(builder, offset, ctx->i32, "");
ptr = ctx->shared_memory;
struct lp_build_emit_data *emit_data)
{
const struct tgsi_full_instruction *inst = emit_data->inst;
- struct lp_build_context *base = &ctx->soa.bld_base.base;
+ struct lp_build_context *base = &ctx->bld_base.base;
struct gallivm_state *gallivm = &ctx->gallivm;
LLVMBuilderRef builder = gallivm->builder;
unsigned writemask = inst->Dst[0].Register.WriteMask;
const struct tgsi_full_instruction *inst = emit_data->inst;
struct gallivm_state *gallivm = &ctx->gallivm;
LLVMBuilderRef builder = gallivm->builder;
- struct lp_build_context *uint_bld = &ctx->soa.bld_base.uint_bld;
+ struct lp_build_context *uint_bld = &ctx->bld_base.uint_bld;
LLVMValueRef base_data = emit_data->args[0];
LLVMValueRef base_offset = emit_data->args[3];
unsigned writemask = inst->Dst[0].Register.WriteMask;
{
const struct tgsi_full_instruction *inst = emit_data->inst;
struct gallivm_state *gallivm = &ctx->gallivm;
- struct lp_build_context *base = &ctx->soa.bld_base.base;
+ struct lp_build_context *base = &ctx->bld_base.base;
LLVMBuilderRef builder = gallivm->builder;
unsigned writemask = inst->Dst[0].Register.WriteMask;
LLVMValueRef ptr, derived_ptr, data, index;
if (!(writemask & (1 << chan))) {
continue;
}
- data = lp_build_emit_fetch(&ctx->soa.bld_base, inst, 1, chan);
+ data = lp_build_emit_fetch(&ctx->bld_base, inst, 1, chan);
index = lp_build_const_int32(gallivm, chan);
derived_ptr = LLVMBuildGEP(builder, ptr, &index, 1, "");
LLVMBuildStore(builder, data, derived_ptr);
ptr = get_memory_ptr(ctx, inst, ctx->i32, 1);
- arg = lp_build_emit_fetch(&ctx->soa.bld_base, inst, 2, 0);
+ arg = lp_build_emit_fetch(&ctx->bld_base, inst, 2, 0);
arg = LLVMBuildBitCast(builder, arg, ctx->i32, "");
if (inst->Instruction.Opcode == TGSI_OPCODE_ATOMCAS) {
LLVMValueRef new_data;
- new_data = lp_build_emit_fetch(&ctx->soa.bld_base,
+ new_data = lp_build_emit_fetch(&ctx->bld_base,
inst, 3, 0);
new_data = LLVMBuildBitCast(builder, new_data, ctx->i32, "");
break;
}
- return build_indexed_load_const(ctx, list, index);
+ return ac_build_indexed_load_const(&ctx->ac, list, index);
}
static LLVMValueRef load_sampler_desc(struct si_shader_context *ctx,
target == TGSI_TEXTURE_CUBE_ARRAY ||
target == TGSI_TEXTURE_SHADOWCUBE ||
target == TGSI_TEXTURE_SHADOWCUBE_ARRAY)
- si_prepare_cube_coords(bld_base, emit_data, coords, derivs);
+ ac_prepare_cube_coords(&ctx->ac,
+ opcode == TGSI_OPCODE_TXD,
+ target == TGSI_TEXTURE_CUBE_ARRAY ||
+ target == TGSI_TEXTURE_SHADOWCUBE_ARRAY,
+ coords, derivs);
if (opcode == TGSI_OPCODE_TXD)
for (int i = 0; i < num_deriv_channels * 2; i++)
/* add tex offsets */
if (inst->Texture.NumOffsets) {
struct lp_build_context *uint_bld = &bld_base->uint_bld;
- struct lp_build_tgsi_soa_context *bld = lp_soa_context(bld_base);
const struct tgsi_texture_offset *off = inst->TexOffsets;
assert(inst->Texture.NumOffsets == 1);
