From: Marek Olšák <marek.ol...@amd.com>
---
src/gallium/drivers/radeonsi/si_pipe.c | 2 +-
src/gallium/drivers/radeonsi/si_shader.c | 96 ++++++-------
src/gallium/drivers/radeonsi/si_shader_internal.h | 70 +++++-----
.../drivers/radeonsi/si_shader_tgsi_setup.c | 150 ++++++++++-----------
4 files changed, 159 insertions(+), 159 deletions(-)
diff --git a/src/gallium/drivers/radeonsi/si_pipe.c
b/src/gallium/drivers/radeonsi/si_pipe.c
index 7924375..a9faa75 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.c
+++ b/src/gallium/drivers/radeonsi/si_pipe.c
@@ -119,21 +119,21 @@ static void si_emit_string_marker(struct pipe_context
*ctx,
struct si_context *sctx = (struct si_context *)ctx;
dd_parse_apitrace_marker(string, len, &sctx->apitrace_call_number);
}
static LLVMTargetMachineRef
si_create_llvm_target_machine(struct si_screen *sscreen)
{
const char *triple = "amdgcn--";
- return LLVMCreateTargetMachine(radeon_llvm_get_r600_target(triple),
triple,
+ return LLVMCreateTargetMachine(si_llvm_get_amdgpu_target(triple),
triple,
r600_get_llvm_processor_name(sscreen->b.family),
#if HAVE_LLVM >= 0x0308
sscreen->b.debug_flags & DBG_SI_SCHED ?
SI_LLVM_DEFAULT_FEATURES
",+si-scheduler" :
#endif
SI_LLVM_DEFAULT_FEATURES,
LLVMCodeGenLevelDefault,
LLVMRelocDefault,
LLVMCodeModelDefault);
}
diff --git a/src/gallium/drivers/radeonsi/si_shader.c
b/src/gallium/drivers/radeonsi/si_shader.c
index bca07ac..cbf2090 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -478,21 +478,21 @@ static LLVMValueRef get_bounded_indirect_index(struct
si_shader_context *ctx,
{
LLVMValueRef result = get_indirect_index(ctx, ind, rel_index);
/* LLVM 3.8: If indirect resource indexing is used:
* - SI & CIK hang
* - VI crashes
*/
if (HAVE_LLVM <= 0x0308)
return LLVMGetUndef(ctx->i32);
- return radeon_llvm_bound_index(ctx, result, num);
+ return si_llvm_bound_index(ctx, result, num);
}
/**
* Calculate a dword address given an input or output register and a stride.
*/
static LLVMValueRef get_dw_address(struct si_shader_context *ctx,
const struct tgsi_full_dst_register *dst,
const struct tgsi_full_src_register *src,
LLVMValueRef vertex_dw_stride,
@@ -869,21 +869,21 @@ static LLVMValueRef buffer_load(struct
lp_build_tgsi_context *bld_base,
return LLVMBuildExtractElement(gallivm->builder, value,
lp_build_const_int32(gallivm, swizzle), "");
}
value = build_buffer_load(ctx, buffer, 1, NULL, base, offset,
swizzle * 4, 1, 0);
value2 = build_buffer_load(ctx, buffer, 1, NULL, base, offset,
swizzle * 4 + 4, 1, 0);
- return radeon_llvm_emit_fetch_64bit(bld_base, type, value, value2);
+ return si_llvm_emit_fetch_64bit(bld_base, type, value, value2);
}
/**
* Load from LDS.
*
* \param type output value type
* \param swizzle offset (typically 0..3); it can be ~0, which loads a
vec4
* \param dw_addr address in dwords
*/
static LLVMValueRef lds_load(struct lp_build_tgsi_context *bld_base,
@@ -906,21 +906,21 @@ static LLVMValueRef lds_load(struct lp_build_tgsi_context
*bld_base,
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);
if (tgsi_type_is_64bit(type)) {
LLVMValueRef value2;
dw_addr = lp_build_add(&bld_base->uint_bld, dw_addr,
lp_build_const_int32(gallivm, swizzle +
1));
value2 = build_indexed_load(ctx, ctx->lds, dw_addr, false);
- return radeon_llvm_emit_fetch_64bit(bld_base, type, value,
value2);
+ return si_llvm_emit_fetch_64bit(bld_base, type, value, value2);
}
return LLVMBuildBitCast(gallivm->builder, value,
tgsi2llvmtype(bld_base, type), "");
}
/**
* Store to LDS.
*
* \param swizzle offset (typically 0..3)
@@ -1008,21 +1008,21 @@ static void store_output_tcs(struct
lp_build_tgsi_context *bld_base,
unsigned chan_index;
LLVMValueRef dw_addr, stride;
LLVMValueRef rw_buffers, buffer, base, buf_addr;
LLVMValueRef values[4];
/* Only handle per-patch and per-vertex outputs here.
* Vectors will be lowered to scalars and this function will be called
again.
*/
if (reg->Register.File != TGSI_FILE_OUTPUT ||
(dst[0] && LLVMGetTypeKind(LLVMTypeOf(dst[0])) ==
LLVMVectorTypeKind)) {
- radeon_llvm_emit_store(bld_base, inst, info, dst);
+ si_llvm_emit_store(bld_base, inst, info, dst);
return;
}
if (reg->Register.Dimension) {
stride = unpack_param(ctx, SI_PARAM_TCS_OUT_LAYOUT, 13, 8);
dw_addr = get_tcs_out_current_patch_offset(ctx);
dw_addr = get_dw_address(ctx, reg, NULL, stride, dw_addr);
} else {
dw_addr = get_tcs_out_current_patch_data_offset(ctx);
dw_addr = get_dw_address(ctx, reg, NULL, NULL, dw_addr);
@@ -1034,21 +1034,21 @@ static void store_output_tcs(struct
lp_build_tgsi_context *bld_base,
lp_build_const_int32(gallivm, SI_HS_RING_TESS_OFFCHIP));
base = LLVMGetParam(ctx->main_fn, ctx->param_oc_lds);
buf_addr = get_tcs_tes_buffer_address_from_reg(ctx, reg, NULL);
TGSI_FOR_EACH_DST0_ENABLED_CHANNEL(inst, chan_index) {
LLVMValueRef value = dst[chan_index];
if (inst->Instruction.Saturate)
- value = radeon_llvm_saturate(bld_base, value);
+ value = si_llvm_saturate(bld_base, value);
lds_store(bld_base, chan_index, dw_addr, value);
value = LLVMBuildBitCast(gallivm->builder, value, ctx->i32, "");
values[chan_index] = value;
if (inst->Dst[0].Register.WriteMask != 0xF) {
build_tbuffer_store_dwords(ctx, buffer, value, 1,
buf_addr, base,
4 * chan_index);
@@ -1127,22 +1127,22 @@ static LLVMValueRef fetch_input_gs(
"llvm.SI.buffer.load.dword.i32.i32",
ctx->i32, args, 9,
LLVMReadOnlyAttribute);
if (tgsi_type_is_64bit(type)) {
LLVMValueRef value2;
args[2] = lp_build_const_int32(gallivm, (param * 4 + swizzle +
1) * 256);
value2 = lp_build_intrinsic(gallivm->builder,
"llvm.SI.buffer.load.dword.i32.i32",
ctx->i32, args, 9,
LLVMReadOnlyAttribute);
- return radeon_llvm_emit_fetch_64bit(bld_base, type,
- value, value2);
+ return si_llvm_emit_fetch_64bit(bld_base, type,
+ value, value2);
}
return LLVMBuildBitCast(gallivm->builder,
value,
tgsi2llvmtype(bld_base, type), "");
}
static int lookup_interp_param_index(unsigned interpolate, unsigned location)
{
switch (interpolate) {
case TGSI_INTERPOLATE_CONSTANT:
@@ -1808,22 +1808,22 @@ static LLVMValueRef fetch_constant(
desc = load_const_buffer_desc(ctx, buf);
c0 = buffer_load_const(ctx, desc,
LLVMConstInt(ctx->i32, idx * 4, 0));
if (!tgsi_type_is_64bit(type))
return bitcast(bld_base, type, c0);
else {
c1 = buffer_load_const(ctx, desc,
LLVMConstInt(ctx->i32,
(idx + 1) * 4, 0));
- return radeon_llvm_emit_fetch_64bit(bld_base, type,
- c0, c1);
+ return si_llvm_emit_fetch_64bit(bld_base, type,
+ c0, c1);
}
}
if (reg->Register.Dimension && reg->Dimension.Indirect) {
