OpenHarmony/third_party_mesa3d
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

ac/llvm: remove gep_2 and others temporary functions

Browse Source 
Reviewed-by: Mihai Preda <mhpreda@gmail.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Acked-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19035>
This commit is contained in:
Pierre-Eric Pelloux-Prayer
2022-10-11 10:39:18 +02:00
committed by Marge Bot
parent 7508cdd2ff
commit b31affddf4
10 changed files with 104 additions and 157 deletions
Download Patch File Download Diff File Expand all files Collapse all files

75
src/amd/llvm/ac_llvm_build.c
View File

@@ -1090,15 +1090,6 @@ LLVMValueRef ac_build_gep_ptr(struct ac_llvm_context *ctx, LLVMTypeRef type, LLV
return LLVMBuildGEP2(ctx->builder, type, base_ptr, &index, 1, "");
}
LLVMValueRef ac_build_gep0(struct ac_llvm_context *ctx, LLVMValueRef base_ptr, LLVMValueRef index)
{
LLVMValueRef indices[2] = {
ctx->i32_0,
index,
};
return LLVMBuildGEP(ctx->builder, base_ptr, indices, 2, "");
}
LLVMTypeRef ac_build_gep0_type(LLVMTypeRef pointee_type, LLVMValueRef index)
{
switch (LLVMGetTypeKind(pointee_type)) {
@@ -1121,7 +1112,7 @@ LLVMTypeRef ac_build_gep0_type(LLVMTypeRef pointee_type, LLVMValueRef index)
return NULL;
}
LLVMValueRef ac_build_gep0_2(struct ac_llvm_context *ctx, LLVMTypeRef pointee_type, LLVMValueRef value, LLVMValueRef index)
LLVMValueRef ac_build_gep0(struct ac_llvm_context *ctx, LLVMTypeRef pointee_type, LLVMValueRef value, LLVMValueRef index)
{
LLVMValueRef indices[2] = {
ctx->i32_0,
@@ -1136,16 +1127,10 @@ LLVMValueRef ac_build_pointer_add(struct ac_llvm_context *ctx, LLVMTypeRef type,
return LLVMBuildGEP2(ctx->builder, type, ptr, &index, 1, "");
}
void ac_build_indexed_store(struct ac_llvm_context *ctx, LLVMValueRef base_ptr, LLVMValueRef index,
void ac_build_indexed_store(struct ac_llvm_context *ctx, LLVMTypeRef type, LLVMValueRef base_ptr, LLVMValueRef index,
LLVMValueRef value)
{
LLVMBuildStore(ctx->builder, value, ac_build_gep0(ctx, base_ptr, index));
}
void ac_build_indexed_store2(struct ac_llvm_context *ctx, LLVMTypeRef type, LLVMValueRef base_ptr, LLVMValueRef index,
LLVMValueRef value)
{
LLVMBuildStore(ctx->builder, value, ac_build_gep0_2(ctx, type, base_ptr, index));
LLVMBuildStore(ctx->builder, value, ac_build_gep0(ctx, type, base_ptr, index));
}
/**
@@ -1176,30 +1161,9 @@ void ac_build_indexed_store2(struct ac_llvm_context *ctx, LLVMTypeRef type, LLVM
* ptr2 = LLVMBuildInBoundsGEP(ptr1, 32 / elemsize);
* sampler = load(ptr2); // becomes "s_load ptr1, 32" thanks to InBounds
*/
static LLVMValueRef ac_build_load_custom(struct ac_llvm_context *ctx, LLVMValueRef base_ptr,
LLVMValueRef index, bool uniform, bool invariant,
bool no_unsigned_wraparound)
{
LLVMValueRef pointer, result;
if (no_unsigned_wraparound &&
LLVMGetPointerAddressSpace(LLVMTypeOf(base_ptr)) == AC_ADDR_SPACE_CONST_32BIT)
pointer = LLVMBuildInBoundsGEP(ctx->builder, base_ptr, &index, 1, "");
else
pointer = LLVMBuildGEP(ctx->builder, base_ptr, &index, 1, "");
if (uniform)
LLVMSetMetadata(pointer, ctx->uniform_md_kind, ctx->empty_md);
result = LLVMBuildLoad(ctx->builder, pointer, "");
if (invariant)
LLVMSetMetadata(result, ctx->invariant_load_md_kind, ctx->empty_md);
LLVMSetAlignment(result, 4);
return result;
}
static LLVMValueRef ac_build_load_custom2(struct ac_llvm_context *ctx, LLVMTypeRef type,
LLVMValueRef base_ptr, LLVMValueRef index,
bool uniform, bool invariant, bool no_unsigned_wraparound)
static LLVMValueRef ac_build_load_custom(struct ac_llvm_context *ctx, LLVMTypeRef type,
LLVMValueRef base_ptr, LLVMValueRef index,
bool uniform, bool invariant, bool no_unsigned_wraparound)
