[Mesa-dev] [PATCH 11/11] radeonsi: use f32_0 and f32_1
Marek Olšák
maraeo at gmail.com
Fri Sep 29 14:49:55 UTC 2017
From: Marek Olšák <marek.olsak at amd.com>
---
src/gallium/drivers/radeonsi/si_shader.c | 28 ++++++++++------------
src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c | 18 +++++++-------
src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c | 4 ++--
.../drivers/radeonsi/si_shader_tgsi_setup.c | 2 +-
4 files changed, 26 insertions(+), 26 deletions(-)
diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index db9a0d7..f3f5b23 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -1664,28 +1664,28 @@ void si_load_system_value(struct si_shader_context *ctx,
* doesn't support smoothing.
*/
value = LLVMGetParam(ctx->main_fn, SI_PARAM_SAMPLE_COVERAGE);
break;
case TGSI_SEMANTIC_TESSCOORD:
{
LLVMValueRef coord[4] = {
LLVMGetParam(ctx->main_fn, ctx->param_tes_u),
LLVMGetParam(ctx->main_fn, ctx->param_tes_v),
- bld->zero,
- bld->zero
+ ctx->ac.f32_0,
+ ctx->ac.f32_0
};
/* For triangles, the vector should be (u, v, 1-u-v). */
if (ctx->shader->selector->info.properties[TGSI_PROPERTY_TES_PRIM_MODE] ==
PIPE_PRIM_TRIANGLES)
- coord[2] = lp_build_sub(bld, bld->one,
+ coord[2] = lp_build_sub(bld, ctx->ac.f32_1,
lp_build_add(bld, coord[0], coord[1]));
value = lp_build_gather_values(&ctx->gallivm, coord, 4);
break;
}
case TGSI_SEMANTIC_VERTICESIN:
if (ctx->type == PIPE_SHADER_TESS_CTRL)
value = unpack_param(ctx, ctx->param_tcs_out_lds_layout, 26, 6);
else if (ctx->type == PIPE_SHADER_TESS_EVAL)
@@ -2083,21 +2083,21 @@ static void si_llvm_init_export_args(struct lp_build_tgsi_context *bld_base,
val[chan] = lp_build_emit_llvm_binary(bld_base, TGSI_OPCODE_MAX,
val[chan],
LLVMConstReal(ctx->f32, -1));
/* Convert to a signed integer in [-32767, 32767]. */
val[chan] = LLVMBuildFMul(builder, val[chan],
LLVMConstReal(ctx->f32, 32767), "");
/* If positive, add 0.5, else add -0.5. */
val[chan] = LLVMBuildFAdd(builder, val[chan],
LLVMBuildSelect(builder,
LLVMBuildFCmp(builder, LLVMRealOGE,
- val[chan], base->zero, ""),
+ val[chan], ctx->ac.f32_0, ""),
LLVMConstReal(ctx->f32, 0.5),
LLVMConstReal(ctx->f32, -0.5), ""), "");
val[chan] = LLVMBuildFPToSI(builder, val[chan], ctx->i32, "");
}
args->compr = 1; /* COMPR flag */
args->out[0] = ac_to_float(&ctx->ac, si_llvm_pack_two_int32_as_int16(ctx, val));
args->out[1] = ac_to_float(&ctx->ac, si_llvm_pack_two_int32_as_int16(ctx, val+2));
break;
@@ -2459,21 +2459,20 @@ static void si_build_param_exports(struct si_shader_context *ctx,
shader->info.nr_param_exports = param_count;
}
/* Generate export instructions for hardware VS shader stage */
static void si_llvm_export_vs(struct lp_build_tgsi_context *bld_base,
struct si_shader_output_values *outputs,
unsigned noutput)
{
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 ac_export_args pos_args[4] = {};
LLVMValueRef psize_value = NULL, edgeflag_value = NULL, layer_value = NULL, viewport_index_value = NULL;
unsigned pos_idx;
int i;
/* Build position exports. */
for (i = 0; i < noutput; i++) {
switch (outputs[i].semantic_name) {
case TGSI_SEMANTIC_POSITION:
si_llvm_init_export_args(bld_base, outputs[i].values,
@@ -2508,43 +2507,43 @@ static void si_llvm_export_vs(struct lp_build_tgsi_context *bld_base,
}
}
/* We need to add the position output manually if it's missing. */
if (!pos_args[0].out[0]) {
pos_args[0].enabled_channels = 0xf; /* writemask */