switch (target) {
case TGSI_TEXTURE_3D:
address[2] = lp_build_add(uint_bld, address[2],
- bld->immediates[off->Index][off->SwizzleZ]);
+ ctx->imms[off->Index * TGSI_NUM_CHANNELS + off->SwizzleZ]);
/* fall through */
case TGSI_TEXTURE_2D:
case TGSI_TEXTURE_SHADOW2D:
case TGSI_TEXTURE_SHADOW2D_ARRAY:
address[1] =
lp_build_add(uint_bld, address[1],
- bld->immediates[off->Index][off->SwizzleY]);
+ ctx->imms[off->Index * TGSI_NUM_CHANNELS + off->SwizzleY]);
/* fall through */
case TGSI_TEXTURE_1D:
case TGSI_TEXTURE_SHADOW1D:
case TGSI_TEXTURE_SHADOW1D_ARRAY:
address[0] =
lp_build_add(uint_bld, address[0],
- bld->immediates[off->Index][off->SwizzleX]);
+ ctx->imms[off->Index * TGSI_NUM_CHANNELS + off->SwizzleX]);
break;
/* texture offsets do not apply to other texture targets */
}
/* Get the component index from src1.x for Gather4. */
if (!tgsi_is_shadow_target(target)) {
- LLVMValueRef (*imms)[4] = lp_soa_context(bld_base)->immediates;
LLVMValueRef comp_imm;
struct tgsi_src_register src1 = inst->Src[1].Register;
assert(src1.File == TGSI_FILE_IMMEDIATE);
- comp_imm = imms[src1.Index][src1.SwizzleX];
+ comp_imm = ctx->imms[src1.Index * TGSI_NUM_CHANNELS + src1.SwizzleX];
gather_comp = LLVMConstIntGetZExtValue(comp_imm);
gather_comp = CLAMP(gather_comp, 0, 3);
}
set_tex_fetch_args(ctx, &txq_emit_data, TGSI_OPCODE_TXQ,
txq_inst.Texture.Texture,
emit_data->args[1], NULL,
- &ctx->soa.bld_base.uint_bld.zero,
+ &ctx->bld_base.uint_bld.zero,
1, 0xf);
- txq_emit(NULL, &ctx->soa.bld_base, &txq_emit_data);
+ txq_emit(NULL, &ctx->bld_base, &txq_emit_data);
/* Compute -0.5 / size. */
for (c = 0; c < 2; c++) {
LLVMConstInt(ctx->i32, c, 0), "");
half_texel[c] = LLVMBuildUIToFP(builder, half_texel[c], ctx->f32, "");
half_texel[c] =
- lp_build_emit_llvm_unary(&ctx->soa.bld_base,
+ lp_build_emit_llvm_unary(&ctx->bld_base,
TGSI_OPCODE_RCP, half_texel[c]);
half_texel[c] = LLVMBuildFMul(builder, half_texel[c],
LLVMConstReal(ctx->f32, -0.5), "");
} else {
LLVMValueRef store_ptr, load_ptr0, load_ptr1;
- store_ptr = build_gep0(ctx, ctx->lds, thread_id);
- load_ptr0 = build_gep0(ctx, ctx->lds, tl_tid);
- load_ptr1 = build_gep0(ctx, ctx->lds, trbl_tid);
+ store_ptr = ac_build_gep0(&ctx->ac, ctx->lds, thread_id);
+ load_ptr0 = ac_build_gep0(&ctx->ac, ctx->lds, tl_tid);
+ load_ptr1 = ac_build_gep0(&ctx->ac, ctx->lds, trbl_tid);
LLVMBuildStore(gallivm->builder, val, store_ptr);
tl = LLVMBuildLoad(gallivm->builder, load_ptr0, "");
gallivm->builder, interp_param, uint->zero, "");
LLVMValueRef j = LLVMBuildExtractElement(
gallivm->builder, interp_param, uint->one, "");
- emit_data->output[chan] = build_fs_interp(bld_base,
+ emit_data->output[chan] = ac_build_fs_interp(&ctx->ac,
llvm_chan, attr_number, params,
i, j);
} else {
- emit_data->output[chan] = build_fs_interp_mov(bld_base,
+ emit_data->output[chan] = ac_build_fs_interp_mov(&ctx->ac,