{
LLVMValueRef pointer, result;
@@ -1220,39 +1184,28 @@ static LLVMValueRef ac_build_load_custom2(struct ac_llvm_context *ctx, LLVMTypeR
LLVMValueRef ac_build_load(struct ac_llvm_context *ctx, LLVMTypeRef type, LLVMValueRef base_ptr, LLVMValueRef index)
{
return ac_build_load_custom2(ctx, type, base_ptr, index, false, false, false);
return ac_build_load_custom(ctx, type, base_ptr, index, false, false, false);
}
LLVMValueRef ac_build_load_invariant(struct ac_llvm_context *ctx, LLVMTypeRef type, LLVMValueRef base_ptr,
LLVMValueRef index)
{
return ac_build_load_custom2(ctx, type, base_ptr, index, false, true, false);
return ac_build_load_custom(ctx, type, base_ptr, index, false, true, false);
}
/* This assumes that there is no unsigned integer wraparound during the address
* computation, excluding all GEPs within base_ptr. */
LLVMValueRef ac_build_load_to_sgpr(struct ac_llvm_context *ctx, LLVMValueRef base_ptr,
LLVMValueRef ac_build_load_to_sgpr(struct ac_llvm_context *ctx, LLVMTypeRef type, LLVMValueRef base_ptr,
LLVMValueRef index)
{
return ac_build_load_custom(ctx, base_ptr, index, true, true, true);
}
LLVMValueRef ac_build_load_to_sgpr2(struct ac_llvm_context *ctx, LLVMTypeRef type, LLVMValueRef base_ptr,
LLVMValueRef index)
{
return ac_build_load_custom2(ctx, type, base_ptr, index, true, true, true);
return ac_build_load_custom(ctx, type, base_ptr, index, true, true, true);
}
/* See ac_build_load_custom() documentation. */
LLVMValueRef ac_build_load_to_sgpr_uint_wraparound(struct ac_llvm_context *ctx,
LLVMValueRef ac_build_load_to_sgpr_uint_wraparound(struct ac_llvm_context *ctx, LLVMTypeRef type,
LLVMValueRef base_ptr, LLVMValueRef index)
{
return ac_build_load_custom(ctx, base_ptr, index, true, true, false);
}
LLVMValueRef ac_build_load_to_sgpr_uint_wraparound2(struct ac_llvm_context *ctx, LLVMTypeRef type,
LLVMValueRef base_ptr, LLVMValueRef index)
{
return ac_build_load_custom2(ctx, type, base_ptr, index, true, true, false);
return ac_build_load_custom(ctx, type, base_ptr, index, true, true, false);
}
static unsigned get_load_cache_policy(struct ac_llvm_context *ctx, unsigned cache_policy)
@@ -2830,14 +2783,14 @@ void ac_declare_lds_as_pointer(struct ac_llvm_context *ctx)
LLVMValueRef ac_lds_load(struct ac_llvm_context *ctx, LLVMValueRef dw_addr)
{
LLVMValueRef v = ac_build_gep0_2(ctx, ctx->lds.t, ctx->lds.v, dw_addr);
LLVMValueRef v = ac_build_gep0(ctx, ctx->lds.t, ctx->lds.v, dw_addr);
return LLVMBuildLoad2(ctx->builder, ctx->i32, v, "");
}
void ac_lds_store(struct ac_llvm_context *ctx, LLVMValueRef dw_addr, LLVMValueRef value)
{
value = ac_to_integer(ctx, value);
ac_build_indexed_store2(ctx, ctx->lds.t, ctx->lds.v, dw_addr, value);
ac_build_indexed_store(ctx, ctx->lds.t, ctx->lds.v, dw_addr, value);
}
LLVMValueRef ac_find_lsb(struct ac_llvm_context *ctx, LLVMTypeRef dst_type, LLVMValueRef src0)

16
src/amd/llvm/ac_llvm_build.h
View File

@@ -254,28 +254,22 @@ LLVMValueRef ac_build_fs_interp_mov(struct ac_llvm_context *ctx, LLVMValueRef pa
LLVMValueRef ac_build_gep_ptr(struct ac_llvm_context *ctx, LLVMTypeRef type, LLVMValueRef base_ptr,
LLVMValueRef index);
LLVMValueRef ac_build_gep0(struct ac_llvm_context *ctx, LLVMValueRef value, LLVMValueRef index);
LLVMValueRef ac_build_pointer_add(struct ac_llvm_context *ctx, LLVMTypeRef type, LLVMValueRef ptr,
LLVMValueRef index);
LLVMTypeRef ac_build_gep0_type(LLVMTypeRef pointee_type, LLVMValueRef index);
LLVMValueRef ac_build_gep0_2(struct ac_llvm_context *ctx, LLVMTypeRef type, LLVMValueRef value, LLVMValueRef index);
LLVMValueRef ac_build_gep0(struct ac_llvm_context *ctx, LLVMTypeRef type, LLVMValueRef value, LLVMValueRef index);
