From: Nicolai Hähnle <nicolai.haeh...@amd.com>
---
src/amd/common/ac_nir_to_llvm.c | 30 +++++++++++++++++++-----------
src/amd/common/ac_shader_abi.h | 1 +
src/gallium/drivers/radeonsi/si_shader.c | 1 +
3 files changed, 21 insertions(+), 11 deletions(-)
diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
index 1cb920c..0457d43 100644
--- a/src/amd/common/ac_nir_to_llvm.c
+++ b/src/amd/common/ac_nir_to_llvm.c
@@ -60,20 +60,21 @@ struct ac_nir_context {
LLVMValueRef main_function;
LLVMBasicBlockRef continue_block;
LLVMBasicBlockRef break_block;
LLVMValueRef outputs[RADEON_LLVM_MAX_OUTPUTS * 4];
int num_locals;
LLVMValueRef *locals;
+ LLVMValueRef frag_pos[4];
LLVMValueRef ddxy_lds;
struct nir_to_llvm_context *nctx; /* TODO get rid of this */
};
struct nir_to_llvm_context {
struct ac_llvm_context ac;
const struct ac_nir_compiler_options *options;
struct ac_shader_variant_info *shader_info;
struct ac_shader_abi abi;
@@ -121,21 +122,20 @@ struct nir_to_llvm_context {
LLVMValueRef esgs_ring;
LLVMValueRef gsvs_ring;
LLVMValueRef hs_ring_tess_offchip;
LLVMValueRef hs_ring_tess_factor;
LLVMValueRef prim_mask;
LLVMValueRef sample_pos_offset;
LLVMValueRef persp_sample, persp_center, persp_centroid;
LLVMValueRef linear_sample, linear_center, linear_centroid;
- LLVMValueRef frag_pos[4];
LLVMTypeRef i1;
LLVMTypeRef i8;
LLVMTypeRef i16;
LLVMTypeRef i32;
LLVMTypeRef i64;
LLVMTypeRef v2i32;
LLVMTypeRef v3i32;
LLVMTypeRef v4i32;
LLVMTypeRef v8i32;
@@ -822,24 +822,25 @@ static void create_function(struct nir_to_llvm_context
*ctx)
add_user_sgpr_argument(&args, ctx->i32,
&ctx->sample_pos_offset); /* sample position offset */
add_sgpr_argument(&args, ctx->i32, &ctx->prim_mask); /* prim
mask */
add_vgpr_argument(&args, ctx->v2i32, &ctx->persp_sample); /*
persp sample */
add_vgpr_argument(&args, ctx->v2i32, &ctx->persp_center); /*
persp center */
add_vgpr_argument(&args, ctx->v2i32, &ctx->persp_centroid); /*
persp centroid */
add_vgpr_argument(&args, ctx->v3i32, NULL); /* persp pull model
*/
add_vgpr_argument(&args, ctx->v2i32, &ctx->linear_sample); /*
linear sample */
add_vgpr_argument(&args, ctx->v2i32, &ctx->linear_center); /*
linear center */
add_vgpr_argument(&args, ctx->v2i32, &ctx->linear_centroid); /*
linear centroid */
add_vgpr_argument(&args, ctx->f32, NULL); /* line stipple tex
*/
- add_vgpr_argument(&args, ctx->f32, &ctx->frag_pos[0]); /* pos
x float */
- add_vgpr_argument(&args, ctx->f32, &ctx->frag_pos[1]); /* pos
y float */
- add_vgpr_argument(&args, ctx->f32, &ctx->frag_pos[2]); /* pos
z float */
- add_vgpr_argument(&args, ctx->f32, &ctx->frag_pos[3]); /* pos
w float */
+ ctx->abi.param_frag_pos =
+ add_vgpr_argument(&args, ctx->f32, NULL); /* pos x
float */
+ add_vgpr_argument(&args, ctx->f32, NULL); /* pos y float */
+ add_vgpr_argument(&args, ctx->f32, NULL); /* pos z float */
+ add_vgpr_argument(&args, ctx->f32, NULL); /* pos w float */
ctx->abi.param_front_face =
add_vgpr_argument(&args, ctx->i32, NULL); /* front
face */
ctx->abi.param_ancillary =
add_vgpr_argument(&args, ctx->i32, NULL); /* ancillary
*/
ctx->abi.param_sample_coverage =
add_vgpr_argument(&args, ctx->i32, NULL); /* sample
coverage */
add_vgpr_argument(&args, ctx->i32, NULL); /* fixed pt */
break;
default:
unreachable("Shader stage not implemented");
@@ -3244,21 +3245,21 @@ static LLVMValueRef get_image_coords(struct
ac_nir_context *ctx,
int chan;
fmask_load_address[0] =
LLVMBuildExtractElement(ctx->ac.builder, src0, masks[0], "");