lp_build_const_int32(gallivm, 2), /* P0 */
llvm_chan, attr_number, params);
}
static unsigned si_llvm_get_stream(struct lp_build_tgsi_context *bld_base,
struct lp_build_emit_data *emit_data)
{
- LLVMValueRef (*imms)[4] = lp_soa_context(bld_base)->immediates;
+ struct si_shader_context *ctx = si_shader_context(bld_base);
struct tgsi_src_register src0 = emit_data->inst->Src[0].Register;
+ LLVMValueRef imm;
unsigned stream;
assert(src0.File == TGSI_FILE_IMMEDIATE);
- stream = LLVMConstIntGetZExtValue(imms[src0.Index][src0.SwizzleX]) & 0x3;
+ imm = ctx->imms[src0.Index * TGSI_NUM_CHANNELS + src0.SwizzleX];
+ stream = LLVMConstIntGetZExtValue(imm) & 0x3;
return stream;
}
LLVMValueRef gs_next_vertex;
LLVMValueRef can_emit, kill;
LLVMValueRef args[2];
- unsigned chan;
+ unsigned chan, offset;
int i;
unsigned stream;
lp_build_if(&if_state, gallivm, can_emit);
}
+ offset = 0;
for (i = 0; i < info->num_outputs; i++) {
- LLVMValueRef *out_ptr =
- ctx->soa.outputs[i];
+ LLVMValueRef *out_ptr = ctx->outputs[i];
for (chan = 0; chan < 4; chan++) {
- if (((info->output_streams[i] >> (2 * chan)) & 3) != stream)
+ if (!(info->output_usagemask[i] & (1 << chan)) ||
+ ((info->output_streams[i] >> (2 * chan)) & 3) != stream)
continue;
LLVMValueRef out_val = LLVMBuildLoad(gallivm->builder, out_ptr[chan], "");
LLVMValueRef voffset =
- lp_build_const_int32(gallivm, (i * 4 + chan) *
+ lp_build_const_int32(gallivm, offset *
shader->selector->gs_max_out_vertices);
+ offset++;
voffset = lp_build_add(uint, voffset, gs_next_vertex);
voffset = lp_build_mul_imm(uint, voffset, 4);
struct si_shader_context *ctx = si_shader_context(bld_base);
struct gallivm_state *gallivm = bld_base->base.gallivm;
- /* The real barrier instruction isn’t needed, because an entire patch
+ /* SI only (thanks to a hw bug workaround):
+ * The real barrier instruction isn’t needed, because an entire patch
* always fits into a single wave.
*/
- if (ctx->type == PIPE_SHADER_TESS_CTRL) {
+ if (HAVE_LLVM >= 0x0309 &&
+ ctx->screen->b.chip_class == SI &&
+ ctx->type == PIPE_SHADER_TESS_CTRL) {
emit_waitcnt(ctx, LGKM_CNT & VM_CNT);
return;
}
static void create_meta_data(struct si_shader_context *ctx)
{
- struct gallivm_state *gallivm = ctx->soa.bld_base.base.gallivm;
+ struct gallivm_state *gallivm = ctx->bld_base.base.gallivm;
- ctx->invariant_load_md_kind = LLVMGetMDKindIDInContext(gallivm->context,
- "invariant.load", 14);
ctx->range_md_kind = LLVMGetMDKindIDInContext(gallivm->context,
"range", 5);
- ctx->uniform_md_kind = LLVMGetMDKindIDInContext(gallivm->context,
- "amdgpu.uniform", 14);
-
- ctx->empty_md = LLVMMDNodeInContext(gallivm->context, NULL, 0);
}
static void declare_streamout_params(struct si_shader_context *ctx,
if (ctx->type != PIPE_SHADER_TESS_EVAL)
params[ctx->param_streamout_config = (*num_params)++] = i32;
else
- ctx->param_streamout_config = ctx->param_tess_offchip;
+ ctx->param_streamout_config = *num_params - 1;