void ac_build_indexed_store(struct ac_llvm_context *ctx, LLVMValueRef base_ptr, LLVMValueRef index,
void ac_build_indexed_store(struct ac_llvm_context *ctx, LLVMTypeRef type, LLVMValueRef base_ptr, LLVMValueRef index,
LLVMValueRef value);
void ac_build_indexed_store2(struct ac_llvm_context *ctx, LLVMTypeRef type, LLVMValueRef base_ptr, LLVMValueRef index,
LLVMValueRef value);
LLVMValueRef ac_build_load(struct ac_llvm_context *ctx, LLVMTypeRef type, LLVMValueRef base_ptr, LLVMValueRef index);
LLVMValueRef ac_build_load_invariant(struct ac_llvm_context *ctx, LLVMTypeRef type, LLVMValueRef base_ptr,
LLVMValueRef index);
LLVMValueRef ac_build_load_to_sgpr(struct ac_llvm_context *ctx, LLVMValueRef base_ptr,
LLVMValueRef ac_build_load_to_sgpr(struct ac_llvm_context *ctx, LLVMTypeRef type, LLVMValueRef base_ptr,
LLVMValueRef index);
LLVMValueRef ac_build_load_to_sgpr2(struct ac_llvm_context *ctx, LLVMTypeRef type, LLVMValueRef base_ptr,
LLVMValueRef index);
LLVMValueRef ac_build_load_to_sgpr_uint_wraparound(struct ac_llvm_context *ctx,
LLVMValueRef base_ptr, LLVMValueRef index);
LLVMValueRef ac_build_load_to_sgpr_uint_wraparound2(struct ac_llvm_context *ctx, LLVMTypeRef type,
LLVMValueRef ac_build_load_to_sgpr_uint_wraparound(struct ac_llvm_context *ctx, LLVMTypeRef type,
LLVMValueRef base_ptr, LLVMValueRef index);
void ac_build_buffer_store_dword(struct ac_llvm_context *ctx, LLVMValueRef rsrc, LLVMValueRef vdata,

6
src/amd/llvm/ac_nir_to_llvm.c
View File

@@ -4115,7 +4115,7 @@ static bool visit_intrinsic(struct ac_nir_context *ctx, nir_intrinsic_instr *ins
break;
case nir_intrinsic_load_scratch: {
LLVMValueRef offset = get_src(ctx, instr->src[0]);
LLVMValueRef ptr = ac_build_gep0_2(&ctx->ac, ctx->scratch.t, ctx->scratch.v, offset);
LLVMValueRef ptr = ac_build_gep0(&ctx->ac, ctx->scratch.t, ctx->scratch.v, offset);
LLVMTypeRef comp_type = LLVMIntTypeInContext(ctx->ac.context, instr->dest.ssa.bit_size);
LLVMTypeRef vec_type = instr->dest.ssa.num_components == 1
? comp_type
@@ -4125,7 +4125,7 @@ static bool visit_intrinsic(struct ac_nir_context *ctx, nir_intrinsic_instr *ins
}
case nir_intrinsic_store_scratch: {
LLVMValueRef offset = get_src(ctx, instr->src[1]);
LLVMValueRef ptr = ac_build_gep0_2(&ctx->ac, ctx->scratch.t, ctx->scratch.v, offset);
LLVMValueRef ptr = ac_build_gep0(&ctx->ac, ctx->scratch.t, ctx->scratch.v, offset);
LLVMTypeRef comp_type = LLVMIntTypeInContext(ctx->ac.context, instr->src[0].ssa->bit_size);
LLVMValueRef src = get_src(ctx, instr->src[0]);
unsigned wrmask = nir_intrinsic_write_mask(instr);
@@ -4154,7 +4154,7 @@ static bool visit_intrinsic(struct ac_nir_context *ctx, nir_intrinsic_instr *ins
LLVMValueRef cond = LLVMBuildICmp(ctx->ac.builder, LLVMIntULT, offset, size, "");
offset = LLVMBuildSelect(ctx->ac.builder, cond, offset, size, "");
LLVMValueRef ptr = ac_build_gep0_2(&ctx->ac, ctx->constant_data.t, ctx->constant_data.v, offset);
LLVMValueRef ptr = ac_build_gep0(&ctx->ac, ctx->constant_data.t, ctx->constant_data.v, offset);
LLVMTypeRef comp_type = LLVMIntTypeInContext(ctx->ac.context, instr->dest.ssa.bit_size);
LLVMTypeRef vec_type = instr->dest.ssa.num_components == 1
? comp_type

16
src/amd/vulkan/radv_nir_to_llvm.c
View File

@@ -114,7 +114,7 @@ load_descriptor_sets(struct radv_shader_context *ctx)
int i = u_bit_scan(&mask);
ctx->descriptor_sets[i] =
ac_build_load_to_sgpr2(&ctx->ac, type, desc_sets, LLVMConstInt(ctx->ac.i32, i, false));
ac_build_load_to_sgpr(&ctx->ac, type, desc_sets, LLVMConstInt(ctx->ac.i32, i, false));
LLVMSetAlignment(ctx->descriptor_sets[i], 4);
}
} else {