pos_args[0].valid_mask = 0; /* EXEC mask */
pos_args[0].done = 0; /* last export? */
pos_args[0].target = V_008DFC_SQ_EXP_POS;
pos_args[0].compr = 0; /* COMPR flag */
- pos_args[0].out[0] = base->zero; /* X */
- pos_args[0].out[1] = base->zero; /* Y */
- pos_args[0].out[2] = base->zero; /* Z */
- pos_args[0].out[3] = base->one; /* W */
+ pos_args[0].out[0] = ctx->ac.f32_0; /* X */
+ pos_args[0].out[1] = ctx->ac.f32_0; /* Y */
+ pos_args[0].out[2] = ctx->ac.f32_0; /* Z */
+ pos_args[0].out[3] = ctx->ac.f32_1; /* W */
}
/* Write the misc vector (point size, edgeflag, layer, viewport). */
if (shader->selector->info.writes_psize ||
shader->selector->info.writes_edgeflag ||
shader->selector->info.writes_viewport_index ||
shader->selector->info.writes_layer) {
pos_args[1].enabled_channels = shader->selector->info.writes_psize |
(shader->selector->info.writes_edgeflag << 1) |
(shader->selector->info.writes_layer << 2);
pos_args[1].valid_mask = 0; /* EXEC mask */
pos_args[1].done = 0; /* last export? */
pos_args[1].target = V_008DFC_SQ_EXP_POS + 1;
pos_args[1].compr = 0; /* COMPR flag */
- pos_args[1].out[0] = base->zero; /* X */
- pos_args[1].out[1] = base->zero; /* Y */
- pos_args[1].out[2] = base->zero; /* Z */
- pos_args[1].out[3] = base->zero; /* W */
+ pos_args[1].out[0] = ctx->ac.f32_0; /* X */
+ pos_args[1].out[1] = ctx->ac.f32_0; /* Y */
+ pos_args[1].out[2] = ctx->ac.f32_0; /* Z */
+ pos_args[1].out[3] = ctx->ac.f32_0; /* W */
if (shader->selector->info.writes_psize)
pos_args[1].out[0] = psize_value;
if (shader->selector->info.writes_edgeflag) {
/* The output is a float, but the hw expects an integer
* with the first bit containing the edge flag. */
edgeflag_value = LLVMBuildFPToUI(ctx->ac.builder,
edgeflag_value,
ctx->i32, "");
@@ -3338,31 +3337,30 @@ static void si_export_mrt_z(struct lp_build_tgsi_context *bld_base,
memcpy(&exp->args[exp->num++], &args, sizeof(args));
}
static void si_export_mrt_color(struct lp_build_tgsi_context *bld_base,
LLVMValueRef *color, unsigned index,
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.part.ps.epilog.clamp_color)
for (i = 0; i < 4; i++)
color[i] = ac_build_clamp(&ctx->ac, color[i]);
/* Alpha to one */
if (ctx->shader->key.part.ps.epilog.alpha_to_one)
- color[3] = base->one;
+ color[3] = ctx->ac.f32_1;
/* Alpha test */
if (index == 0 &&
ctx->shader->key.part.ps.epilog.alpha_func != PIPE_FUNC_ALWAYS)
si_alpha_test(bld_base, color[3]);
/* Line & polygon smoothing */
if (ctx->shader->key.part.ps.epilog.poly_line_smoothing)
color[3] = si_scale_alpha_by_sample_mask(bld_base, color[3],
samplemask_param);
diff --git a/src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c b/src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c
index 563b305..fc705c3 100644
--- a/src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c
+++ b/src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c
@@ -33,33 +33,33 @@ static void kill_if_fetch_args(struct lp_build_tgsi_context *bld_base,
{
const struct tgsi_full_instruction *inst = emit_data->inst;
struct si_shader_context *ctx = si_shader_context(bld_base);
LLVMBuilderRef builder = ctx->ac.builder;
unsigned i;
LLVMValueRef conds[TGSI_NUM_CHANNELS];
for (i = 0; i < TGSI_NUM_CHANNELS; i++) {
LLVMValueRef value = lp_build_emit_fetch(bld_base, inst, 0, i);
conds[i] = LLVMBuildFCmp(builder, LLVMRealOLT, value,
- bld_base->base.zero, "");
+ ctx->ac.f32_0, "");
}
/* Or the conditions together */