fmask_load_address[1] =
LLVMBuildExtractElement(ctx->ac.builder, src0, masks[1], "");
if (glsl_sampler_type_is_array(type))
fmask_load_address[2] =
LLVMBuildExtractElement(ctx->ac.builder, src0, masks[2], "");
else
fmask_load_address[2] = NULL;
if (add_frag_pos) {
for (chan = 0; chan < 2; ++chan)
- fmask_load_address[chan] =
LLVMBuildAdd(ctx->ac.builder, fmask_load_address[chan],
LLVMBuildFPToUI(ctx->ac.builder, ctx->nctx->frag_pos[chan], ctx->ac.i32, ""),
"");
+ fmask_load_address[chan] =
LLVMBuildAdd(ctx->ac.builder, fmask_load_address[chan],
LLVMBuildFPToUI(ctx->ac.builder, ctx->frag_pos[chan], ctx->ac.i32, ""), "");
}
sample_index = adjust_sample_index_using_fmask(&ctx->ac,
fmask_load_address[0],
fmask_load_address[1],
fmask_load_address[2],
sample_index,
get_sampler_desc(ctx, instr->variables[0], AC_DESC_FMASK, true, false));
}
if (count == 1) {
if (instr->src[0].ssa->num_components)
@@ -3268,21 +3269,21 @@ static LLVMValueRef get_image_coords(struct
ac_nir_context *ctx,
} else {
int chan;
if (is_ms)
count--;
for (chan = 0; chan < count; ++chan) {
coords[chan] = LLVMBuildExtractElement(ctx->ac.builder,
src0, masks[chan], "");
}
if (add_frag_pos) {
for (chan = 0; chan < count; ++chan)
- coords[chan] = LLVMBuildAdd(ctx->ac.builder,
coords[chan], LLVMBuildFPToUI(ctx->ac.builder, ctx->nctx->frag_pos[chan],
ctx->ac.i32, ""), "");
+ coords[chan] = LLVMBuildAdd(ctx->ac.builder,
coords[chan], LLVMBuildFPToUI(ctx->ac.builder, ctx->frag_pos[chan],
ctx->ac.i32, ""), "");
}
if (is_ms) {
coords[count] = sample_index;
count++;
}
if (count == 3) {
coords[3] = LLVMGetUndef(ctx->ac.i32);
count = 4;
}
@@ -3672,21 +3673,21 @@ static LLVMValueRef load_sample_position(struct
nir_to_llvm_context *ctx,
ptr = LLVMBuildBitCast(ctx->builder, ptr,
const_array(ctx->v2f32, 64), "");
sample_id = LLVMBuildAdd(ctx->builder, sample_id,
ctx->sample_pos_offset, "");
result = ac_build_indexed_load(&ctx->ac, ptr, sample_id, false);
return result;
}
-static LLVMValueRef load_sample_pos(struct nir_to_llvm_context *ctx)
+static LLVMValueRef load_sample_pos(struct ac_nir_context *ctx)
{
LLVMValueRef values[2];
values[0] = emit_ffract(&ctx->ac, ctx->frag_pos[0]);
values[1] = emit_ffract(&ctx->ac, ctx->frag_pos[1]);
return ac_build_gather_values(&ctx->ac, values, 2);
}
static LLVMValueRef visit_interp(struct nir_to_llvm_context *ctx,
const nir_intrinsic_instr *instr)
@@ -3930,21 +3931,21 @@ static void visit_intrinsic(struct ac_nir_context *ctx,
result = ctx->nctx->tes_patch_id;
else
fprintf(stderr, "Unknown primitive id intrinsic: %d",
ctx->stage);
break;
case nir_intrinsic_load_sample_id:
result = unpack_param(ctx->nctx,
LLVMGetParam(ctx->main_function,
ctx->abi->param_ancillary),
8, 4);
break;
case nir_intrinsic_load_sample_pos:
- result = load_sample_pos(ctx->nctx);
+ result = load_sample_pos(ctx);
break;
case nir_intrinsic_load_sample_mask_in:
result = LLVMGetParam(ctx->main_function,
ctx->abi->param_sample_coverage);
break;
case nir_intrinsic_load_front_face:
result = LLVMGetParam(ctx->main_function,
ctx->abi->param_front_face);
break;
case nir_intrinsic_load_instance_id:
result = LLVMGetParam(ctx->main_function,
ctx->abi->param_instance_id);
break;
@@ -4940,23 +4941,25 @@ handle_fs_inputs_pre(struct nir_to_llvm_context *ctx,
i == VARYING_SLOT_PRIMITIVE_ID || i == VARYING_SLOT_LAYER) {
interp_param = *inputs;
interp_fs_input(ctx, index, interp_param,
ctx->prim_mask,
inputs);
if (!interp_param)