LLVMValueRef ptr = LLVMGetParam(ctx->main_fn,
SI_PARAM_CONST_BUFFERS);
LLVMValueRef index;
index = get_bounded_indirect_index(ctx, ®->DimIndirect,
reg->Dimension.Index,
SI_NUM_CONST_BUFFERS);
bufp = build_indexed_load_const(ctx, ptr, index);
@@ -1843,22 +1843,22 @@ static LLVMValueRef fetch_constant(
else {
LLVMValueRef addr2, result2;
addr2 = ctx->soa.addr[ireg->Index][ireg->Swizzle + 1];
addr2 = LLVMBuildLoad(base->gallivm->builder, addr2, "load addr
reg2");
addr2 = lp_build_mul_imm(&bld_base->uint_bld, addr2, 16);
addr2 = lp_build_add(&bld_base->uint_bld, addr2,
lp_build_const_int32(base->gallivm, idx *
4));
result2 = buffer_load_const(ctx, bufp, addr2);
- result = radeon_llvm_emit_fetch_64bit(bld_base, type,
- result, result2);
+ result = si_llvm_emit_fetch_64bit(bld_base, type,
+ result, result2);
}
return result;
}
/* Upper 16 bits must be zero. */
static LLVMValueRef si_llvm_pack_two_int16(struct gallivm_state *gallivm,
LLVMValueRef val[2])
{
return LLVMBuildOr(gallivm->builder, val[0],
LLVMBuildShl(gallivm->builder, val[1],
@@ -1961,21 +1961,21 @@ static void si_llvm_init_export_args(struct
lp_build_tgsi_context *bld_base,
ctx->i32, pack_args, 2,
LLVMReadNoneAttribute);
args[chan + 5] =
LLVMBuildBitCast(base->gallivm->builder,
packed, ctx->f32, "");
}
break;
case V_028714_SPI_SHADER_UNORM16_ABGR:
for (chan = 0; chan < 4; chan++) {
- val[chan] = radeon_llvm_saturate(bld_base,
values[chan]);
+ val[chan] = si_llvm_saturate(bld_base, values[chan]);
val[chan] = LLVMBuildFMul(builder, val[chan],
lp_build_const_float(gallivm, 65535),
"");
val[chan] = LLVMBuildFAdd(builder, val[chan],
lp_build_const_float(gallivm, 0.5),
"");
val[chan] = LLVMBuildFPToUI(builder, val[chan],
ctx->i32, "");
}
args[4] = uint->one; /* COMPR flag */
args[5] = bitcast(bld_base, TGSI_TYPE_FLOAT,
@@ -2806,21 +2806,21 @@ static void si_llvm_emit_vs_epilogue(struct
lp_build_tgsi_context *bld_base)
cond = LLVMGetParam(ctx->main_fn,
SI_PARAM_VS_STATE_BITS);
cond = LLVMBuildTrunc(gallivm->builder, cond,
ctx->i1, "");
lp_build_if(&if_ctx, gallivm, cond);
}
for (j = 0; j < 4; j++) {
addr = ctx->soa.outputs[i][j];
val = LLVMBuildLoad(gallivm->builder, addr, "");
- val = radeon_llvm_saturate(bld_base, val);
+ val = si_llvm_saturate(bld_base, val);
LLVMBuildStore(gallivm->builder, val, addr);
}
}
if (cond)
lp_build_endif(&if_ctx);
}
for (i = 0; i < info->num_outputs; i++) {
outputs[i].name = info->output_semantic_name[i];
@@ -2959,21 +2959,21 @@ static void si_export_mrt_color(struct
lp_build_tgsi_context *bld_base,
unsigned samplemask_param,
bool is_last, struct si_ps_exports *exp)
{
struct si_shader_context *ctx = si_shader_context(bld_base);
struct lp_build_context *base = &bld_base->base;
int i;
/* Clamp color */
if (ctx->shader->key.ps.epilog.clamp_color)
for (i = 0; i < 4; i++)
- color[i] = radeon_llvm_saturate(bld_base, color[i]);
+ color[i] = si_llvm_saturate(bld_base, color[i]);
/* Alpha to one */
if (ctx->shader->key.ps.epilog.alpha_to_one)
color[3] = base->one;
/* Alpha test */
if (index == 0 &&
ctx->shader->key.ps.epilog.alpha_func != PIPE_FUNC_ALWAYS)
si_alpha_test(bld_base, color[3]);
@@ -4545,21 +4545,21 @@ static void tex_fetch_args(
}
/* TC-compatible HTILE promotes Z16 and Z24 to Z32_FLOAT,
* so the depth comparison value isn't clamped for Z16 and
* Z24 anymore. Do it manually here.
*
* It's unnecessary if the original texture format was
* Z32_FLOAT, but we don't know that here.
*/
if (ctx->screen->b.chip_class == VI)
- z = radeon_llvm_saturate(bld_base, z);
+ z = si_llvm_saturate(bld_base, z);
address[count++] = z;
}
/* Pack user derivatives */
if (opcode == TGSI_OPCODE_TXD) {
int param, num_src_deriv_channels;
switch (target) {
case TGSI_TEXTURE_3D:
@@ -5381,23 +5381,23 @@ static const struct lp_build_tgsi_action interp_action
= {
.emit = build_interp_intrinsic,
};
static void si_create_function(struct si_shader_context *ctx,
LLVMTypeRef *returns, unsigned num_returns,
LLVMTypeRef *params, unsigned num_params,
int last_sgpr)
{
int i;
- radeon_llvm_create_func(ctx, returns, num_returns,
- params, num_params);
- radeon_llvm_shader_type(ctx->main_fn, ctx->type);
+ si_llvm_create_func(ctx, returns, num_returns,
+ params, num_params);
+ si_llvm_shader_type(ctx->main_fn, ctx->type);
ctx->return_value = LLVMGetUndef(ctx->return_type);
for (i = 0; i <= last_sgpr; ++i) {
LLVMValueRef P = LLVMGetParam(ctx->main_fn, i);
/* The combination of:
* - ByVal
* - dereferenceable
* - invariant.load
* allows the optimization passes to move loads and reduces
@@ -5705,47 +5705,47 @@ static void create_function(struct si_shader_context
*ctx)
}
assert(num_params <= ARRAY_SIZE(params));
si_create_function(ctx, returns, num_returns, params,
num_params, last_sgpr);
/* Reserve register locations for VGPR inputs the PS prolog may need. */
if (ctx->type == PIPE_SHADER_FRAGMENT &&
!ctx->is_monolithic) {
- radeon_llvm_add_attribute(ctx->main_fn,
- "InitialPSInputAddr",
- S_0286D0_PERSP_SAMPLE_ENA(1) |
- S_0286D0_PERSP_CENTER_ENA(1) |
- S_0286D0_PERSP_CENTROID_ENA(1) |
- S_0286D0_LINEAR_SAMPLE_ENA(1) |
- S_0286D0_LINEAR_CENTER_ENA(1) |
- S_0286D0_LINEAR_CENTROID_ENA(1) |
- S_0286D0_FRONT_FACE_ENA(1) |
- S_0286D0_POS_FIXED_PT_ENA(1));
+ si_llvm_add_attribute(ctx->main_fn,
+ "InitialPSInputAddr",
+ S_0286D0_PERSP_SAMPLE_ENA(1) |
+ S_0286D0_PERSP_CENTER_ENA(1) |
+ S_0286D0_PERSP_CENTROID_ENA(1) |
+ S_0286D0_LINEAR_SAMPLE_ENA(1) |
+ S_0286D0_LINEAR_CENTER_ENA(1) |
+ S_0286D0_LINEAR_CENTROID_ENA(1) |
+ S_0286D0_FRONT_FACE_ENA(1) |
+ S_0286D0_POS_FIXED_PT_ENA(1));
} else if (ctx->type == PIPE_SHADER_COMPUTE) {
const unsigned *properties = shader->selector->info.properties;
unsigned max_work_group_size =
properties[TGSI_PROPERTY_CS_FIXED_BLOCK_WIDTH] *
properties[TGSI_PROPERTY_CS_FIXED_BLOCK_HEIGHT] *
properties[TGSI_PROPERTY_CS_FIXED_BLOCK_DEPTH];
if (!max_work_group_size) {
/* This is a variable group size compute shader,
* compile it for the maximum possible group size.
*/
max_work_group_size = SI_MAX_VARIABLE_THREADS_PER_BLOCK;
}
- radeon_llvm_add_attribute(ctx->main_fn,
- "amdgpu-max-work-group-size",
- max_work_group_size);
+ si_llvm_add_attribute(ctx->main_fn,
+ "amdgpu-max-work-group-size",
+ max_work_group_size);
}
shader->info.num_input_sgprs = 0;
shader->info.num_input_vgprs = 0;
for (i = 0; i <= last_sgpr; ++i)
shader->info.num_input_sgprs += llvm_get_type_size(params[i]) /
4;
/* Unused fragment shader inputs are eliminated by the compiler,
* so we don't know yet how many there will be.