for (i = TGSI_NUM_CHANNELS - 1; i > 0; i--) {
conds[i - 1] = LLVMBuildOr(builder, conds[i], conds[i - 1], "");
}
emit_data->dst_type = ctx->voidt;
emit_data->arg_count = 1;
emit_data->args[0] = LLVMBuildSelect(builder, conds[0],
LLVMConstReal(ctx->f32, -1.0f),
- bld_base->base.zero, "");
+ ctx->ac.f32_0, "");
}
static void kil_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);
LLVMBuilderRef builder = ctx->ac.builder;
if (ctx->postponed_kill) {
@@ -137,21 +137,21 @@ static void emit_ucmp(const struct lp_build_tgsi_action *action,
}
static void emit_cmp(const struct lp_build_tgsi_action *action,
struct lp_build_tgsi_context *bld_base,
struct lp_build_emit_data *emit_data)
{
struct si_shader_context *ctx = si_shader_context(bld_base);
LLVMValueRef cond, *args = emit_data->args;
cond = LLVMBuildFCmp(ctx->ac.builder, LLVMRealOLT, args[0],
- bld_base->base.zero, "");
+ ctx->ac.f32_0, "");
emit_data->output[emit_data->chan] =
LLVMBuildSelect(ctx->ac.builder, cond, args[1], args[2], "");
}
static void emit_set_cond(const struct lp_build_tgsi_action *action,
struct lp_build_tgsi_context *bld_base,
struct lp_build_emit_data *emit_data)
{
struct si_shader_context *ctx = si_shader_context(bld_base);
@@ -168,21 +168,21 @@ static void emit_set_cond(const struct lp_build_tgsi_action *action,
case TGSI_OPCODE_SLT: pred = LLVMRealOLT; break;
case TGSI_OPCODE_SNE: pred = LLVMRealUNE; break;
case TGSI_OPCODE_SGT: pred = LLVMRealOGT; break;
default: assert(!"unknown instruction"); pred = 0; break;
}
cond = LLVMBuildFCmp(ctx->ac.builder,
pred, emit_data->args[0], emit_data->args[1], "");
emit_data->output[emit_data->chan] = LLVMBuildSelect(ctx->ac.builder,
- cond, bld_base->base.one, bld_base->base.zero, "");
+ cond, ctx->ac.f32_1, ctx->ac.f32_0, "");
}
static void emit_fcmp(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);
LLVMRealPredicate pred;
/* Use ordered for everything but NE (which is usual for
@@ -365,23 +365,23 @@ static void emit_ssg(const struct lp_build_tgsi_action *action,
cmp = LLVMBuildICmp(builder, LLVMIntSGT, emit_data->args[0], ctx->i32_0, "");
val = LLVMBuildSelect(builder, cmp, ctx->i32_1, emit_data->args[0], "");
cmp = LLVMBuildICmp(builder, LLVMIntSGE, val, ctx->i32_0, "");
val = LLVMBuildSelect(builder, cmp, val, LLVMConstInt(ctx->i32, -1, true), "");
} else if (emit_data->inst->Instruction.Opcode == TGSI_OPCODE_DSSG) {
cmp = LLVMBuildFCmp(builder, LLVMRealOGT, emit_data->args[0], bld_base->dbl_bld.zero, "");
val = LLVMBuildSelect(builder, cmp, bld_base->dbl_bld.one, emit_data->args[0], "");
cmp = LLVMBuildFCmp(builder, LLVMRealOGE, val, bld_base->dbl_bld.zero, "");
val = LLVMBuildSelect(builder, cmp, val, LLVMConstReal(bld_base->dbl_bld.elem_type, -1), "");
} else { // float SSG
- cmp = LLVMBuildFCmp(builder, LLVMRealOGT, emit_data->args[0], bld_base->base.zero, "");
- val = LLVMBuildSelect(builder, cmp, bld_base->base.one, emit_data->args[0], "");
- cmp = LLVMBuildFCmp(builder, LLVMRealOGE, val, bld_base->base.zero, "");
+ cmp = LLVMBuildFCmp(builder, LLVMRealOGT, emit_data->args[0], ctx->ac.f32_0, "");
+ val = LLVMBuildSelect(builder, cmp, ctx->ac.f32_1, emit_data->args[0], "");
+ cmp = LLVMBuildFCmp(builder, LLVMRealOGE, val, ctx->ac.f32_0, "");
val = LLVMBuildSelect(builder, cmp, val, LLVMConstReal(ctx->f32, -1), "");
}
emit_data->output[emit_data->chan] = val;
}
static void emit_ineg(const struct lp_build_tgsi_action *action,
