From: Nicolai Hähnle <[email protected]>
---
src/gallium/drivers/radeonsi/si_shader.c | 50 +++++++++++++++-----------------
1 file changed, 23 insertions(+), 27 deletions(-)
diff --git a/src/gallium/drivers/radeonsi/si_shader.c
b/src/gallium/drivers/radeonsi/si_shader.c
index e6b14f92056..61adb6b1dcc 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -2074,27 +2074,26 @@ static LLVMValueRef
si_llvm_pack_two_int32_as_int16(struct si_shader_context *ct
{
LLVMValueRef v[2] = {
LLVMBuildAnd(ctx->ac.builder, val[0],
LLVMConstInt(ctx->i32, 0xffff, 0), ""),
val[1],
};
return si_llvm_pack_two_int16(ctx, v);
}
/* Initialize arguments for the shader export intrinsic */
-static void si_llvm_init_export_args(struct lp_build_tgsi_context *bld_base,
+static void si_llvm_init_export_args(struct si_shader_context *ctx,
LLVMValueRef *values,
unsigned target,
struct ac_export_args *args)
{
- struct si_shader_context *ctx = si_shader_context(bld_base);
- struct lp_build_context *base = &bld_base->base;
+ LLVMValueRef f32undef = LLVMGetUndef(ctx->ac.f32);
LLVMBuilderRef builder = ctx->ac.builder;
LLVMValueRef val[4];
unsigned spi_shader_col_format = V_028714_SPI_SHADER_32_ABGR;
unsigned chan;
bool is_int8, is_int10;
/* Default is 0xf. Adjusted below depending on the format. */
args->enabled_channels = 0xf; /* writemask */
/* Specify whether the EXEC mask represents the valid mask */
@@ -2111,24 +2110,24 @@ static void si_llvm_init_export_args(struct
lp_build_tgsi_context *bld_base,
unsigned col_formats =
key->part.ps.epilog.spi_shader_col_format;
int cbuf = target - V_008DFC_SQ_EXP_MRT;
assert(cbuf >= 0 && cbuf < 8);
spi_shader_col_format = (col_formats >> (cbuf * 4)) & 0xf;
is_int8 = (key->part.ps.epilog.color_is_int8 >> cbuf) & 0x1;
is_int10 = (key->part.ps.epilog.color_is_int10 >> cbuf) & 0x1;
}
args->compr = false;
- args->out[0] = base->undef;
- args->out[1] = base->undef;
- args->out[2] = base->undef;
- args->out[3] = base->undef;
+ args->out[0] = f32undef;
+ args->out[1] = f32undef;
+ args->out[2] = f32undef;
+ args->out[3] = f32undef;
switch (spi_shader_col_format) {
case V_028714_SPI_SHADER_ZERO:
args->enabled_channels = 0; /* writemask */
args->target = V_008DFC_SQ_EXP_NULL;
break;
case V_028714_SPI_SHADER_32_R:
args->enabled_channels = 1; /* writemask */
args->out[0] = values[0];
@@ -2173,24 +2172,24 @@ static void si_llvm_init_export_args(struct
lp_build_tgsi_context *bld_base,
}
args->compr = 1; /* COMPR flag */
args->out[0] = ac_to_float(&ctx->ac,
si_llvm_pack_two_int16(ctx, val));
args->out[1] = ac_to_float(&ctx->ac,
si_llvm_pack_two_int16(ctx, val+2));
break;
case V_028714_SPI_SHADER_SNORM16_ABGR:
for (chan = 0; chan < 4; chan++) {
/* Clamp between [-1, 1]. */
- val[chan] = lp_build_emit_llvm_binary(bld_base,
TGSI_OPCODE_MIN,
+ val[chan] = lp_build_emit_llvm_binary(&ctx->bld_base,
TGSI_OPCODE_MIN,
values[chan],
LLVMConstReal(ctx->f32, 1));
- val[chan] = lp_build_emit_llvm_binary(bld_base,
TGSI_OPCODE_MAX,
+ val[chan] = lp_build_emit_llvm_binary(&ctx->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],
ctx->ac.f32_0, ""),
@@ -2206,21 +2205,21 @@ static void si_llvm_init_export_args(struct
lp_build_tgsi_context *bld_base,
case V_028714_SPI_SHADER_UINT16_ABGR: {
LLVMValueRef max_rgb = LLVMConstInt(ctx->i32,
is_int8 ? 255 : is_int10 ? 1023 : 65535, 0);
LLVMValueRef max_alpha =
!is_int10 ? max_rgb : LLVMConstInt(ctx->i32, 3, 0);
/* Clamp. */
for (chan = 0; chan < 4; chan++) {
val[chan] = ac_to_integer(&ctx->ac, values[chan]);
- val[chan] = lp_build_emit_llvm_binary(bld_base,
TGSI_OPCODE_UMIN,
+ val[chan] = lp_build_emit_llvm_binary(&ctx->bld_base,
TGSI_OPCODE_UMIN,
val[chan],
chan == 3 ? max_alpha : max_rgb);
}
args->compr = 1; /* COMPR flag */
args->out[0] = ac_to_float(&ctx->ac,