@@ -6240,21 +6240,21 @@ int si_compile_llvm(struct si_screen *sscreen,
}
}
if (sscreen->record_llvm_ir) {
char *ir = LLVMPrintModuleToString(mod);
binary->llvm_ir_string = strdup(ir);
LLVMDisposeMessage(ir);
}
if (!si_replace_shader(count, binary)) {
- r = radeon_llvm_compile(mod, binary, tm, debug);
+ r = si_llvm_compile(mod, binary, tm, debug);
if (r)
return r;
}
si_shader_binary_read_config(binary, conf, 0);
/* Enable 64-bit and 16-bit denormals, because there is no performance
* cost.
*
* If denormals are enabled, all floating-point output modifiers are
@@ -6356,37 +6356,37 @@ static int si_generate_gs_copy_shader(struct si_screen
*sscreen,
si_llvm_export_vs(bld_base, outputs, gsinfo->num_outputs);
LLVMBuildRetVoid(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);
- radeon_llvm_finalize_module(ctx,
+ si_llvm_finalize_module(ctx,
r600_extra_shader_checks(&sscreen->b, PIPE_SHADER_GEOMETRY));
r = si_compile_llvm(sscreen, &ctx->shader->binary,
&ctx->shader->config, ctx->tm,
bld_base->base.gallivm->module,
debug, PIPE_SHADER_GEOMETRY,
"GS Copy Shader");
if (!r) {
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);
r = si_shader_binary_upload(sscreen, ctx->shader);
}
- radeon_llvm_dispose(ctx);
+ si_llvm_dispose(ctx);
FREE(outputs);
return r;
}
static void si_dump_shader_key(unsigned shader, union si_shader_key *key,
FILE *f)
{
int i;
@@ -6443,21 +6443,21 @@ static void si_dump_shader_key(unsigned shader, union
si_shader_key *key,
static void si_init_shader_ctx(struct si_shader_context *ctx,
struct si_screen *sscreen,
struct si_shader *shader,
LLVMTargetMachineRef tm)
{
struct lp_build_tgsi_context *bld_base;
struct lp_build_tgsi_action tmpl = {};
memset(ctx, 0, sizeof(*ctx));
- radeon_llvm_context_init(
+ si_llvm_context_init(
ctx, "amdgcn--",
(shader && shader->selector) ? &shader->selector->info : NULL,
(shader && shader->selector) ? shader->selector->tokens : NULL);
si_shader_context_init_alu(&ctx->soa.bld_base);
ctx->tm = tm;
ctx->screen = sscreen;
if (shader && shader->selector)
ctx->type = shader->selector->info.processor;
else
ctx->type = -1;
@@ -6637,31 +6637,31 @@ int si_compile_tgsi_shader(struct si_screen *sscreen,
}
si_llvm_build_ret(&ctx, ctx.return_value);
mod = bld_base->base.gallivm->module;
/* 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);
- radeon_llvm_finalize_module(&ctx,
+ si_llvm_finalize_module(&ctx,
r600_extra_shader_checks(&sscreen->b,
ctx.type));
r = si_compile_llvm(sscreen, &shader->binary, &shader->config, tm,
mod, debug, ctx.type, "TGSI shader");
if (r) {
fprintf(stderr, "LLVM failed to compile shader\n");
goto out;
}
- radeon_llvm_dispose(&ctx);
+ si_llvm_dispose(&ctx);
/* Validate SGPR and VGPR usage for compute to detect compiler bugs.
* LLVM 3.9svn has this bug.
*/
if (sel->type == PIPE_SHADER_COMPUTE) {
unsigned *props = sel->info.properties;
unsigned wave_size = 64;
unsigned max_vgprs = 256;
unsigned max_sgprs = sscreen->b.chip_class >= VI ? 800 : 512;
unsigned max_sgprs_per_wave = 128;
@@ -6904,29 +6904,29 @@ static bool si_compile_vs_prolog(struct si_screen
*sscreen,
LLVMGetParam(func, SI_SGPR_BASE_VERTEX),
"");
}
index = LLVMBuildBitCast(gallivm->builder, index, ctx.f32, "");
ret = LLVMBuildInsertValue(gallivm->builder, ret, index,
num_params++, "");
}
/* Compile. */
si_llvm_build_ret(&ctx, ret);
- radeon_llvm_finalize_module(&ctx,
+ si_llvm_finalize_module(&ctx,
r600_extra_shader_checks(&sscreen->b, PIPE_SHADER_VERTEX));
if (si_compile_llvm(sscreen, &out->binary, &out->config, tm,
gallivm->module, debug, ctx.type,
"Vertex Shader Prolog"))
status = false;
- radeon_llvm_dispose(&ctx);
+ si_llvm_dispose(&ctx);
return status;
}
/**
* Compile the vertex shader epilog. This is also used by the tessellation
* evaluation shader compiled as VS.
*
* The input is PrimitiveID.
*
* If PrimitiveID is required by the pixel shader, export it.
@@ -6977,29 +6977,29 @@ static bool si_compile_vs_epilog(struct si_screen
*sscreen,
args[7] = uint->undef; /* Z */
args[8] = uint->undef; /* W */
lp_build_intrinsic(base->gallivm->builder, "llvm.SI.export",
LLVMVoidTypeInContext(base->gallivm->context),
args, 9, 0);
}
/* Compile. */
LLVMBuildRetVoid(gallivm->builder);
- radeon_llvm_finalize_module(&ctx,
+ si_llvm_finalize_module(&ctx,
r600_extra_shader_checks(&sscreen->b, PIPE_SHADER_VERTEX));
if (si_compile_llvm(sscreen, &out->binary, &out->config, tm,
gallivm->module, debug, ctx.type,
"Vertex Shader Epilog"))
status = false;
- radeon_llvm_dispose(&ctx);
+ si_llvm_dispose(&ctx);
return status;
}
/**
* Create & compile a vertex shader epilog. This a helper used by VS and TES.
*/
static bool si_get_vs_epilog(struct si_screen *sscreen,
LLVMTargetMachineRef tm,
struct si_shader *shader,
struct pipe_debug_callback *debug,
@@ -7131,29 +7131,29 @@ static bool si_compile_tcs_epilog(struct si_screen
*sscreen,
declare_tess_lds(&ctx);
func = ctx.main_fn;
si_write_tess_factors(bld_base,
LLVMGetParam(func, last_sgpr + 1),
LLVMGetParam(func, last_sgpr + 2),
LLVMGetParam(func, last_sgpr + 3));
/* Compile. */
LLVMBuildRetVoid(gallivm->builder);
- radeon_llvm_finalize_module(&ctx,
+ si_llvm_finalize_module(&ctx,
r600_extra_shader_checks(&sscreen->b, PIPE_SHADER_TESS_CTRL));
if (si_compile_llvm(sscreen, &out->binary, &out->config, tm,
gallivm->module, debug, ctx.type,
"Tessellation Control Shader Epilog"))
status = false;
- radeon_llvm_dispose(&ctx);
+ si_llvm_dispose(&ctx);
return status;
}
/**
* Select and compile (or reuse) TCS parts (epilog).