params[ctx->param_streamout_write_index = (*num_params)++] = i32;
}
static void declare_tess_lds(struct si_shader_context *ctx)
{
struct gallivm_state *gallivm = &ctx->gallivm;
- struct lp_build_tgsi_context *bld_base = &ctx->soa.bld_base;
+ struct lp_build_tgsi_context *bld_base = &ctx->bld_base;
struct lp_build_context *uint = &bld_base->uint_bld;
unsigned lds_size = ctx->screen->b.chip_class >= CIK ? 65536 : 32768;
static void create_function(struct si_shader_context *ctx)
{
- struct lp_build_tgsi_context *bld_base = &ctx->soa.bld_base;
+ struct lp_build_tgsi_context *bld_base = &ctx->bld_base;
struct gallivm_state *gallivm = bld_base->base.gallivm;
struct si_shader *shader = ctx->shader;
LLVMTypeRef params[SI_NUM_PARAMS + SI_NUM_VERTEX_BUFFERS], v3i32;
if (shader->key.as_es) {
params[ctx->param_oc_lds = num_params++] = ctx->i32;
- params[ctx->param_tess_offchip = num_params++] = ctx->i32;
+ params[num_params++] = ctx->i32;
params[ctx->param_es2gs_offset = num_params++] = ctx->i32;
} else {
- params[ctx->param_tess_offchip = num_params++] = ctx->i32;
+ params[num_params++] = ctx->i32;
declare_streamout_params(ctx, &shader->selector->so,
params, ctx->i32, &num_params);
params[ctx->param_oc_lds = num_params++] = ctx->i32;
LOCAL_ADDR_SPACE);
if ((ctx->type == PIPE_SHADER_VERTEX && shader->key.as_ls) ||
- ctx->type == PIPE_SHADER_TESS_CTRL ||
- ctx->type == PIPE_SHADER_TESS_EVAL)
+ ctx->type == PIPE_SHADER_TESS_CTRL)
declare_tess_lds(ctx);
}
*/
static void preload_ring_buffers(struct si_shader_context *ctx)
{
- struct gallivm_state *gallivm =
- ctx->soa.bld_base.base.gallivm;
+ struct gallivm_state *gallivm = ctx->bld_base.base.gallivm;
+ LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef buf_ptr = LLVMGetParam(ctx->main_fn,
SI_PARAM_RW_BUFFERS);
LLVMValueRef offset = lp_build_const_int32(gallivm, ring);
ctx->esgs_ring =
- build_indexed_load_const(ctx, buf_ptr, offset);
+ ac_build_indexed_load_const(&ctx->ac, buf_ptr, offset);
}
if (ctx->shader->is_gs_copy_shader) {
- LLVMValueRef offset = lp_build_const_int32(gallivm, SI_VS_RING_GSVS);
+ LLVMValueRef offset = lp_build_const_int32(gallivm, SI_RING_GSVS);
ctx->gsvs_ring[0] =
- build_indexed_load_const(ctx, buf_ptr, offset);
- }
- if (ctx->type == PIPE_SHADER_GEOMETRY) {
- int i;
- for (i = 0; i < 4; i++) {
- LLVMValueRef offset = lp_build_const_int32(gallivm, SI_GS_RING_GSVS0 + i);
+ ac_build_indexed_load_const(&ctx->ac, buf_ptr, offset);
+ } else if (ctx->type == PIPE_SHADER_GEOMETRY) {
+ const struct si_shader_selector *sel = ctx->shader->selector;
+ struct lp_build_context *uint = &ctx->bld_base.uint_bld;
+ LLVMValueRef offset = lp_build_const_int32(gallivm, SI_RING_GSVS);
+ LLVMValueRef base_ring;
+
+ base_ring = ac_build_indexed_load_const(&ctx->ac, buf_ptr, offset);
+
+ /* The conceptual layout of the GSVS ring is
+ * v0c0 .. vLv0 v0c1 .. vLc1 ..