@@ -400,7 +400,7 @@ load_vs_input(struct radv_shader_context *ctx, unsigned driver_location, LLVMTyp
desc_index = util_bitcount(ctx->shader_info->vs.vb_desc_usage_mask &
u_bit_consecutive(0, desc_index));
t_offset = LLVMConstInt(ctx->ac.i32, desc_index, false);
t_list = ac_build_load_to_sgpr2(&ctx->ac, t_list_type, t_list_ptr, t_offset);
t_list = ac_build_load_to_sgpr(&ctx->ac, t_list_type, t_list_ptr, t_offset);
/* Always split typed vertex buffer loads on GFX6 and GFX10+ to avoid any alignment issues that
* triggers memory violations and eventually a GPU hang. This can happen if the stride (static or
@@ -807,7 +807,7 @@ radv_emit_streamout(struct radv_shader_context *ctx, unsigned stream)
LLVMValueRef offset = LLVMConstInt(ctx->ac.i32, i, false);
so_buffers[i] = ac_build_load_to_sgpr2(&ctx->ac, type, buf_ptr, offset);
so_buffers[i] = ac_build_load_to_sgpr(&ctx->ac, type, buf_ptr, offset);
LLVMValueRef so_offset = ac_get_arg(&ctx->ac, ctx->args->ac.streamout_offset[i]);
@@ -1181,11 +1181,11 @@ ac_setup_rings(struct radv_shader_context *ctx)
unsigned ring = ctx->stage == MESA_SHADER_GEOMETRY ? RING_ESGS_GS : RING_ESGS_VS;
LLVMValueRef offset = LLVMConstInt(ctx->ac.i32, ring, false);
ctx->esgs_ring = ac_build_load_to_sgpr2(&ctx->ac, ctx->ac.i8, ctx->ring_offsets, offset);
ctx->esgs_ring = ac_build_load_to_sgpr(&ctx->ac, ctx->ac.i8, ctx->ring_offsets, offset);
}
if (ctx->args->is_gs_copy_shader) {
ctx->gsvs_ring[0] = ac_build_load_to_sgpr2(&ctx->ac, ctx->ac.i8, ctx->ring_offsets,
ctx->gsvs_ring[0] = ac_build_load_to_sgpr(&ctx->ac, ctx->ac.i8, ctx->ring_offsets,
LLVMConstInt(ctx->ac.i32, RING_GSVS_VS, false));
}
@@ -1203,7 +1203,7 @@ ac_setup_rings(struct radv_shader_context *ctx)
unsigned num_records = ctx->ac.wave_size;
LLVMValueRef base_ring;
base_ring = ac_build_load_to_sgpr2(&ctx->ac, ctx->ac.i8, ctx->ring_offsets,
base_ring = ac_build_load_to_sgpr(&ctx->ac, ctx->ac.i8, ctx->ring_offsets,
LLVMConstInt(ctx->ac.i32, RING_GSVS_GS, false));
for (unsigned stream = 0; stream < 4; stream++) {
@@ -1243,9 +1243,9 @@ ac_setup_rings(struct radv_shader_context *ctx)
}
if (ctx->stage == MESA_SHADER_TESS_CTRL || ctx->stage == MESA_SHADER_TESS_EVAL) {
ctx->hs_ring_tess_offchip = ac_build_load_to_sgpr2(
ctx->hs_ring_tess_offchip = ac_build_load_to_sgpr(
&ctx->ac, ctx->ac.i8, ctx->ring_offsets, LLVMConstInt(ctx->ac.i32, RING_HS_TESS_OFFCHIP, false));
ctx->hs_ring_tess_factor = ac_build_load_to_sgpr2(
ctx->hs_ring_tess_factor = ac_build_load_to_sgpr(
&ctx->ac, ctx->ac.i8, ctx->ring_offsets, LLVMConstInt(ctx->ac.i32, RING_HS_TESS_FACTOR, false));
}
}

84
src/gallium/drivers/radeonsi/gfx10_shader_ngg.c
View File

@@ -67,8 +67,8 @@ static LLVMValueRef ngg_get_query_buf(struct si_shader_context *ctx)
{
LLVMValueRef buf_ptr = ac_get_arg(&ctx->ac, ctx->internal_bindings);
LLVMTypeRef type = ac_get_arg_pointee_type(&ctx->ac, &ctx->args, ctx->internal_bindings);
return ac_build_load_to_sgpr2(&ctx->ac, type, buf_ptr,
LLVMConstInt(ctx->ac.i32, SI_GS_QUERY_BUF, false));
return ac_build_load_to_sgpr(&ctx->ac, type, buf_ptr,
LLVMConstInt(ctx->ac.i32, SI_GS_QUERY_BUF, false));
}
static LLVMValueRef ngg_get_emulated_counters_buf(struct si_shader_context *ctx)
@@ -76,8 +76,8 @@ static LLVMValueRef ngg_get_emulated_counters_buf(struct si_shader_context *ctx)
LLVMValueRef buf_ptr = ac_get_arg(&ctx->ac, ctx->internal_bindings);
LLVMTypeRef type = ac_get_arg_pointee_type(&ctx->ac, &ctx->args, ctx->internal_bindings);
return ac_build_load_to_sgpr2(&ctx->ac, type, buf_ptr,
LLVMConstInt(ctx->ac.i32, SI_GS_QUERY_EMULATED_COUNTERS_BUF, false));
return ac_build_load_to_sgpr(&ctx->ac, type, buf_ptr,
LLVMConstInt(ctx->ac.i32, SI_GS_QUERY_EMULATED_COUNTERS_BUF, false));