si_llvm_pack_two_int16(ctx, val));
args->out[1] = ac_to_float(&ctx->ac,
si_llvm_pack_two_int16(ctx, val+2));
break;
}
@@ -2230,24 +2229,24 @@ static void si_llvm_init_export_args(struct
lp_build_tgsi_context *bld_base,
LLVMValueRef min_rgb = LLVMConstInt(ctx->i32,
is_int8 ? -128 : is_int10 ? -512 : -32768, 0);
LLVMValueRef max_alpha =
!is_int10 ? max_rgb : ctx->i32_1;
LLVMValueRef min_alpha =
!is_int10 ? min_rgb : LLVMConstInt(ctx->i32, -2, 0);
/* Clamp. */
for (chan = 0; chan < 4; chan++) {
val[chan] = ac_to_integer(&ctx->ac, values[chan]);
- val[chan] = lp_build_emit_llvm_binary(bld_base,
+ val[chan] = lp_build_emit_llvm_binary(&ctx->bld_base,
TGSI_OPCODE_IMIN,
val[chan], chan == 3 ? max_alpha :
max_rgb);
- val[chan] = lp_build_emit_llvm_binary(bld_base,
+ val[chan] = lp_build_emit_llvm_binary(&ctx->bld_base,
TGSI_OPCODE_IMAX,
val[chan], chan == 3 ? min_alpha :
min_rgb);
}
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;
}
@@ -2303,25 +2302,23 @@ static LLVMValueRef
si_scale_alpha_by_sample_mask(struct lp_build_tgsi_context *
coverage = LLVMBuildUIToFP(ctx->ac.builder, coverage,
ctx->f32, "");
coverage = LLVMBuildFMul(ctx->ac.builder, coverage,
LLVMConstReal(ctx->f32,
1.0 / SI_NUM_SMOOTH_AA_SAMPLES), "");
return LLVMBuildFMul(ctx->ac.builder, alpha, coverage, "");
}
-static void si_llvm_emit_clipvertex(struct lp_build_tgsi_context *bld_base,
+static void si_llvm_emit_clipvertex(struct si_shader_context *ctx,
struct ac_export_args *pos, LLVMValueRef
*out_elts)
{
- struct si_shader_context *ctx = si_shader_context(bld_base);
- struct lp_build_context *base = &bld_base->base;
unsigned reg_index;
unsigned chan;
unsigned const_chan;
LLVMValueRef base_elt;
LLVMValueRef ptr = LLVMGetParam(ctx->main_fn, ctx->param_rw_buffers);
LLVMValueRef constbuf_index = LLVMConstInt(ctx->i32,
SI_VS_CONST_CLIP_PLANES, 0);
LLVMValueRef const_resource = ac_build_load_to_sgpr(&ctx->ac, ptr,
constbuf_index);
for (reg_index = 0; reg_index < 2; reg_index ++) {
@@ -2334,22 +2331,22 @@ static void si_llvm_emit_clipvertex(struct
lp_build_tgsi_context *bld_base,
/* Compute dot products of position and user clip plane vectors
*/
for (chan = 0; chan < TGSI_NUM_CHANNELS; chan++) {
for (const_chan = 0; const_chan < TGSI_NUM_CHANNELS;
const_chan++) {
LLVMValueRef addr =
LLVMConstInt(ctx->i32, ((reg_index * 4
+ chan) * 4 +
const_chan) *
4, 0);
base_elt = buffer_load_const(ctx,
const_resource,
addr);
args->out[chan] =
- lp_build_add(base, args->out[chan],
- lp_build_mul(base,
base_elt,
+ lp_build_add(&ctx->bld_base.base,
args->out[chan],
+
lp_build_mul(&ctx->bld_base.base, base_elt,
out_elts[const_chan]));
}
}
args->enabled_channels = 0xf;
args->valid_mask = 0;
args->done = 0;
args->target = V_008DFC_SQ_EXP_POS + 2 + reg_index;
args->compr = 0;
}
@@ -2505,21 +2502,21 @@ static void si_llvm_emit_streamout(struct
si_shader_context *ctx,
}
}
lp_build_endif(&if_ctx);
}
static void si_export_param(struct si_shader_context *ctx, unsigned index,
LLVMValueRef *values)
{
struct ac_export_args args;
- si_llvm_init_export_args(&ctx->bld_base, values,
+ si_llvm_init_export_args(ctx, values,
V_008DFC_SQ_EXP_PARAM + index, &args);
ac_build_export(&ctx->ac, &args);
}
static void si_build_param_exports(struct si_shader_context *ctx,
struct si_shader_output_values *outputs,
unsigned noutput)
{
struct si_shader *shader = ctx->shader;
unsigned param_count = 0;
@@ -2558,61 +2555,60 @@ static void si_build_param_exports(struct
si_shader_context *ctx,
si_export_param(ctx, param_count, outputs[i].values);
assert(i < ARRAY_SIZE(shader->info.vs_output_param_offset));
shader->info.vs_output_param_offset[i] = param_count++;
}
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,
+static void si_llvm_export_vs(struct si_shader_context *ctx,