*/
static bool si_shader_select_tcs_parts(struct si_screen *sscreen,
LLVMTargetMachineRef tm,
struct si_shader *shader,
struct pipe_debug_callback *debug)
@@ -7416,29 +7416,29 @@ static bool si_compile_ps_prolog(struct si_screen
*sscreen,
}
/* Tell LLVM to insert WQM instruction sequence when needed. */
if (key->ps_prolog.wqm) {
LLVMAddTargetDependentFunctionAttr(func,
"amdgpu-ps-wqm-outputs", "");
}
/* Compile. */
si_llvm_build_ret(&ctx, ret);
- radeon_llvm_finalize_module(&ctx,
+ si_llvm_finalize_module(&ctx,
r600_extra_shader_checks(&sscreen->b, PIPE_SHADER_FRAGMENT));
if (si_compile_llvm(sscreen, &out->binary, &out->config, tm,
gallivm->module, debug, ctx.type,
"Fragment Shader Prolog"))
status = false;
- radeon_llvm_dispose(&ctx);
+ si_llvm_dispose(&ctx);
return status;
}
/**
* Compile the pixel shader epilog. This handles everything that must be
* emulated for pixel shader exports. (alpha-test, format conversions, etc)
*/
static bool si_compile_ps_epilog(struct si_screen *sscreen,
LLVMTargetMachineRef tm,
struct pipe_debug_callback *debug,
@@ -7479,21 +7479,21 @@ static bool si_compile_ps_epilog(struct si_screen
*sscreen,
last_sgpr + 1 + PS_EPILOG_SAMPLEMASK_MIN_LOC + 1);
assert(num_params <= ARRAY_SIZE(params));
for (i = last_sgpr + 1; i < num_params; i++)
params[i] = ctx.f32;
/* Create the function. */
si_create_function(&ctx, NULL, 0, params, num_params, last_sgpr);
/* Disable elimination of unused inputs. */
- radeon_llvm_add_attribute(ctx.main_fn,
+ si_llvm_add_attribute(ctx.main_fn,
"InitialPSInputAddr", 0xffffff);
/* Process colors. */
unsigned vgpr = last_sgpr + 1;
unsigned colors_written = key->ps_epilog.colors_written;
int last_color_export = -1;
/* Find the last color export. */
if (!key->ps_epilog.writes_z &&
!key->ps_epilog.writes_stencil &&
@@ -7537,29 +7537,29 @@ static bool si_compile_ps_epilog(struct si_screen
*sscreen,
if (depth || stencil || samplemask)
si_export_mrt_z(bld_base, depth, stencil, samplemask, &exp);
else if (last_color_export == -1)
si_export_null(bld_base);
if (exp.num)
si_emit_ps_exports(&ctx, &exp);
/* Compile. */
LLVMBuildRetVoid(gallivm->builder);
- radeon_llvm_finalize_module(&ctx,
+ si_llvm_finalize_module(&ctx,
r600_extra_shader_checks(&sscreen->b, PIPE_SHADER_FRAGMENT));
if (si_compile_llvm(sscreen, &out->binary, &out->config, tm,
gallivm->module, debug, ctx.type,
"Fragment Shader Epilog"))
status = false;
- radeon_llvm_dispose(&ctx);
+ si_llvm_dispose(&ctx);
return status;
}
/**
* Select and compile (or reuse) pixel shader parts (prolog & epilog).
*/
static bool si_shader_select_ps_parts(struct si_screen *sscreen,
LLVMTargetMachineRef tm,
struct si_shader *shader,
struct pipe_debug_callback *debug)
diff --git a/src/gallium/drivers/radeonsi/si_shader_internal.h
b/src/gallium/drivers/radeonsi/si_shader_internal.h
index 37001c0..84d8ed5 100644
--- a/src/gallium/drivers/radeonsi/si_shader_internal.h
+++ b/src/gallium/drivers/radeonsi/si_shader_internal.h
@@ -36,21 +36,21 @@ struct pipe_debug_callback;
struct radeon_shader_binary;
#define RADEON_LLVM_MAX_INPUT_SLOTS 32
#define RADEON_LLVM_MAX_INPUTS 32 * 4
#define RADEON_LLVM_MAX_OUTPUTS 32 * 4
#define RADEON_LLVM_INITIAL_CF_DEPTH 4
#define RADEON_LLVM_MAX_SYSTEM_VALUES 4
-struct radeon_llvm_flow;
+struct si_llvm_flow;
struct si_shader_context {
struct lp_build_tgsi_soa_context soa;
struct gallivm_state gallivm;
struct si_shader *shader;
struct si_screen *screen;
unsigned type; /* PIPE_SHADER_* specifies the type of shader. */
bool is_gs_copy_shader;
/* Whether to generate the optimized shader variant compiled as a whole
@@ -82,21 +82,21 @@ struct si_shader_context {
LLVMValueRef outputs[RADEON_LLVM_MAX_OUTPUTS][TGSI_NUM_CHANNELS];
/** This pointer is used to contain the temporary values.
* The amount of temporary used in tgsi can't be bound to a max value
and
* thus we must allocate this array at runtime.
*/
LLVMValueRef *temps;
unsigned temps_count;
LLVMValueRef system_values[RADEON_LLVM_MAX_SYSTEM_VALUES];
- struct radeon_llvm_flow *flow;
+ struct si_llvm_flow *flow;
unsigned flow_depth;
unsigned flow_depth_max;
struct tgsi_array_info *temp_arrays;
LLVMValueRef *temp_array_allocas;
LLVMValueRef undef_alloca;
LLVMValueRef main_fn;
LLVMTypeRef return_type;
@@ -153,68 +153,68 @@ struct si_shader_context {
LLVMValueRef shared_memory;
};
static inline struct si_shader_context *
si_shader_context(struct lp_build_tgsi_context *bld_base)
{
return (struct si_shader_context*)bld_base;
}
-void radeon_llvm_add_attribute(LLVMValueRef F, const char *name, int value);
-void radeon_llvm_shader_type(LLVMValueRef F, unsigned type);
+void si_llvm_add_attribute(LLVMValueRef F, const char *name, int value);
+void si_llvm_shader_type(LLVMValueRef F, unsigned type);
-LLVMTargetRef radeon_llvm_get_r600_target(const char *triple);
+LLVMTargetRef si_llvm_get_amdgpu_target(const char *triple);
-unsigned radeon_llvm_compile(LLVMModuleRef M, struct radeon_shader_binary
*binary,
- LLVMTargetMachineRef tm,
- struct pipe_debug_callback *debug);
+unsigned si_llvm_compile(LLVMModuleRef M, struct radeon_shader_binary *binary,
+ LLVMTargetMachineRef tm,
+ struct pipe_debug_callback *debug);
LLVMTypeRef tgsi2llvmtype(struct lp_build_tgsi_context *bld_base,
enum tgsi_opcode_type type);
LLVMValueRef bitcast(struct lp_build_tgsi_context *bld_base,
enum tgsi_opcode_type type, LLVMValueRef value);
-LLVMValueRef radeon_llvm_bound_index(struct si_shader_context *ctx,
- LLVMValueRef index,
- unsigned num);
+LLVMValueRef si_llvm_bound_index(struct si_shader_context *ctx,
+ LLVMValueRef index,
+ unsigned num);
-void radeon_llvm_context_init(struct si_shader_context *ctx,
- const char *triple,
- const struct tgsi_shader_info *info,
- const struct tgsi_token *tokens);
+void si_llvm_context_init(struct si_shader_context *ctx,
+ const char *triple,
+ const struct tgsi_shader_info *info,
+ const struct tgsi_token *tokens);
-void radeon_llvm_create_func(struct si_shader_context *ctx,
- LLVMTypeRef *return_types, unsigned
num_return_elems,
- LLVMTypeRef *ParamTypes, unsigned ParamCount);
+void si_llvm_create_func(struct si_shader_context *ctx,
+ LLVMTypeRef *return_types, unsigned num_return_elems,
+ LLVMTypeRef *ParamTypes, unsigned ParamCount);
-void radeon_llvm_dispose(struct si_shader_context *ctx);
+void si_llvm_dispose(struct si_shader_context *ctx);
-void radeon_llvm_finalize_module(struct si_shader_context *ctx,
- bool run_verifier);
+void si_llvm_finalize_module(struct si_shader_context *ctx,
+ bool run_verifier);
-LLVMValueRef radeon_llvm_emit_fetch_64bit(struct lp_build_tgsi_context
*bld_base,
- enum tgsi_opcode_type type,
- LLVMValueRef ptr,
- LLVMValueRef ptr2);
+LLVMValueRef si_llvm_emit_fetch_64bit(struct lp_build_tgsi_context *bld_base,
+ enum tgsi_opcode_type type,
+ LLVMValueRef ptr,
+ LLVMValueRef ptr2);
-LLVMValueRef radeon_llvm_saturate(struct lp_build_tgsi_context *bld_base,
- LLVMValueRef value);
+LLVMValueRef si_llvm_saturate(struct lp_build_tgsi_context *bld_base,
+ LLVMValueRef value);
-LLVMValueRef radeon_llvm_emit_fetch(struct lp_build_tgsi_context *bld_base,
- const struct tgsi_full_src_register *reg,
- enum tgsi_opcode_type type,
- unsigned swizzle);
+LLVMValueRef si_llvm_emit_fetch(struct lp_build_tgsi_context *bld_base,
+ const struct tgsi_full_src_register *reg,
+ enum tgsi_opcode_type type,
+ unsigned swizzle);
-void radeon_llvm_emit_store(struct lp_build_tgsi_context *bld_base,
- const struct tgsi_full_instruction *inst,
- const struct tgsi_opcode_info *info,
- LLVMValueRef dst[4]);
+void si_llvm_emit_store(struct lp_build_tgsi_context *bld_base,
+ const struct tgsi_full_instruction *inst,
+ const struct tgsi_opcode_info *info,
+ LLVMValueRef dst[4]);
void si_shader_context_init_alu(struct lp_build_tgsi_context *bld_base);
void si_prepare_cube_coords(struct lp_build_tgsi_context *bld_base,
struct lp_build_emit_data *emit_data,
LLVMValueRef *coords_arg,
LLVMValueRef *derivs_arg);
#endif
diff --git a/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c
b/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c
index b0174b1..c7a08cc 100644
--- a/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c
+++ b/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c
@@ -35,64 +35,64 @@
#include "tgsi/tgsi_parse.h"
#include "util/u_math.h"
#include "util/u_memory.h"
#include "util/u_debug.h"
#include <stdio.h>
#include <llvm-c/Transforms/Scalar.h>
/* Data for if/else/endif and bgnloop/endloop control flow structures.