+ * but the real memory layout is swizzled across
+ * threads:
+ * t0v0c0 .. t15v0c0 t0v1c0 .. t15v1c0 ... t15vLcL
+ * t16v0c0 ..
+ * Override the buffer descriptor accordingly.
+ */
+ LLVMTypeRef v2i64 = LLVMVectorType(ctx->i64, 2);
+ uint64_t stream_offset = 0;
+
+ for (unsigned stream = 0; stream < 4; ++stream) {
+ unsigned num_components;
+ unsigned stride;
+ unsigned num_records;
+ LLVMValueRef ring, tmp;
+
+ num_components = sel->info.num_stream_output_components[stream];
+ if (!num_components)
+ continue;
- ctx->gsvs_ring[i] =
- build_indexed_load_const(ctx, buf_ptr, offset);
+ stride = 4 * num_components * sel->gs_max_out_vertices;
+
+ /* Limit on the stride field for <= CIK. */
+ assert(stride < (1 << 14));
+
+ num_records = 64;
+
+ ring = LLVMBuildBitCast(builder, base_ring, v2i64, "");
+ tmp = LLVMBuildExtractElement(builder, ring, uint->zero, "");
+ tmp = LLVMBuildAdd(builder, tmp,
+ LLVMConstInt(ctx->i64,
+ stream_offset, 0), "");
+ stream_offset += stride * 64;
+
+ ring = LLVMBuildInsertElement(builder, ring, tmp, uint->zero, "");
+ ring = LLVMBuildBitCast(builder, ring, ctx->v4i32, "");
+ tmp = LLVMBuildExtractElement(builder, ring, uint->one, "");
+ tmp = LLVMBuildOr(builder, tmp,
+ LLVMConstInt(ctx->i32,
+ S_008F04_STRIDE(stride) |
+ S_008F04_SWIZZLE_ENABLE(1), 0), "");
+ ring = LLVMBuildInsertElement(builder, ring, tmp, uint->one, "");
+ ring = LLVMBuildInsertElement(builder, ring,
+ LLVMConstInt(ctx->i32, num_records, 0),
+ LLVMConstInt(ctx->i32, 2, 0), "");
+ ring = LLVMBuildInsertElement(builder, ring,
+ LLVMConstInt(ctx->i32,
+ S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X) |
+ S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y) |
+ S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z) |
+ S_008F0C_DST_SEL_W(V_008F0C_SQ_SEL_W) |
+ S_008F0C_NUM_FORMAT(V_008F0C_BUF_NUM_FORMAT_FLOAT) |
+ S_008F0C_DATA_FORMAT(V_008F0C_BUF_DATA_FORMAT_32) |
+ S_008F0C_ELEMENT_SIZE(1) | /* element_size = 4 (bytes) */
+ S_008F0C_INDEX_STRIDE(1) | /* index_stride = 16 (elements) */
+ S_008F0C_ADD_TID_ENABLE(1),
+ 0),
+ LLVMConstInt(ctx->i32, 3, 0), "");
+ ring = LLVMBuildBitCast(builder, ring, ctx->v16i8, "");
+
+ ctx->gsvs_ring[stream] = ring;
}
}
}
LLVMValueRef param_rw_buffers,
unsigned param_pos_fixed_pt)
{
- struct lp_build_tgsi_context *bld_base =
- &ctx->soa.bld_base;
+ struct lp_build_tgsi_context *bld_base = &ctx->bld_base;
struct gallivm_state *gallivm = bld_base->base.gallivm;
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef slot, desc, offset, row, bit, address[2];
/* Load the buffer descriptor. */
slot = lp_build_const_int32(gallivm, SI_PS_CONST_POLY_STIPPLE);
- desc = build_indexed_load_const(ctx, param_rw_buffers, slot);
+ desc = ac_build_indexed_load_const(&ctx->ac, param_rw_buffers, slot);
/* The stipple pattern is 32x32, each row has 32 bits. */
offset = LLVMBuildMul(builder, address[1],
struct si_shader *shader,
struct pipe_debug_callback *debug,