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 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,
+ si_llvm_init_export_args(ctx, outputs[i].values,
V_008DFC_SQ_EXP_POS,
&pos_args[0]);
break;
case TGSI_SEMANTIC_PSIZE:
psize_value = outputs[i].values[0];
break;
case TGSI_SEMANTIC_LAYER:
layer_value = outputs[i].values[0];
break;
case TGSI_SEMANTIC_VIEWPORT_INDEX:
viewport_index_value = outputs[i].values[0];
break;
case TGSI_SEMANTIC_EDGEFLAG:
edgeflag_value = outputs[i].values[0];
break;
case TGSI_SEMANTIC_CLIPDIST:
if (!shader->key.opt.clip_disable) {
unsigned index = 2 + outputs[i].semantic_index;
- si_llvm_init_export_args(bld_base,
outputs[i].values,
+ si_llvm_init_export_args(ctx, outputs[i].values,
V_008DFC_SQ_EXP_POS +
index,
&pos_args[index]);
}
break;
case TGSI_SEMANTIC_CLIPVERTEX:
if (!shader->key.opt.clip_disable) {
- si_llvm_emit_clipvertex(bld_base, pos_args,
+ si_llvm_emit_clipvertex(ctx, pos_args,
outputs[i].values);
}
break;
}
}
/* 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 */
@@ -3332,21 +3328,21 @@ static void si_llvm_emit_vs_epilogue(struct
ac_shader_abi *abi,
outputs[i].semantic_index = 0;
outputs[i].values[0] = ac_to_float(&ctx->ac,
get_primitive_id(ctx, 0));
for (j = 1; j < 4; j++)
outputs[i].values[j] = LLVMConstReal(ctx->f32, 0);
memset(outputs[i].vertex_stream, 0,
sizeof(outputs[i].vertex_stream));
i++;
}
- si_llvm_export_vs(&ctx->bld_base, outputs, i);
+ si_llvm_export_vs(ctx, outputs, i);
FREE(outputs);
}
static void si_tgsi_emit_epilogue(struct lp_build_tgsi_context *bld_base)
{
struct si_shader_context *ctx = si_shader_context(bld_base);
ctx->abi.emit_outputs(&ctx->abi, RADEON_LLVM_MAX_OUTPUTS,
&ctx->outputs[0][0]);
}
@@ -3473,41 +3469,41 @@ static void si_export_mrt_color(struct
lp_build_tgsi_context *bld_base,
color[3] = si_scale_alpha_by_sample_mask(bld_base, color[3],
samplemask_param);
/* If last_cbuf > 0, FS_COLOR0_WRITES_ALL_CBUFS is true. */
if (ctx->shader->key.part.ps.epilog.last_cbuf > 0) {
struct ac_export_args args[8];
int c, last = -1;
/* Get the export arguments, also find out what the last one
is. */
for (c = 0; c <= ctx->shader->key.part.ps.epilog.last_cbuf;
c++) {
- si_llvm_init_export_args(bld_base, color,
+ si_llvm_init_export_args(ctx, color,
V_008DFC_SQ_EXP_MRT + c,
&args[c]);
if (args[c].enabled_channels)
last = c;
}
/* Emit all exports. */
for (c = 0; c <= ctx->shader->key.part.ps.epilog.last_cbuf;
c++) {
if (is_last && last == c) {
args[c].valid_mask = 1; /* whether the EXEC
mask is valid */
args[c].done = 1; /* DONE bit */
} else if (!args[c].enabled_channels)
continue; /* unnecessary NULL export */
memcpy(&exp->args[exp->num++], &args[c],
sizeof(args[c]));
}
} else {
struct ac_export_args args;
/* Export */
- si_llvm_init_export_args(bld_base, color, V_008DFC_SQ_EXP_MRT +
index,
+ si_llvm_init_export_args(ctx, color, V_008DFC_SQ_EXP_MRT +
index,
&args);
if (is_last) {
args.valid_mask = 1; /* whether the EXEC mask is valid
*/
args.done = 1; /* DONE bit */
} else if (!args.enabled_channels)
return; /* unnecessary NULL export */
memcpy(&exp->args[exp->num++], &args, sizeof(args));
}
}
@@ -5474,21 +5470,21 @@ si_generate_gs_copy_shader(struct si_screen *sscreen,
}
/* Streamout and exports. */
if (gs_selector->so.num_outputs) {
si_llvm_emit_streamout(&ctx, outputs,
gsinfo->num_outputs,
stream);
}
if (stream == 0)
- si_llvm_export_vs(bld_base, outputs,
gsinfo->num_outputs);
+ si_llvm_export_vs(&ctx, outputs, gsinfo->num_outputs);
LLVMBuildBr(builder, end_bb);
}
LLVMPositionBuilderAtEnd(builder, end_bb);
LLVMBuildRetVoid(ctx.ac.builder);
ctx.type = PIPE_SHADER_GEOMETRY; /* override for shader dumping */
si_llvm_optimize_module(&ctx);
--
2.11.0
_______________________________________________
mesa-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