*/
-struct radeon_llvm_flow {
+struct si_llvm_flow {
/* Loop exit or next part of if/else/endif. */
LLVMBasicBlockRef next_block;
LLVMBasicBlockRef loop_entry_block;
};
#define CPU_STRING_LEN 30
#define FS_STRING_LEN 30
#define TRIPLE_STRING_LEN 7
/**
* Shader types for the LLVM backend.
*/
-enum radeon_llvm_shader_type {
+enum si_llvm_shader_type {
RADEON_LLVM_SHADER_PS = 0,
RADEON_LLVM_SHADER_VS = 1,
RADEON_LLVM_SHADER_GS = 2,
RADEON_LLVM_SHADER_CS = 3,
};
-enum radeon_llvm_calling_convention {
+enum si_llvm_calling_convention {
RADEON_LLVM_AMDGPU_VS = 87,
RADEON_LLVM_AMDGPU_GS = 88,
RADEON_LLVM_AMDGPU_PS = 89,
RADEON_LLVM_AMDGPU_CS = 90,
};
-void radeon_llvm_add_attribute(LLVMValueRef F, const char *name, int value)
+void si_llvm_add_attribute(LLVMValueRef F, const char *name, int value)
{
char str[16];
snprintf(str, sizeof(str), "%i", value);
LLVMAddTargetDependentFunctionAttr(F, name, str);
}
/**
* Set the shader type we want to compile
*
* @param type shader type to set
*/
-void radeon_llvm_shader_type(LLVMValueRef F, unsigned type)
+void si_llvm_shader_type(LLVMValueRef F, unsigned type)
{
- enum radeon_llvm_shader_type llvm_type;
- enum radeon_llvm_calling_convention calling_conv;
+ enum si_llvm_shader_type llvm_type;
+ enum si_llvm_calling_convention calling_conv;
switch (type) {
case PIPE_SHADER_VERTEX:
case PIPE_SHADER_TESS_CTRL:
case PIPE_SHADER_TESS_EVAL:
llvm_type = RADEON_LLVM_SHADER_VS;
calling_conv = RADEON_LLVM_AMDGPU_VS;
break;
case PIPE_SHADER_GEOMETRY:
llvm_type = RADEON_LLVM_SHADER_GS;
@@ -106,68 +106,68 @@ void radeon_llvm_shader_type(LLVMValueRef F, unsigned
type)
llvm_type = RADEON_LLVM_SHADER_CS;
calling_conv = RADEON_LLVM_AMDGPU_CS;
break;
default:
unreachable("Unhandle shader type");
}
if (HAVE_LLVM >= 0x309)
LLVMSetFunctionCallConv(F, calling_conv);
else
- radeon_llvm_add_attribute(F, "ShaderType", llvm_type);
+ si_llvm_add_attribute(F, "ShaderType", llvm_type);
}
-static void init_r600_target()
+static void init_amdgpu_target()
{
gallivm_init_llvm_targets();
#if HAVE_LLVM < 0x0307
LLVMInitializeR600TargetInfo();
LLVMInitializeR600Target();
LLVMInitializeR600TargetMC();
LLVMInitializeR600AsmPrinter();
#else
LLVMInitializeAMDGPUTargetInfo();
LLVMInitializeAMDGPUTarget();
LLVMInitializeAMDGPUTargetMC();
LLVMInitializeAMDGPUAsmPrinter();
#endif
}
-static once_flag init_r600_target_once_flag = ONCE_FLAG_INIT;
+static once_flag init_amdgpu_target_once_flag = ONCE_FLAG_INIT;
-LLVMTargetRef radeon_llvm_get_r600_target(const char *triple)
+LLVMTargetRef si_llvm_get_amdgpu_target(const char *triple)
{
LLVMTargetRef target = NULL;
char *err_message = NULL;
- call_once(&init_r600_target_once_flag, init_r600_target);
+ call_once(&init_amdgpu_target_once_flag, init_amdgpu_target);
if (LLVMGetTargetFromTriple(triple, &target, &err_message)) {
fprintf(stderr, "Cannot find target for triple %s ", triple);
if (err_message) {
fprintf(stderr, "%s\n", err_message);
}
LLVMDisposeMessage(err_message);
return NULL;
}
return target;
}
-struct radeon_llvm_diagnostics {
+struct si_llvm_diagnostics {
struct pipe_debug_callback *debug;
unsigned retval;
};
-static void radeonDiagnosticHandler(LLVMDiagnosticInfoRef di, void *context)
+static void si_diagnostic_handler(LLVMDiagnosticInfoRef di, void *context)
{
- struct radeon_llvm_diagnostics *diag = (struct radeon_llvm_diagnostics
*)context;
+ struct si_llvm_diagnostics *diag = (struct si_llvm_diagnostics
*)context;
LLVMDiagnosticSeverity severity = LLVMGetDiagInfoSeverity(di);
char *description = LLVMGetDiagInfoDescription(di);
const char *severity_str = NULL;
switch (severity) {
case LLVMDSError:
severity_str = "error";
break;
case LLVMDSWarning:
severity_str = "warning";
@@ -191,39 +191,39 @@ static void radeonDiagnosticHandler(LLVMDiagnosticInfoRef
di, void *context)
}
LLVMDisposeMessage(description);
}
/**
* Compile an LLVM module to machine code.
*
* @returns 0 for success, 1 for failure
*/
-unsigned radeon_llvm_compile(LLVMModuleRef M, struct radeon_shader_binary
*binary,
- LLVMTargetMachineRef tm,
- struct pipe_debug_callback *debug)
+unsigned si_llvm_compile(LLVMModuleRef M, struct radeon_shader_binary *binary,
+ LLVMTargetMachineRef tm,
+ struct pipe_debug_callback *debug)
{
- struct radeon_llvm_diagnostics diag;
+ struct si_llvm_diagnostics diag;
char *err;
LLVMContextRef llvm_ctx;
LLVMMemoryBufferRef out_buffer;
unsigned buffer_size;
const char *buffer_data;
LLVMBool mem_err;
diag.debug = debug;
diag.retval = 0;
/* Setup Diagnostic Handler*/
llvm_ctx = LLVMGetModuleContext(M);
- LLVMContextSetDiagnosticHandler(llvm_ctx, radeonDiagnosticHandler,
&diag);
+ LLVMContextSetDiagnosticHandler(llvm_ctx, si_diagnostic_handler, &diag);
/* Compile IR*/
mem_err = LLVMTargetMachineEmitToMemoryBuffer(tm, M, LLVMObjectFile,
&err,
&out_buffer);
/* Process Errors/Warnings */
if (mem_err) {
fprintf(stderr, "%s: %s", __FUNCTION__, err);
pipe_debug_message(debug, SHADER_INFO,
"LLVM emit error: %s", err);
@@ -278,23 +278,23 @@ LLVMValueRef bitcast(struct lp_build_tgsi_context
*bld_base,
if (dst_type)
return LLVMBuildBitCast(builder, value, dst_type, "");
else
return value;
}
/**
* Return a value that is equal to the given i32 \p index if it lies in [0,num)
* or an undefined value in the same interval otherwise.
*/
-LLVMValueRef radeon_llvm_bound_index(struct si_shader_context *ctx,
- LLVMValueRef index,
- unsigned num)
+LLVMValueRef si_llvm_bound_index(struct si_shader_context *ctx,
+ LLVMValueRef index,
+ unsigned num)
{
struct gallivm_state *gallivm = &ctx->gallivm;
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef c_max = lp_build_const_int32(gallivm, num - 1);
LLVMValueRef cc;
if (util_is_power_of_two(num)) {
index = LLVMBuildAnd(builder, index, c_max, "");
} else {
/* In theory, this MAX pattern should result in code that is
@@ -303,42 +303,42 @@ LLVMValueRef radeon_llvm_bound_index(struct
si_shader_context *ctx,
* In practice, LLVM generates worse code (at the time of
* writing), because its value tracking is not strong enough.