}
/**
@@ -253,7 +253,7 @@ static void build_streamout_vertex(struct si_shader_context *ctx, LLVMValueRef *
for (unsigned comp = 0; comp < 4; comp++) {
LLVMValueRef idx = LLVMConstInt(ctx->ac.i32, 4 * reg + comp, false);
LLVMValueRef v = ac_build_gep0_2(&ctx->ac, vertexptr.t, vertexptr.v, idx);
LLVMValueRef v = ac_build_gep0(&ctx->ac, vertexptr.t, vertexptr.v, idx);
out.values[comp] = LLVMBuildLoad2(builder, ac_build_gep0_type(vertexptr.t, idx), v, "");
out.vertex_streams = info->output_streams[reg];
}
@@ -327,7 +327,7 @@ static void build_streamout(struct si_shader_context *ctx, struct ngg_streamout
ac_build_writelane(&ctx->ac, prim_stride_dw_vgpr, prim_stride_dw[buffer],
LLVMConstInt(ctx->ac.i32, buffer, false));
so_buffer[buffer] = ac_build_load_to_sgpr2(
so_buffer[buffer] = ac_build_load_to_sgpr(
&ctx->ac, type, buf_ptr, LLVMConstInt(ctx->ac.i32, SI_VS_STREAMOUT_BUF0 + buffer, false));
}
@@ -345,7 +345,7 @@ static void build_streamout(struct si_shader_context *ctx, struct ngg_streamout
ac_build_ifcc(&ctx->ac, tmp, 5210);
{
if (isgs) {
LLVMValueRef vt = ac_build_gep0_2(&ctx->ac, ctx->gs_ngg_scratch.t, ctx->gs_ngg_scratch.v, tid);
LLVMValueRef vt = ac_build_gep0(&ctx->ac, ctx->gs_ngg_scratch.t, ctx->gs_ngg_scratch.v, tid);
tmp = LLVMBuildLoad2(builder, ac_build_gep0_type(ctx->gs_ngg_scratch.t, tid), vt, "");
} else {
tmp = ac_build_writelane(&ctx->ac, ctx->ac.i32_0, ngg_get_prim_cnt(ctx), ctx->ac.i32_0);
@@ -452,7 +452,7 @@ static void build_streamout(struct si_shader_context *ctx, struct ngg_streamout
LLVMBuildStore(builder, tmp, offsets_vgpr);
tmp2 = LLVMBuildAdd(builder, ac_get_thread_id(&ctx->ac), scratch_offset_basev, "");
tmp2 = ac_build_gep0_2(&ctx->ac, ctx->gs_ngg_scratch.t, ctx->gs_ngg_scratch.v, tmp2);
tmp2 = ac_build_gep0(&ctx->ac, ctx->gs_ngg_scratch.t, ctx->gs_ngg_scratch.v, tmp2);
LLVMBuildStore(builder, tmp, tmp2);
}
ac_build_endif(&ctx->ac, 5210);
@@ -546,7 +546,7 @@ static void build_streamout(struct si_shader_context *ctx, struct ngg_streamout
ac_build_ifcc(&ctx->ac, tmp, 5225);
{
tmp = LLVMBuildAdd(builder, ac_get_thread_id(&ctx->ac), scratch_emit_basev, "");
tmp = ac_build_gep0_2(&ctx->ac, ctx->gs_ngg_scratch.t, ctx->gs_ngg_scratch.v, tmp);
tmp = ac_build_gep0(&ctx->ac, ctx->gs_ngg_scratch.t, ctx->gs_ngg_scratch.v, tmp);
LLVMBuildStore(builder, emit_vgpr, tmp);
}
ac_build_endif(&ctx->ac, 5225);
@@ -566,7 +566,7 @@ static void build_streamout(struct si_shader_context *ctx, struct ngg_streamout
primemit_scan[stream].enable_exclusive = true;
primemit_scan[stream].op = nir_op_iadd;
primemit_scan[stream].src = nggso->prim_enable[stream];
primemit_scan[stream].scratch = ac_build_gep0_2(
primemit_scan[stream].scratch = ac_build_gep0(
&ctx->ac, ctx->gs_ngg_scratch.t, ctx->gs_ngg_scratch.v,
LLVMConstInt(ctx->ac.i32, 12 + 8 * stream, false));
primemit_scan[stream].waveidx = get_wave_id_in_tg(ctx);
@@ -592,7 +592,7 @@ static void build_streamout(struct si_shader_context *ctx, struct ngg_streamout
LLVMValueRef scratch_vgpr;
LLVMValueRef idx = ac_get_thread_id(&ctx->ac);
LLVMValueRef v = ac_build_gep0_2(&ctx->ac, ctx->gs_ngg_scratch.t, ctx->gs_ngg_scratch.v, idx);
LLVMValueRef v = ac_build_gep0(&ctx->ac, ctx->gs_ngg_scratch.t, ctx->gs_ngg_scratch.v, idx);
scratch_vgpr = LLVMBuildLoad2(builder, ac_build_gep0_type(ctx->gs_ngg_scratch.t, idx), v, "");
for (unsigned buffer = 0; buffer < 4; ++buffer) {
@@ -779,7 +779,7 @@ static void load_vertex_counts(struct si_shader_context *ctx, struct ac_llvm_poi
*/
ac_build_ifcc(&ctx->ac, LLVMBuildICmp(builder, LLVMIntULT, tid,
LLVMConstInt(ctx->ac.i32, num_i8vec4, 0), ""), 17771);
LLVMValueRef v = ac_build_gep0_2(&ctx->ac, lds.t, lds.v, tid);