struct lp_build_tgsi_context *bld_base,
struct lp_build_emit_data *emit_data)
{
@@ -703,27 +703,29 @@ static void emit_fdiv(const struct lp_build_tgsi_action *action,
LLVMSetMetadata(emit_data->output[emit_data->chan],
ctx->fpmath_md_kind, ctx->fpmath_md_2p5_ulp);
}
/* 1/sqrt is translated to rsq for f32 if fp32 denormals are not enabled in
* the target machine. f64 needs global unsafe math flags to get rsq. */
static void emit_rsq(const struct lp_build_tgsi_action *action,
struct lp_build_tgsi_context *bld_base,
struct lp_build_emit_data *emit_data)
{
+ struct si_shader_context *ctx = si_shader_context(bld_base);
+
LLVMValueRef sqrt =
lp_build_emit_llvm_unary(bld_base, TGSI_OPCODE_SQRT,
emit_data->args[0]);
emit_data->output[emit_data->chan] =
lp_build_emit_llvm_binary(bld_base, TGSI_OPCODE_DIV,
- bld_base->base.one, sqrt);
+ ctx->ac.f32_1, sqrt);
}
static void dfracexp_fetch_args(struct lp_build_tgsi_context *bld_base,
struct lp_build_emit_data *emit_data)
{
emit_data->args[0] = lp_build_emit_fetch(bld_base, emit_data->inst, 0, TGSI_CHAN_X);
emit_data->arg_count = 1;
}
static void dfracexp_emit(const struct lp_build_tgsi_action *action,
diff --git a/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c b/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c
index 7c2afe3..b472b19 100644
--- a/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c
+++ b/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c
@@ -1322,21 +1322,21 @@ static void tex_fetch_args(
emit_data->inst, 0,
chan);
if (opcode == TGSI_OPCODE_TXP)
coords[chan] = lp_build_emit_llvm_binary(bld_base,
TGSI_OPCODE_DIV,
coords[chan],
coords[3]);
}
if (opcode == TGSI_OPCODE_TXP)
- coords[3] = bld_base->base.one;
+ coords[3] = ctx->ac.f32_1;
/* Pack offsets. */
if (has_offset &&
opcode != TGSI_OPCODE_TXF &&
opcode != TGSI_OPCODE_TXF_LZ) {
/* The offsets are six-bit signed integers packed like this:
* X=[5:0], Y=[13:8], and Z=[21:16].
*/
LLVMValueRef offset[3], pack;
@@ -1450,21 +1450,21 @@ static void tex_fetch_args(
for (param = 0; param < 2; param++) {
for (chan = 0; chan < num_src_deriv_channels; chan++)
derivs[param * num_dst_deriv_channels + chan] =
lp_build_emit_fetch(bld_base, inst, param+1, chan);
/* Fill in the rest with zeros. */
for (chan = num_src_deriv_channels;
chan < num_dst_deriv_channels; chan++)
derivs[param * num_dst_deriv_channels + chan] =
- bld_base->base.zero;
+ ctx->ac.f32_0;
}
}
if (target == TGSI_TEXTURE_CUBE ||
target == TGSI_TEXTURE_CUBE_ARRAY ||
target == TGSI_TEXTURE_SHADOWCUBE ||
target == TGSI_TEXTURE_SHADOWCUBE_ARRAY) {
ac_prepare_cube_coords(&ctx->ac,
opcode == TGSI_OPCODE_TXD,
target == TGSI_TEXTURE_CUBE_ARRAY ||
diff --git a/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c b/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c
index 5c397f7..1d4696f 100644
--- a/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c
+++ b/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c
@@ -1091,21 +1091,21 @@ static void if_cond_emit(const struct lp_build_tgsi_action *action,
static void if_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);
LLVMValueRef cond;
cond = LLVMBuildFCmp(ctx->ac.builder, LLVMRealUNE,
emit_data->args[0],
- bld_base->base.zero, "");
+ ctx->ac.f32_0, "");
if_cond_emit(action, bld_base, emit_data, cond);
}
static void uif_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);
LLVMValueRef cond;
--
2.7.4
More information about the mesa-dev
mailing list