*/
cc = LLVMBuildICmp(builder, LLVMIntULE, index, c_max, "");
index = LLVMBuildSelect(builder, cc, index, c_max, "");
}
return index;
}
-static struct radeon_llvm_flow *
+static struct si_llvm_flow *
get_current_flow(struct si_shader_context *ctx)
{
if (ctx->flow_depth > 0)
return &ctx->flow[ctx->flow_depth - 1];
return NULL;
}
-static struct radeon_llvm_flow *
+static struct si_llvm_flow *
get_innermost_loop(struct si_shader_context *ctx)
{
for (unsigned i = ctx->flow_depth; i > 0; --i) {
if (ctx->flow[i - 1].loop_entry_block)
return &ctx->flow[i - 1];
}
return NULL;
}
-static struct radeon_llvm_flow *
+static struct si_llvm_flow *
push_flow(struct si_shader_context *ctx)
{
- struct radeon_llvm_flow *flow;
+ struct si_llvm_flow *flow;
if (ctx->flow_depth >= ctx->flow_depth_max) {
unsigned new_max = MAX2(ctx->flow_depth << 1,
RADEON_LLVM_INITIAL_CF_DEPTH);
ctx->flow = REALLOC(ctx->flow,
ctx->flow_depth_max * sizeof(*ctx->flow),
new_max * sizeof(*ctx->flow));
ctx->flow_depth_max = new_max;
}
flow = &ctx->flow[ctx->flow_depth];
@@ -476,42 +476,42 @@ get_pointer_into_array(struct si_shader_context *ctx,
* descriptors).
*
* TODO It should be possible to avoid the additional instructions
* if LLVM is changed so that it guarantuees:
* 1. the scratch space descriptor isolates the current wave (this
* could even save the scratch offset SGPR at the cost of an
* additional SALU instruction)
* 2. the memory for allocas must be allocated at the _end_ of the
* scratch space (after spilled registers)
*/
- index = radeon_llvm_bound_index(ctx, index, array->range.Last -
array->range.First + 1);
+ index = si_llvm_bound_index(ctx, index, array->range.Last -
array->range.First + 1);
index = LLVMBuildMul(
builder, index,
lp_build_const_int32(gallivm, util_bitcount(array->writemask)),
"");
index = LLVMBuildAdd(
builder, index,
lp_build_const_int32(
gallivm,
util_bitcount(array->writemask & ((1 << swizzle) - 1))),
"");
idxs[0] = ctx->soa.bld_base.uint_bld.zero;
idxs[1] = index;
return LLVMBuildGEP(builder, alloca, idxs, 2, "");
}
LLVMValueRef
-radeon_llvm_emit_fetch_64bit(struct lp_build_tgsi_context *bld_base,
- enum tgsi_opcode_type type,
- LLVMValueRef ptr,
- LLVMValueRef ptr2)
+si_llvm_emit_fetch_64bit(struct lp_build_tgsi_context *bld_base,
+ enum tgsi_opcode_type type,
+ LLVMValueRef ptr,
+ LLVMValueRef ptr2)
{
LLVMBuilderRef builder = bld_base->base.gallivm->builder;
LLVMValueRef result;
result =
LLVMGetUndef(LLVMVectorType(LLVMIntTypeInContext(bld_base->base.gallivm->context,
32), bld_base->base.type.length * 2));
result = LLVMBuildInsertElement(builder,
result,
bitcast(bld_base, TGSI_TYPE_UNSIGNED,
ptr),
bld_base->int_bld.zero, "");
@@ -534,21 +534,21 @@ emit_array_fetch(struct lp_build_tgsi_context *bld_base,
unsigned i, size = range.Last - range.First + 1;
LLVMTypeRef vec = LLVMVectorType(tgsi2llvmtype(bld_base, type), size);
LLVMValueRef result = LLVMGetUndef(vec);
struct tgsi_full_src_register tmp_reg = {};
tmp_reg.Register.File = File;
for (i = 0; i < size; ++i) {
tmp_reg.Register.Index = i + range.First;
- LLVMValueRef temp = radeon_llvm_emit_fetch(bld_base, &tmp_reg,
type, swizzle);
+ LLVMValueRef temp = si_llvm_emit_fetch(bld_base, &tmp_reg,
type, swizzle);
result = LLVMBuildInsertElement(builder, result, temp,
lp_build_const_int32(gallivm, i), "array_vector");
}
return result;
}
static LLVMValueRef
load_value_from_array(struct lp_build_tgsi_context *bld_base,
unsigned file,
enum tgsi_opcode_type type,
@@ -562,21 +562,21 @@ load_value_from_array(struct lp_build_tgsi_context
*bld_base,
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef ptr;
ptr = get_pointer_into_array(ctx, file, swizzle, reg_index,
reg_indirect);
if (ptr) {
LLVMValueRef val = LLVMBuildLoad(builder, ptr, "");
if (tgsi_type_is_64bit(type)) {
LLVMValueRef ptr_hi, val_hi;
ptr_hi = LLVMBuildGEP(builder, ptr, &bld_base->uint_bld.one, 1,
"");
val_hi = LLVMBuildLoad(builder, ptr_hi, "");
- val = radeon_llvm_emit_fetch_64bit(bld_base, type, val,
val_hi);
+ val = si_llvm_emit_fetch_64bit(bld_base, type, val,
val_hi);
}
return val;
} else {
struct tgsi_declaration_range range =
get_array_range(bld_base, file, reg_index,
reg_indirect);
LLVMValueRef index =
emit_array_index(bld, reg_indirect, reg_index -
range.First);
LLVMValueRef array =
emit_array_fetch(bld_base, file, type, range, swizzle);
@@ -627,35 +627,35 @@ store_value_to_array(struct lp_build_tgsi_context
*bld_base,
default:
continue;
}
value = LLVMBuildExtractElement(builder, array,
lp_build_const_int32(gallivm, i), "");
LLVMBuildStore(builder, value, temp_ptr);
}
}
}
-LLVMValueRef radeon_llvm_emit_fetch(struct lp_build_tgsi_context *bld_base,
- const struct tgsi_full_src_register *reg,
- enum tgsi_opcode_type type,
- unsigned swizzle)
+LLVMValueRef si_llvm_emit_fetch(struct lp_build_tgsi_context *bld_base,
+ const struct tgsi_full_src_register *reg,
+ enum tgsi_opcode_type type,
+ unsigned swizzle)
{
struct si_shader_context *ctx = si_shader_context(bld_base);
struct lp_build_tgsi_soa_context *bld = lp_soa_context(bld_base);
LLVMBuilderRef builder = bld_base->base.gallivm->builder;
LLVMValueRef result = NULL, ptr, ptr2;
if (swizzle == ~0) {
LLVMValueRef values[TGSI_NUM_CHANNELS];
unsigned chan;
for (chan = 0; chan < TGSI_NUM_CHANNELS; chan++) {
- values[chan] = radeon_llvm_emit_fetch(bld_base, reg,
type, chan);
+ values[chan] = si_llvm_emit_fetch(bld_base, reg, type,
chan);
}
return lp_build_gather_values(bld_base->base.gallivm, values,
TGSI_NUM_CHANNELS);
}
if (reg->Register.Indirect) {
LLVMValueRef load = load_value_from_array(bld_base,
reg->Register.File, type,
swizzle, reg->Register.Index, ®->Indirect);
return bitcast(bld_base, type, load);
}
@@ -689,45 +689,45 @@ LLVMValueRef radeon_llvm_emit_fetch(struct
lp_build_tgsi_context *bld_base,
if (ctx->soa.bld_base.info->processor == PIPE_SHADER_FRAGMENT)
ctx->load_input(ctx, index, &ctx->input_decls[index],
input);
else
memcpy(input, &ctx->inputs[index * 4], sizeof(input));
result = input[swizzle];
if (tgsi_type_is_64bit(type)) {
ptr = result;
ptr2 = input[swizzle + 1];
- return radeon_llvm_emit_fetch_64bit(bld_base, type,
ptr, ptr2);
+ return si_llvm_emit_fetch_64bit(bld_base, type, ptr,
ptr2);
}
break;
}
case TGSI_FILE_TEMPORARY:
if (reg->Register.Index >= ctx->temps_count)
return LLVMGetUndef(tgsi2llvmtype(bld_base, type));
ptr = ctx->temps[reg->Register.Index * TGSI_NUM_CHANNELS +
swizzle];
if (tgsi_type_is_64bit(type)) {
ptr2 = ctx->temps[reg->Register.Index *
TGSI_NUM_CHANNELS + swizzle + 1];
- return radeon_llvm_emit_fetch_64bit(bld_base, type,
- LLVMBuildLoad(builder, ptr,
""),
- LLVMBuildLoad(builder, ptr2,
""));
+ return si_llvm_emit_fetch_64bit(bld_base, type,
+ LLVMBuildLoad(builder, ptr,
""),
+ LLVMBuildLoad(builder, ptr2,
""));
}
result = LLVMBuildLoad(builder, ptr, "");
break;
case TGSI_FILE_OUTPUT:
ptr = lp_get_output_ptr(bld, reg->Register.Index, swizzle);
if (tgsi_type_is_64bit(type)) {
ptr2 = lp_get_output_ptr(bld, reg->Register.Index,
swizzle + 1);
- return radeon_llvm_emit_fetch_64bit(bld_base, type,
- LLVMBuildLoad(builder, ptr,
""),
- LLVMBuildLoad(builder, ptr2,
""));
+ return si_llvm_emit_fetch_64bit(bld_base, type,
+ LLVMBuildLoad(builder, ptr,
""),
+ LLVMBuildLoad(builder, ptr2,
""));
}
result = LLVMBuildLoad(builder, ptr, "");
break;
default:
return LLVMGetUndef(tgsi2llvmtype(bld_base, type));
}
return bitcast(bld_base, type, result);
}
@@ -917,39 +917,39 @@ static void emit_declaration(struct lp_build_tgsi_context
*bld_base,
case TGSI_FILE_MEMORY:
ctx->declare_memory_region(ctx, decl);
break;
default:
break;
}
}
-LLVMValueRef radeon_llvm_saturate(struct lp_build_tgsi_context *bld_base,
- LLVMValueRef value)
+LLVMValueRef si_llvm_saturate(struct lp_build_tgsi_context *bld_base,
+ LLVMValueRef value)
{
struct lp_build_emit_data clamp_emit_data;
memset(&clamp_emit_data, 0, sizeof(clamp_emit_data));
clamp_emit_data.arg_count = 3;