LLVMValueRef v = ac_build_gep0(&ctx->ac, lds.t, lds.v, tid);
LLVMBuildStore(builder, LLVMBuildLoad2(builder, ac_build_gep0_type(lds.t, tid), v, ""), i8vec4_lane);
ac_build_endif(&ctx->ac, 17771);
@@ -956,8 +956,8 @@ static void cull_primitive(struct si_shader_context *ctx,
LLVMValueRef ptr = ac_get_arg(&ctx->ac, ctx->small_prim_cull_info);
LLVMTypeRef type = ac_get_arg_pointee_type(&ctx->ac, &ctx->args, ctx->small_prim_cull_info);
/* Lines will always use the non-AA viewport transformation. */
LLVMValueRef vp = ac_build_load_to_sgpr2(&ctx->ac, type, ptr,
prim_is_lines ? ctx->ac.i32_1 : ctx->ac.i32_0);
LLVMValueRef vp = ac_build_load_to_sgpr(&ctx->ac, type, ptr,
prim_is_lines ? ctx->ac.i32_1 : ctx->ac.i32_0);
vp = LLVMBuildBitCast(builder, vp, ctx->ac.v4f32, "");
vp_scale[0] = ac_llvm_extract_elem(&ctx->ac, vp, 0);
vp_scale[1] = ac_llvm_extract_elem(&ctx->ac, vp, 1);
@@ -970,7 +970,7 @@ static void cull_primitive(struct si_shader_context *ctx,
options.cull_w = true;
if (prim_is_lines) {
LLVMValueRef terms = ac_build_load_to_sgpr2(&ctx->ac, ctx->ac.v2f32, ptr, LLVMConstInt(ctx->ac.i32, 4, 0));
LLVMValueRef terms = ac_build_load_to_sgpr(&ctx->ac, ctx->ac.v2f32, ptr, LLVMConstInt(ctx->ac.i32, 4, 0));
terms = LLVMBuildBitCast(builder, terms, ctx->ac.v2f32, "");
clip_half_line_width[0] = ac_llvm_extract_elem(&ctx->ac, terms, 0);
clip_half_line_width[1] = ac_llvm_extract_elem(&ctx->ac, terms, 1);
@@ -1062,14 +1062,14 @@ void gfx10_ngg_culling_build_end(struct si_shader_context *ctx)
/* Store Position.W into LDS. */
LLVMBuildStore(
builder, ac_to_integer(&ctx->ac, position[3]),
ac_build_gep0_2(&ctx->ac, es_vtxptr.t, es_vtxptr.v, LLVMConstInt(ctx->ac.i32, lds_pos_cull_w, 0)));
ac_build_gep0(&ctx->ac, es_vtxptr.t, es_vtxptr.v, LLVMConstInt(ctx->ac.i32, lds_pos_cull_w, 0)));
/* Store Position.XY / W into LDS. */
for (unsigned chan = 0; chan < 2; chan++) {
LLVMValueRef val = ac_build_fdiv(&ctx->ac, position[chan], position[3]);
LLVMBuildStore(
builder, ac_to_integer(&ctx->ac, val),
ac_build_gep0_2(&ctx->ac, es_vtxptr.t, es_vtxptr.v, LLVMConstInt(ctx->ac.i32, lds_pos_cull_x_div_w + chan, 0)));
ac_build_gep0(&ctx->ac, es_vtxptr.t, es_vtxptr.v, LLVMConstInt(ctx->ac.i32, lds_pos_cull_x_div_w + chan, 0)));
}
break;
@@ -1111,7 +1111,7 @@ void gfx10_ngg_culling_build_end(struct si_shader_context *ctx)
/* Initialize the packed data. */
LLVMBuildStore(
builder, packed_data,
ac_build_gep0_2(&ctx->ac, es_vtxptr.t, es_vtxptr.v, LLVMConstInt(ctx->ac.i32, lds_packed_data, 0)));
ac_build_gep0(&ctx->ac, es_vtxptr.t, es_vtxptr.v, LLVMConstInt(ctx->ac.i32, lds_packed_data, 0)));
ac_build_endif(&ctx->ac, ctx->merged_wrap_if_label);
ac_build_waitcnt(&ctx->ac, AC_WAIT_LGKM);
@@ -1208,7 +1208,7 @@ void gfx10_ngg_culling_build_end(struct si_shader_context *ctx)
continue;
LLVMValueRef idx = LLVMConstInt(ctx->ac.i32, index, 0);
LLVMValueRef v = ac_build_gep0_2(&ctx->ac, gs_vtxptr[vtx].t, gs_vtxptr[vtx].v, idx);
LLVMValueRef v = ac_build_gep0(&ctx->ac, gs_vtxptr[vtx].t, gs_vtxptr[vtx].v, idx);
pos[vtx][chan] = LLVMBuildLoad2(builder, ac_build_gep0_type(gs_vtxptr[vtx].t, idx), v, "");
pos[vtx][chan] = ac_to_float(&ctx->ac, pos[vtx][chan]);
}
@@ -1305,7 +1305,7 @@ void gfx10_ngg_culling_build_end(struct si_shader_context *ctx)
LLVMBuildStore(
builder, ac_to_integer(&ctx->ac,
LLVMBuildLoad2(builder, ctx->ac.f32, addrs[4 * pos_index + chan], "")),
ac_build_gep0_2(&ctx->ac, new_vtx.t, new_vtx.v, LLVMConstInt(ctx->ac.i32, lds_pos_x + chan, 0)));
ac_build_gep0(&ctx->ac, new_vtx.t, new_vtx.v, LLVMConstInt(ctx->ac.i32, lds_pos_x + chan, 0)));
}
/* Store VertexID and InstanceID into LDS. ES threads will have to load them