ctx->shader_info->fs.flat_shaded_mask |= 1u <<
index;
++index;
} else if (i == VARYING_SLOT_POS) {
for(int i = 0; i < 3; ++i)
- inputs[i] = ctx->frag_pos[i];
+ inputs[i] = LLVMGetParam(ctx->main_function,
ctx->abi.param_frag_pos + i);
- inputs[3] = ac_build_fdiv(&ctx->ac, ctx->f32one,
ctx->frag_pos[3]);
+ inputs[3] = ac_build_fdiv(&ctx->ac, ctx->f32one,
+
LLVMGetParam(ctx->main_function,
+
ctx->abi.param_frag_pos + 3));
}
}
ctx->shader_info->fs.num_interp = index;
if (ctx->input_mask & (1 << VARYING_SLOT_PNTC))
ctx->shader_info->fs.has_pcoord = true;
if (ctx->input_mask & (1 << VARYING_SLOT_PRIMITIVE_ID))
ctx->shader_info->fs.prim_id_input = true;
if (ctx->input_mask & (1 << VARYING_SLOT_LAYER))
ctx->shader_info->fs.layer_input = true;
ctx->shader_info->fs.input_mask = ctx->input_mask >> VARYING_SLOT_VAR0;
@@ -6016,20 +6019,25 @@ void ac_nir_translate(struct ac_llvm_context *ac,
struct ac_shader_abi *abi,
ctx.defs = _mesa_hash_table_create(NULL, _mesa_hash_pointer,
_mesa_key_pointer_equal);
ctx.phis = _mesa_hash_table_create(NULL, _mesa_hash_pointer,
_mesa_key_pointer_equal);
func = (struct nir_function *)exec_list_get_head(&nir->functions);
setup_locals(&ctx, func);
+ if (nir->stage == MESA_SHADER_FRAGMENT) {
+ for (unsigned i = 0; i < 4; ++i)
+ ctx.frag_pos[i] = LLVMGetParam(ctx.main_function,
ctx.abi->param_frag_pos + i);
+ }
+
visit_cf_list(&ctx, &func->impl->body);
phi_post_pass(&ctx);
ctx.abi->emit_outputs(ctx.abi, RADEON_LLVM_MAX_OUTPUTS,
ctx.outputs);
free(ctx.locals);
ralloc_free(ctx.defs);
ralloc_free(ctx.phis);
diff --git a/src/amd/common/ac_shader_abi.h b/src/amd/common/ac_shader_abi.h
index 5caee91..9fc4390 100644
--- a/src/amd/common/ac_shader_abi.h
+++ b/src/amd/common/ac_shader_abi.h
@@ -35,20 +35,21 @@ enum ac_descriptor_type {
* radv to share a compiler backend.
*/
struct ac_shader_abi {
enum chip_class chip_class;
int param_base_vertex;
int param_start_instance;
int param_draw_id;
int param_vertex_id;
int param_instance_id;
+ int param_frag_pos;
int param_front_face;
int param_ancillary;
int param_sample_coverage;
/* For VS and PS: pre-loaded shader inputs.
*
* Currently only used for NIR shaders; indexed by variables'
* driver_location.
*/
LLVMValueRef *inputs;
diff --git a/src/gallium/drivers/radeonsi/si_shader.c
b/src/gallium/drivers/radeonsi/si_shader.c
index 5e53797..91bc067 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -4453,20 +4453,21 @@ static void create_function(struct si_shader_context
*ctx)
last_sgpr = SI_PARAM_PRIM_MASK;
params[SI_PARAM_PERSP_SAMPLE] = ctx->v2i32;
params[SI_PARAM_PERSP_CENTER] = ctx->v2i32;
params[SI_PARAM_PERSP_CENTROID] = ctx->v2i32;
params[SI_PARAM_PERSP_PULL_MODEL] = v3i32;
params[SI_PARAM_LINEAR_SAMPLE] = ctx->v2i32;
params[SI_PARAM_LINEAR_CENTER] = ctx->v2i32;
params[SI_PARAM_LINEAR_CENTROID] = ctx->v2i32;
params[SI_PARAM_LINE_STIPPLE_TEX] = ctx->f32;
params[SI_PARAM_POS_X_FLOAT] = ctx->f32;
+ ctx->abi.param_frag_pos = SI_PARAM_POS_X_FLOAT;
params[SI_PARAM_POS_Y_FLOAT] = ctx->f32;
params[SI_PARAM_POS_Z_FLOAT] = ctx->f32;
params[SI_PARAM_POS_W_FLOAT] = ctx->f32;
params[SI_PARAM_FRONT_FACE] = ctx->i32;
ctx->abi.param_front_face = SI_PARAM_FRONT_FACE;
shader->info.face_vgpr_index = 20;
params[SI_PARAM_ANCILLARY] = ctx->i32;
ctx->abi.param_ancillary = SI_PARAM_ANCILLARY;
params[SI_PARAM_SAMPLE_COVERAGE] = ctx->f32;
ctx->abi.param_sample_coverage = SI_PARAM_SAMPLE_COVERAGE;
--
2.9.3
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