clamp_emit_data.args[0] = value;
clamp_emit_data.args[2] = bld_base->base.one;
clamp_emit_data.args[1] = bld_base->base.zero;
return lp_build_emit_llvm(bld_base, TGSI_OPCODE_CLAMP,
&clamp_emit_data);
}
-void radeon_llvm_emit_store(struct lp_build_tgsi_context *bld_base,
- const struct tgsi_full_instruction *inst,
- const struct tgsi_opcode_info *info,
- LLVMValueRef dst[4])
+void si_llvm_emit_store(struct lp_build_tgsi_context *bld_base,
+ const struct tgsi_full_instruction *inst,
+ const struct tgsi_opcode_info *info,
+ LLVMValueRef dst[4])
{
struct si_shader_context *ctx = si_shader_context(bld_base);
struct lp_build_tgsi_soa_context *bld = lp_soa_context(bld_base);
struct gallivm_state *gallivm = bld->bld_base.base.gallivm;
const struct tgsi_full_dst_register *reg = &inst->Dst[0];
LLVMBuilderRef builder = bld->bld_base.base.gallivm->builder;
LLVMValueRef temp_ptr, temp_ptr2 = NULL;
unsigned chan, chan_index;
bool is_vec_store = false;
enum tgsi_opcode_type dtype =
tgsi_opcode_infer_dst_type(inst->Instruction.Opcode);
@@ -969,21 +969,21 @@ void radeon_llvm_emit_store(struct lp_build_tgsi_context
*bld_base,
bld_base->emit_store(bld_base, inst, info, values);
return;
}
TGSI_FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
LLVMValueRef value = dst[chan_index];
if (tgsi_type_is_64bit(dtype) && (chan_index == 1 || chan_index
== 3))
continue;
if (inst->Instruction.Saturate)
- value = radeon_llvm_saturate(bld_base, value);
+ value = si_llvm_saturate(bld_base, value);
if (reg->Register.File == TGSI_FILE_ADDRESS) {
temp_ptr = bld->addr[reg->Register.Index][chan_index];
LLVMBuildStore(builder, value, temp_ptr);
continue;
}
if (!tgsi_type_is_64bit(dtype))
value = bitcast(bld_base, TGSI_TYPE_FLOAT, value);
@@ -1046,21 +1046,21 @@ static void set_basicblock_name(LLVMBasicBlockRef bb,
const char *base, int pc)
/* Append a basic block at the level of the parent flow.
*/
static LLVMBasicBlockRef append_basic_block(struct si_shader_context *ctx,
const char *name)
{
struct gallivm_state *gallivm = &ctx->gallivm;
assert(ctx->flow_depth >= 1);
if (ctx->flow_depth >= 2) {
- struct radeon_llvm_flow *flow = &ctx->flow[ctx->flow_depth - 2];
+ struct si_llvm_flow *flow = &ctx->flow[ctx->flow_depth - 2];
return LLVMInsertBasicBlockInContext(gallivm->context,
flow->next_block, name);
}
return LLVMAppendBasicBlockInContext(gallivm->context, ctx->main_fn,
name);
}
/* Emit a branch to the given default target for the current block if
* applicable -- that is, if the current block does not already contain a
@@ -1071,112 +1071,112 @@ static void emit_default_branch(LLVMBuilderRef
builder, LLVMBasicBlockRef target
if (!LLVMGetBasicBlockTerminator(LLVMGetInsertBlock(builder)))
LLVMBuildBr(builder, target);
}
static void bgnloop_emit(const struct lp_build_tgsi_action *action,
struct lp_build_tgsi_context *bld_base,
struct lp_build_emit_data *emit_data)
{
struct si_shader_context *ctx = si_shader_context(bld_base);
struct gallivm_state *gallivm = bld_base->base.gallivm;
- struct radeon_llvm_flow *flow = push_flow(ctx);
+ struct si_llvm_flow *flow = push_flow(ctx);
flow->loop_entry_block = append_basic_block(ctx, "LOOP");
flow->next_block = append_basic_block(ctx, "ENDLOOP");
set_basicblock_name(flow->loop_entry_block, "loop", bld_base->pc);
LLVMBuildBr(gallivm->builder, flow->loop_entry_block);
LLVMPositionBuilderAtEnd(gallivm->builder, flow->loop_entry_block);
}
static void brk_emit(const struct lp_build_tgsi_action *action,
struct lp_build_tgsi_context *bld_base,
struct lp_build_emit_data *emit_data)
{
struct si_shader_context *ctx = si_shader_context(bld_base);
struct gallivm_state *gallivm = bld_base->base.gallivm;
- struct radeon_llvm_flow *flow = get_innermost_loop(ctx);
+ struct si_llvm_flow *flow = get_innermost_loop(ctx);
LLVMBuildBr(gallivm->builder, flow->next_block);
}
static void cont_emit(const struct lp_build_tgsi_action *action,
struct lp_build_tgsi_context *bld_base,
struct lp_build_emit_data *emit_data)
{
struct si_shader_context *ctx = si_shader_context(bld_base);
struct gallivm_state *gallivm = bld_base->base.gallivm;
- struct radeon_llvm_flow *flow = get_innermost_loop(ctx);
+ struct si_llvm_flow *flow = get_innermost_loop(ctx);
LLVMBuildBr(gallivm->builder, flow->loop_entry_block);
}
static void else_emit(const struct lp_build_tgsi_action *action,
struct lp_build_tgsi_context *bld_base,
struct lp_build_emit_data *emit_data)
{
struct si_shader_context *ctx = si_shader_context(bld_base);
struct gallivm_state *gallivm = bld_base->base.gallivm;
- struct radeon_llvm_flow *current_branch = get_current_flow(ctx);
+ struct si_llvm_flow *current_branch = get_current_flow(ctx);
LLVMBasicBlockRef endif_block;
assert(!current_branch->loop_entry_block);
endif_block = append_basic_block(ctx, "ENDIF");
emit_default_branch(gallivm->builder, endif_block);
LLVMPositionBuilderAtEnd(gallivm->builder, current_branch->next_block);
set_basicblock_name(current_branch->next_block, "else", bld_base->pc);
current_branch->next_block = endif_block;
}
static void endif_emit(const struct lp_build_tgsi_action *action,
struct lp_build_tgsi_context *bld_base,
struct lp_build_emit_data *emit_data)
{
struct si_shader_context *ctx = si_shader_context(bld_base);
struct gallivm_state *gallivm = bld_base->base.gallivm;
- struct radeon_llvm_flow *current_branch = get_current_flow(ctx);
+ struct si_llvm_flow *current_branch = get_current_flow(ctx);
assert(!current_branch->loop_entry_block);
emit_default_branch(gallivm->builder, current_branch->next_block);
LLVMPositionBuilderAtEnd(gallivm->builder, current_branch->next_block);
set_basicblock_name(current_branch->next_block, "endif", bld_base->pc);
ctx->flow_depth--;
}
static void endloop_emit(const struct lp_build_tgsi_action *action,
struct lp_build_tgsi_context *bld_base,
struct lp_build_emit_data *emit_data)
{
struct si_shader_context *ctx = si_shader_context(bld_base);
struct gallivm_state *gallivm = bld_base->base.gallivm;
- struct radeon_llvm_flow *current_loop = get_current_flow(ctx);
+ struct si_llvm_flow *current_loop = get_current_flow(ctx);
assert(current_loop->loop_entry_block);
emit_default_branch(gallivm->builder, current_loop->loop_entry_block);
LLVMPositionBuilderAtEnd(gallivm->builder, current_loop->next_block);
set_basicblock_name(current_loop->next_block, "endloop", bld_base->pc);
ctx->flow_depth--;
}
static void if_cond_emit(const struct lp_build_tgsi_action *action,
struct lp_build_tgsi_context *bld_base,
struct lp_build_emit_data *emit_data,
LLVMValueRef cond)
{
struct si_shader_context *ctx = si_shader_context(bld_base);
struct gallivm_state *gallivm = bld_base->base.gallivm;
- struct radeon_llvm_flow *flow = push_flow(ctx);
+ struct si_llvm_flow *flow = push_flow(ctx);
LLVMBasicBlockRef if_block;
if_block = append_basic_block(ctx, "IF");
flow->next_block = append_basic_block(ctx, "ELSE");
set_basicblock_name(if_block, "if", bld_base->pc);
LLVMBuildCondBr(gallivm->builder, cond, if_block, flow->next_block);
LLVMPositionBuilderAtEnd(gallivm->builder, if_block);
}
static void if_emit(const struct lp_build_tgsi_action *action,
@@ -1214,23 +1214,23 @@ static void emit_immediate(struct lp_build_tgsi_context
*bld_base,
struct si_shader_context *ctx = si_shader_context(bld_base);
for (i = 0; i < 4; ++i) {
ctx->soa.immediates[ctx->soa.num_immediates][i] =
LLVMConstInt(bld_base->uint_bld.elem_type,
imm->u[i].Uint, false );
}
ctx->soa.num_immediates++;
}
-void radeon_llvm_context_init(struct si_shader_context *ctx, const char
*triple,
- const struct tgsi_shader_info *info,
- const struct tgsi_token *tokens)
+void si_llvm_context_init(struct si_shader_context *ctx, const char *triple,
+ const struct tgsi_shader_info *info,
+ const struct tgsi_token *tokens)
{
struct lp_type type;
/* Initialize the gallivm object:
* We are only using the module, context, and builder fields of this
struct.