@@ -1315,24 +1315,24 @@ void gfx10_ngg_culling_build_end(struct si_shader_context *ctx)
if (ctx->stage == MESA_SHADER_VERTEX) {
LLVMBuildStore(
builder, ctx->abi.vertex_id,
ac_build_gep0_2(&ctx->ac, new_vtx.t, new_vtx.v, LLVMConstInt(ctx->ac.i32, lds_vertex_id, 0)));
ac_build_gep0(&ctx->ac, new_vtx.t, new_vtx.v, LLVMConstInt(ctx->ac.i32, lds_vertex_id, 0)));
if (uses_instance_id) {
LLVMBuildStore(
builder, ctx->abi.instance_id,
ac_build_gep0_2(&ctx->ac, new_vtx.t, new_vtx.v, LLVMConstInt(ctx->ac.i32, lds_instance_id, 0)));
ac_build_gep0(&ctx->ac, new_vtx.t, new_vtx.v, LLVMConstInt(ctx->ac.i32, lds_instance_id, 0)));
}
} else {
assert(ctx->stage == MESA_SHADER_TESS_EVAL);
LLVMBuildStore(builder, ac_to_integer(&ctx->ac, ac_get_arg(&ctx->ac, ctx->args.tes_u)),
ac_build_gep0_2(&ctx->ac, new_vtx.t, new_vtx.v, LLVMConstInt(ctx->ac.i32, lds_tes_u, 0)));
ac_build_gep0(&ctx->ac, new_vtx.t, new_vtx.v, LLVMConstInt(ctx->ac.i32, lds_tes_u, 0)));
LLVMBuildStore(builder, ac_to_integer(&ctx->ac, ac_get_arg(&ctx->ac, ctx->args.tes_v)),
ac_build_gep0_2(&ctx->ac, new_vtx.t, new_vtx.v, LLVMConstInt(ctx->ac.i32, lds_tes_v, 0)));
ac_build_gep0(&ctx->ac, new_vtx.t, new_vtx.v, LLVMConstInt(ctx->ac.i32, lds_tes_v, 0)));
LLVMBuildStore(builder, LLVMBuildTrunc(builder, ac_get_arg(&ctx->ac, ctx->args.tes_rel_patch_id), ctx->ac.i8, ""),
si_build_gep_i8(ctx, new_vtx.value, lds_byte2_tes_rel_patch_id));
if (uses_tes_prim_id) {
LLVMBuildStore(
builder, ac_get_arg(&ctx->ac, ctx->args.tes_patch_id),
ac_build_gep0_2(&ctx->ac, new_vtx.t, new_vtx.v, LLVMConstInt(ctx->ac.i32, lds_tes_patch_id, 0)));
ac_build_gep0(&ctx->ac, new_vtx.t, new_vtx.v, LLVMConstInt(ctx->ac.i32, lds_tes_patch_id, 0)));
}
}
}
@@ -1411,13 +1411,13 @@ void gfx10_ngg_culling_build_end(struct si_shader_context *ctx)
/* Prepare LDS addresses of the new ES input VGPRs. */
LLVMValueRef input_vgpr_addresses[4] = {
ac_build_gep0_2(&ctx->ac, es_vtxptr.t, es_vtxptr.v, LLVMConstInt(ctx->ac.i32, lds_vertex_id, 0)),
ac_build_gep0_2(&ctx->ac, es_vtxptr.t, es_vtxptr.v, LLVMConstInt(ctx->ac.i32, lds_instance_id, 0)),
ac_build_gep0(&ctx->ac, es_vtxptr.t, es_vtxptr.v, LLVMConstInt(ctx->ac.i32, lds_vertex_id, 0)),
ac_build_gep0(&ctx->ac, es_vtxptr.t, es_vtxptr.v, LLVMConstInt(ctx->ac.i32, lds_instance_id, 0)),
};
if (ctx->stage == MESA_SHADER_TESS_EVAL) {
input_vgpr_addresses[2] = si_build_gep_i8(ctx, es_vtxptr.v, lds_byte2_tes_rel_patch_id);
if (uses_tes_prim_id) {
input_vgpr_addresses[3] = ac_build_gep0_2(&ctx->ac, es_vtxptr.t, es_vtxptr.v,
input_vgpr_addresses[3] = ac_build_gep0(&ctx->ac, es_vtxptr.t, es_vtxptr.v,
LLVMConstInt(ctx->ac.i32, lds_tes_patch_id, 0));
}
}
@@ -1545,7 +1545,7 @@ void gfx10_ngg_build_end(struct si_shader_context *ctx)
*/
if (ctx->so.num_outputs) {
LLVMValueRef idx = LLVMConstInt(ctx->ac.i32, 4 * i + j, false);
tmp = ac_build_gep0_2(&ctx->ac, vertex_ptr.t, vertex_ptr.v, idx);
tmp = ac_build_gep0(&ctx->ac, vertex_ptr.t, vertex_ptr.v, idx);
tmp2 = LLVMBuildLoad2(builder, ac_build_gep0_type(vertex_ptr.t, idx), addrs[4 * i + j], "");
LLVMTypeRef type = ac_to_integer_type(&ctx->ac, ctx->ac.f32);
tmp2 = LLVMBuildBitCast(ctx->ac.builder, tmp2, type, "");
@@ -1561,7 +1561,7 @@ void gfx10_ngg_build_end(struct si_shader_context *ctx)
edgeflag = ac_build_umin(&ctx->ac, edgeflag, ctx->ac.i32_1);
tmp = LLVMConstInt(ctx->ac.i32, ngg_nogs_vertex_size(ctx->shader) - 1, 0);
tmp = ac_build_gep0_2(&ctx->ac, vertex_ptr.t, vertex_ptr.v, tmp);
tmp = ac_build_gep0(&ctx->ac, vertex_ptr.t, vertex_ptr.v, tmp);
LLVMBuildStore(builder, edgeflag, tmp);
}
}
@@ -1624,7 +1624,7 @@ void gfx10_ngg_build_end(struct si_shader_context *ctx)
