KonstantinSeurer
/
mesa
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

radeonsi: implement ARB_shader_group_vote 

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This commit is contained in:
Nicolai Hähnle 2017-03-29 20:29:37 +02:00
parent cd3f386069
commit 02112c3ef7
4 changed files with 87 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
docs/features.txt
View File

@ -295,7 +295,7 @@ Khronos, ARB, and OES extensions that are not part of any OpenGL or OpenGL ES ve
GL_ARB_shader_ballot not started
GL_ARB_shader_clock DONE (i965/gen7+, radeonsi)
GL_ARB_shader_draw_parameters DONE (i965, nvc0, radeonsi)
GL_ARB_shader_group_vote DONE (nvc0)
GL_ARB_shader_group_vote DONE (nvc0, radeonsi)
GL_ARB_shader_stencil_export DONE (i965/gen9+, radeonsi, softpipe, llvmpipe, swr)
GL_ARB_shader_viewport_layer_array DONE (i965/gen6+)
GL_ARB_sparse_buffer not started

1
docs/relnotes/17.1.0.html
View File

@ -46,6 +46,7 @@ Note: some of the new features are only available with certain drivers.
<ul>
<li>GL_ARB_gpu_shader_int64 on i965/gen8+, nvc0, radeonsi, softpipe, llvmpipe</li>
<li>GL_ARB_shader_clock on radeonsi</li>
<li>GL_ARB_shader_group_vote on radeonsi</li>
<li>GL_ARB_transform_feedback2 on i965/gen6</li>
<li>GL_ARB_transform_feedback_overflow_query on i965/gen6+</li>
<li>Geometry shaders enabled on swr</li>

4
src/gallium/drivers/radeonsi/si_pipe.c
View File

@ -424,6 +424,9 @@ static int si_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
case PIPE_CAP_TGSI_CLOCK:
return HAVE_LLVM >= 0x0309;
case PIPE_CAP_TGSI_VOTE:
return HAVE_LLVM >= 0x0400;
case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
return !SI_BIG_ENDIAN && sscreen->b.info.has_userptr;
@ -480,7 +483,6 @@ static int si_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
case PIPE_CAP_VERTEXID_NOBASE:
case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES:
case PIPE_CAP_TGSI_VOTE:
case PIPE_CAP_MAX_WINDOW_RECTANGLES:
case PIPE_CAP_NATIVE_FENCE_FD:
case PIPE_CAP_TGSI_FS_FBFETCH:

82
src/gallium/drivers/radeonsi/si_shader.c
View File

@ -5125,6 +5125,84 @@ static void build_interp_intrinsic(const struct lp_build_tgsi_action *action,
}
}
static LLVMValueRef si_emit_ballot(struct si_shader_context *ctx,
LLVMValueRef value)
{
struct gallivm_state *gallivm = &ctx->gallivm;
LLVMValueRef args[3] = {
value,
ctx->i32_0,
LLVMConstInt(ctx->i32, LLVMIntNE, 0)
};
if (LLVMTypeOf(value) != ctx->i32)
args[0] = LLVMBuildBitCast(gallivm->builder, value, ctx->i32, "");
return lp_build_intrinsic(gallivm->builder,
"llvm.amdgcn.icmp.i32",
ctx->i64, args, 3,
LP_FUNC_ATTR_NOUNWIND |
LP_FUNC_ATTR_READNONE |
LP_FUNC_ATTR_CONVERGENT);
}
static void vote_all_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 = &ctx->gallivm;
LLVMValueRef active_set, vote_set;
LLVMValueRef tmp;
active_set = si_emit_ballot(ctx, ctx->i32_1);
vote_set = si_emit_ballot(ctx, emit_data->args[0]);
tmp = LLVMBuildICmp(gallivm->builder, LLVMIntEQ, vote_set, active_set, "");
emit_data->output[emit_data->chan] =
LLVMBuildSExt(gallivm->builder, tmp, ctx->i32, "");
}
static void vote_any_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 = &ctx->gallivm;
LLVMValueRef vote_set;
LLVMValueRef tmp;
vote_set = si_emit_ballot(ctx, emit_data->args[0]);
tmp = LLVMBuildICmp(gallivm->builder, LLVMIntNE,
vote_set, LLVMConstInt(ctx->i64, 0, 0), "");
emit_data->output[emit_data->chan] =
LLVMBuildSExt(gallivm->builder, tmp, ctx->i32, "");
}
static void vote_eq_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 = &ctx->gallivm;
LLVMValueRef active_set, vote_set;
LLVMValueRef all, none, tmp;
active_set = si_emit_ballot(ctx, ctx->i32_1);
vote_set = si_emit_ballot(ctx, emit_data->args[0]);
all = LLVMBuildICmp(gallivm->builder, LLVMIntEQ, vote_set, active_set, "");
none = LLVMBuildICmp(gallivm->builder, LLVMIntEQ,
vote_set, LLVMConstInt(ctx->i64, 0, 0), "");
tmp = LLVMBuildOr(gallivm->builder, all, none, "");
emit_data->output[emit_data->chan] =
LLVMBuildSExt(gallivm->builder, tmp, ctx->i32, "");
}
static unsigned si_llvm_get_stream(struct lp_build_tgsi_context *bld_base,
struct lp_build_emit_data *emit_data)
{
@ -6574,6 +6652,10 @@ static void si_init_shader_ctx(struct si_shader_context *ctx,
bld_base->op_actions[TGSI_OPCODE_DDX_FINE].emit = si_llvm_emit_ddxy;
bld_base->op_actions[TGSI_OPCODE_DDY_FINE].emit = si_llvm_emit_ddxy;
bld_base->op_actions[TGSI_OPCODE_VOTE_ALL].emit = vote_all_emit;
bld_base->op_actions[TGSI_OPCODE_VOTE_ANY].emit = vote_any_emit;
bld_base->op_actions[TGSI_OPCODE_VOTE_EQ].emit = vote_eq_emit;
bld_base->op_actions[TGSI_OPCODE_EMIT].emit = si_llvm_emit_vertex;
bld_base->op_actions[TGSI_OPCODE_ENDPRIM].emit = si_llvm_emit_primitive;
bld_base->op_actions[TGSI_OPCODE_BARRIER].emit = si_llvm_emit_barrier;