unsigned processor,
- FILE *file)
+ FILE *file,
+ bool check_debug_option)
{
struct si_shader_config *conf = &shader->config;
unsigned num_inputs = shader->selector ? shader->selector->info.num_inputs : 0;
if (lds_per_wave)
max_simd_waves = MIN2(max_simd_waves, 16384 / lds_per_wave);
- if (file != stderr ||
+ if (!check_debug_option ||
r600_can_dump_shader(&sscreen->b, processor)) {
if (processor == PIPE_SHADER_FRAGMENT) {
fprintf(file, "*** SHADER CONFIG ***\n"
void si_shader_dump(struct si_screen *sscreen, struct si_shader *shader,
struct pipe_debug_callback *debug, unsigned processor,
- FILE *file)
+ FILE *file, bool check_debug_option)
{
- if (file != stderr ||
+ if (!check_debug_option ||
r600_can_dump_shader(&sscreen->b, processor))
si_dump_shader_key(processor, &shader->key, file);
- if (file != stderr && shader->binary.llvm_ir_string) {
+ if (!check_debug_option && shader->binary.llvm_ir_string) {
fprintf(file, "\n%s - main shader part - LLVM IR:\n\n",
si_get_shader_name(shader, processor));
fprintf(file, "%s\n", shader->binary.llvm_ir_string);
}
- if (file != stderr ||
+ if (!check_debug_option ||
(r600_can_dump_shader(&sscreen->b, processor) &&
!(sscreen->b.debug_flags & DBG_NO_ASM))) {
fprintf(file, "\n%s:\n", si_get_shader_name(shader, processor));
fprintf(file, "\n");
}
- si_shader_dump_stats(sscreen, shader, debug, processor, file);
+ si_shader_dump_stats(sscreen, shader, debug, processor, file,
+ check_debug_option);
}
int si_compile_llvm(struct si_screen *sscreen,
if (!(sscreen->b.debug_flags & (DBG_NO_IR | DBG_PREOPT_IR))) {
fprintf(stderr, "%s LLVM IR:\n\n", name);
- LLVMDumpModule(mod);
+ ac_dump_module(mod);
fprintf(stderr, "\n");
}
}
struct si_shader *shader;
struct gallivm_state *gallivm = &ctx.gallivm;
LLVMBuilderRef builder;
- struct lp_build_tgsi_context *bld_base = &ctx.soa.bld_base;
+ struct lp_build_tgsi_context *bld_base = &ctx.bld_base;
struct lp_build_context *uint = &bld_base->uint_bld;
struct si_shader_output_values *outputs;
struct tgsi_shader_info *gsinfo = &gs_selector->info;
for (int stream = 0; stream < 4; stream++) {
LLVMBasicBlockRef bb;
+ unsigned offset;
if (!gsinfo->num_stream_output_components[stream])
continue;
LLVMPositionBuilderAtEnd(builder, bb);
/* Fetch vertex data from GSVS ring */
+ offset = 0;
for (i = 0; i < gsinfo->num_outputs; ++i) {
for (unsigned chan = 0; chan < 4; chan++) {
- if (outputs[i].vertex_stream[chan] != stream) {
- outputs[i].values[chan] = ctx.soa.bld_base.base.undef;
+ if (!(gsinfo->output_usagemask[i] & (1 << chan)) ||
+ outputs[i].vertex_stream[chan] != stream) {
+ outputs[i].values[chan] = ctx.bld_base.base.undef;
continue;
}
args[2] = lp_build_const_int32(
gallivm,
- (i * 4 + chan) * gs_selector->gs_max_out_vertices * 16 * 4);
+ offset * gs_selector->gs_max_out_vertices * 16 * 4);
+ offset++;
outputs[i].values[chan] =
LLVMBuildBitCast(gallivm->builder,
/* Dump LLVM IR before any optimization passes */
if (sscreen->b.debug_flags & DBG_PREOPT_IR &&