* This should be enough for us to be able to pass our gallivm struct
to the
* helper functions in the gallivm module.
*/
memset(&ctx->gallivm, 0, sizeof (ctx->gallivm));
memset(&ctx->soa, 0, sizeof(ctx->soa));
@@ -1264,29 +1264,29 @@ void radeon_llvm_context_init(struct si_shader_context
*ctx, const char *triple,
lp_build_context_init(&bld_base->base, &ctx->gallivm, type);
lp_build_context_init(&ctx->soa.bld_base.uint_bld, &ctx->gallivm,
lp_uint_type(type));
lp_build_context_init(&ctx->soa.bld_base.int_bld, &ctx->gallivm,
lp_int_type(type));
type.width *= 2;
lp_build_context_init(&ctx->soa.bld_base.dbl_bld, &ctx->gallivm, type);
lp_build_context_init(&ctx->soa.bld_base.uint64_bld, &ctx->gallivm,
lp_uint_type(type));
lp_build_context_init(&ctx->soa.bld_base.int64_bld, &ctx->gallivm,
lp_int_type(type));
bld_base->soa = 1;
- bld_base->emit_store = radeon_llvm_emit_store;
+ bld_base->emit_store = si_llvm_emit_store;
bld_base->emit_swizzle = emit_swizzle;
bld_base->emit_declaration = emit_declaration;
bld_base->emit_immediate = emit_immediate;
- bld_base->emit_fetch_funcs[TGSI_FILE_IMMEDIATE] =
radeon_llvm_emit_fetch;
- bld_base->emit_fetch_funcs[TGSI_FILE_INPUT] = radeon_llvm_emit_fetch;
- bld_base->emit_fetch_funcs[TGSI_FILE_TEMPORARY] =
radeon_llvm_emit_fetch;
- bld_base->emit_fetch_funcs[TGSI_FILE_OUTPUT] = radeon_llvm_emit_fetch;
+ bld_base->emit_fetch_funcs[TGSI_FILE_IMMEDIATE] = si_llvm_emit_fetch;
+ bld_base->emit_fetch_funcs[TGSI_FILE_INPUT] = si_llvm_emit_fetch;
+ bld_base->emit_fetch_funcs[TGSI_FILE_TEMPORARY] = si_llvm_emit_fetch;
+ bld_base->emit_fetch_funcs[TGSI_FILE_OUTPUT] = si_llvm_emit_fetch;
bld_base->emit_fetch_funcs[TGSI_FILE_SYSTEM_VALUE] = fetch_system_value;
/* metadata allowing 2.5 ULP */
ctx->fpmath_md_kind = LLVMGetMDKindIDInContext(ctx->gallivm.context,
"fpmath", 6);
LLVMValueRef arg = lp_build_const_float(&ctx->gallivm, 2.5);
ctx->fpmath_md_2p5_ulp = LLVMMDNodeInContext(ctx->gallivm.context,
&arg, 1);
/* Allocate outputs */
@@ -1295,45 +1295,45 @@ void radeon_llvm_context_init(struct si_shader_context
*ctx, const char *triple,
bld_base->op_actions[TGSI_OPCODE_BGNLOOP].emit = bgnloop_emit;
bld_base->op_actions[TGSI_OPCODE_BRK].emit = brk_emit;
bld_base->op_actions[TGSI_OPCODE_CONT].emit = cont_emit;
bld_base->op_actions[TGSI_OPCODE_IF].emit = if_emit;
bld_base->op_actions[TGSI_OPCODE_UIF].emit = uif_emit;
bld_base->op_actions[TGSI_OPCODE_ELSE].emit = else_emit;
bld_base->op_actions[TGSI_OPCODE_ENDIF].emit = endif_emit;
bld_base->op_actions[TGSI_OPCODE_ENDLOOP].emit = endloop_emit;
}
-void radeon_llvm_create_func(struct si_shader_context *ctx,
- LLVMTypeRef *return_types, unsigned
num_return_elems,
- LLVMTypeRef *ParamTypes, unsigned ParamCount)
+void si_llvm_create_func(struct si_shader_context *ctx,
+ LLVMTypeRef *return_types, unsigned num_return_elems,
+ LLVMTypeRef *ParamTypes, unsigned ParamCount)
{
LLVMTypeRef main_fn_type, ret_type;
LLVMBasicBlockRef main_fn_body;
if (num_return_elems)
ret_type = LLVMStructTypeInContext(ctx->gallivm.context,
return_types,
num_return_elems, true);
else
ret_type = LLVMVoidTypeInContext(ctx->gallivm.context);
/* Setup the function */
ctx->return_type = ret_type;
main_fn_type = LLVMFunctionType(ret_type, ParamTypes, ParamCount, 0);
ctx->main_fn = LLVMAddFunction(ctx->gallivm.module, "main",
main_fn_type);
main_fn_body = LLVMAppendBasicBlockInContext(ctx->gallivm.context,
ctx->main_fn, "main_body");
LLVMPositionBuilderAtEnd(ctx->gallivm.builder, main_fn_body);
}
-void radeon_llvm_finalize_module(struct si_shader_context *ctx,
- bool run_verifier)
+void si_llvm_finalize_module(struct si_shader_context *ctx,
+ bool run_verifier)
{
struct gallivm_state *gallivm = ctx->soa.bld_base.base.gallivm;
const char *triple = LLVMGetTarget(gallivm->module);
LLVMTargetLibraryInfoRef target_library_info;
/* Create the pass manager */
gallivm->passmgr = LLVMCreateFunctionPassManagerForModule(
gallivm->module);
target_library_info = gallivm_create_target_library_info(triple);
@@ -1355,21 +1355,21 @@ void radeon_llvm_finalize_module(struct
si_shader_context *ctx,
/* Run the pass */
LLVMInitializeFunctionPassManager(gallivm->passmgr);
LLVMRunFunctionPassManager(gallivm->passmgr, ctx->main_fn);
LLVMFinalizeFunctionPassManager(gallivm->passmgr);
LLVMDisposeBuilder(gallivm->builder);
LLVMDisposePassManager(gallivm->passmgr);
gallivm_dispose_target_library_info(target_library_info);
}
-void radeon_llvm_dispose(struct si_shader_context *ctx)
+void si_llvm_dispose(struct si_shader_context *ctx)
{
LLVMDisposeModule(ctx->soa.bld_base.base.gallivm->module);
LLVMContextDispose(ctx->soa.bld_base.base.gallivm->context);
FREE(ctx->temp_arrays);
ctx->temp_arrays = NULL;
FREE(ctx->temp_array_allocas);
ctx->temp_array_allocas = NULL;
FREE(ctx->temps);
ctx->temps = NULL;
ctx->temps_count = 0;