r600_can_dump_shader(&sscreen->b, PIPE_SHADER_GEOMETRY))
- LLVMDumpModule(bld_base->base.gallivm->module);
+ ac_dump_module(bld_base->base.gallivm->module);
si_llvm_finalize_module(&ctx,
r600_extra_shader_checks(&sscreen->b, PIPE_SHADER_GEOMETRY));
if (r600_can_dump_shader(&sscreen->b, PIPE_SHADER_GEOMETRY))
fprintf(stderr, "GS Copy Shader:\n");
si_shader_dump(sscreen, ctx.shader, debug,
- PIPE_SHADER_GEOMETRY, stderr);
+ PIPE_SHADER_GEOMETRY, stderr, true);
r = si_shader_binary_upload(sscreen, ctx.shader);
}
fprintf(f, " part.vs.epilog.export_prim_id = %u\n", key->part.vs.epilog.export_prim_id);
fprintf(f, " as_es = %u\n", key->as_es);
fprintf(f, " as_ls = %u\n", key->as_ls);
- fprintf(f, " mono.vs.fix_fetch = 0x%x\n", key->mono.vs.fix_fetch);
+ fprintf(f, " mono.vs.fix_fetch = 0x%"PRIx64"\n", key->mono.vs.fix_fetch);
break;
case PIPE_SHADER_TESS_CTRL:
(shader && shader->selector) ? &shader->selector->info : NULL,
(shader && shader->selector) ? shader->selector->tokens : NULL);
- bld_base = &ctx->soa.bld_base;
+ bld_base = &ctx->bld_base;
bld_base->emit_fetch_funcs[TGSI_FILE_CONSTANT] = fetch_constant;
bld_base->op_actions[TGSI_OPCODE_INTERP_CENTROID] = interp_action;
struct si_shader *shader)
{
struct si_shader_selector *sel = shader->selector;
- struct lp_build_tgsi_context *bld_base = &ctx->soa.bld_base;
+ struct lp_build_tgsi_context *bld_base = &ctx->bld_base;
switch (ctx->type) {
case PIPE_SHADER_VERTEX:
shader->info.uses_instanceid = sel->info.uses_instanceid;
- bld_base = &ctx.soa.bld_base;
+ bld_base = &ctx.bld_base;
ctx.load_system_value = declare_system_value;
if (!si_compile_tgsi_main(&ctx, shader)) {
/* Dump LLVM IR before any optimization passes */
if (sscreen->b.debug_flags & DBG_PREOPT_IR &&
r600_can_dump_shader(&sscreen->b, ctx.type))
- LLVMDumpModule(mod);
+ ac_dump_module(mod);
si_llvm_finalize_module(&ctx,
r600_extra_shader_checks(&sscreen->b, ctx.type));
union si_shader_part_key *key)
{
struct gallivm_state *gallivm = &ctx->gallivm;
- struct lp_build_tgsi_context *bld_base = &ctx->soa.bld_base;
+ struct lp_build_tgsi_context *bld_base = &ctx->bld_base;
LLVMTypeRef params[5];
int num_params, i;
union si_shader_part_key *key)
{
struct gallivm_state *gallivm = &ctx->gallivm;
- struct lp_build_tgsi_context *bld_base = &ctx->soa.bld_base;
+ struct lp_build_tgsi_context *bld_base = &ctx->bld_base;
LLVMTypeRef params[16];
LLVMValueRef func;
int last_sgpr, num_params;
union si_shader_part_key *key)
{
struct gallivm_state *gallivm = &ctx->gallivm;
- struct lp_build_tgsi_context *bld_base = &ctx->soa.bld_base;
+ struct lp_build_tgsi_context *bld_base = &ctx->bld_base;
LLVMTypeRef params[16+8*4+3];
LLVMValueRef depth = NULL, stencil = NULL, samplemask = NULL;
int last_sgpr, num_params, i;
si_fix_resource_usage(sscreen, shader);
si_shader_dump(sscreen, shader, debug, sel->info.processor,
- stderr);
+ stderr, true);
/* Upload. */
r = si_shader_binary_upload(sscreen, shader);