struct ac_llvm_pointer vt = ngg_nogs_vertex_ptr(ctx, vtxindex[i]);
tmp2 = LLVMConstInt(ctx->ac.i32, ngg_nogs_vertex_size(ctx->shader) - 1, 0);
tmp = LLVMBuildLoad2(builder, ac_build_gep0_type(vt.t, tmp2),
ac_build_gep0_2(&ctx->ac, vt.t, vt.v, tmp2), "");
ac_build_gep0(&ctx->ac, vt.t, vt.v, tmp2), "");
tmp = LLVMBuildTrunc(builder, tmp, ctx->ac.i1, "");
user_edgeflags[i] = ac_build_alloca_init(&ctx->ac, tmp, "");
@@ -1655,7 +1655,7 @@ void gfx10_ngg_build_end(struct si_shader_context *ctx)
struct ac_llvm_pointer vertex_ptr = ngg_nogs_vertex_ptr(ctx, provoking_vtx_index);
LLVMBuildStore(builder, ac_get_arg(&ctx->ac, ctx->args.gs_prim_id),
ac_build_gep0_2(&ctx->ac, vertex_ptr.t, vertex_ptr.v, ctx->ac.i32_0));
ac_build_gep0(&ctx->ac, vertex_ptr.t, vertex_ptr.v, ctx->ac.i32_0));
ac_build_endif(&ctx->ac, 5400);
}
@@ -1719,7 +1719,7 @@ void gfx10_ngg_build_end(struct si_shader_context *ctx)
for (unsigned j = 0; j < 4; j++) {
tmp = LLVMConstInt(ctx->ac.i32, lds_pos_x + j, 0);
LLVMValueRef v = ac_build_gep0_2(&ctx->ac, vertex_ptr.t, vertex_ptr.v, tmp);
LLVMValueRef v = ac_build_gep0(&ctx->ac, vertex_ptr.t, vertex_ptr.v, tmp);
tmp = LLVMBuildLoad2(builder, ac_build_gep0_type(vertex_ptr.t, tmp), v, "");
outputs[i].values[j] = LLVMBuildBitCast(ctx->ac.builder, tmp,
ac_to_float_type(&ctx->ac, ctx->ac.f32), "");
@@ -1742,7 +1742,7 @@ void gfx10_ngg_build_end(struct si_shader_context *ctx)
struct ac_llvm_pointer vt = ngg_nogs_vertex_ptr(ctx, gfx10_get_thread_id_in_tg(ctx));
outputs[i].values[0] = LLVMBuildLoad2(
builder, ac_build_gep0_type(vt.t, ctx->ac.i32_0), ac_build_gep0_2(&ctx->ac, vt.t, vt.v, ctx->ac.i32_0), "");
builder, ac_build_gep0_type(vt.t, ctx->ac.i32_0), ac_build_gep0(&ctx->ac, vt.t, vt.v, ctx->ac.i32_0), "");
} else {
assert(ctx->stage == MESA_SHADER_TESS_EVAL);
outputs[i].values[0] = si_get_primitive_id(ctx, 0);
@@ -1821,7 +1821,7 @@ static struct ac_llvm_pointer ngg_gs_vertex_ptr(struct si_shader_context *ctx, L
}
return (struct ac_llvm_pointer) {
.value = ac_build_gep0_2(&ctx->ac, storage.t, storage.v, vertexidx),
.value = ac_build_gep0(&ctx->ac, storage.t, storage.v, vertexidx),
.pointee_type = ac_build_gep0_type(storage.t, vertexidx)
};
}
@@ -1948,7 +1948,7 @@ void gfx10_ngg_gs_emit_begin(struct si_shader_context *ctx)
tmp = LLVMBuildICmp(builder, LLVMIntULT, tid, LLVMConstInt(ctx->ac.i32, 4, false), "");
ac_build_ifcc(&ctx->ac, tmp, 5090);
{
LLVMValueRef ptr = ac_build_gep0_2(&ctx->ac, scratchptr.t, scratchptr.v, tid);
LLVMValueRef ptr = ac_build_gep0(&ctx->ac, scratchptr.t, scratchptr.v, tid);
LLVMBuildStore(builder, ctx->ac.i32_0, ptr);
}
ac_build_endif(&ctx->ac, 5090);
@@ -2033,7 +2033,7 @@ void gfx10_ngg_gs_build_end(struct si_shader_context *ctx)
{
LLVMBuildAtomicRMW(
builder, LLVMAtomicRMWBinOpAdd,
ac_build_gep0_2(&ctx->ac, ctx->gs_ngg_scratch.t, ctx->gs_ngg_scratch.v, LLVMConstInt(ctx->ac.i32, stream, false)),
ac_build_gep0(&ctx->ac, ctx->gs_ngg_scratch.t, ctx->gs_ngg_scratch.v, LLVMConstInt(ctx->ac.i32, stream, false)),
numprims, LLVMAtomicOrderingMonotonic, false);
}
ac_build_endif(&ctx->ac, 5105);
@@ -2069,7 +2069,7 @@ void gfx10_ngg_gs_build_end(struct si_shader_context *ctx)
"");
struct ac_llvm_pointer vt = ngg_gs_vertex_ptr(ctx, tmp);
nggso.vertices[i].t = ac_build_gep0_type(vt.t, ctx->ac.i32_0);
nggso.vertices[i].v = ac_build_gep0_2(&ctx->ac, vt.t, vt.v, ctx->ac.i32_0);
nggso.vertices[i].v = ac_build_gep0(&ctx->ac, vt.t, vt.v, ctx->ac.i32_0);
}
build_streamout(ctx, &nggso);
@@ -2097,7 +2097,7 @@ void gfx10_ngg_gs_build_end(struct si_shader_context *ctx)
}
tmp = LLVMBuildLoad2(builder, ctx->ac.i32,
ac_build_gep0_2(&ctx->ac, ctx->gs_ngg_scratch.t, ctx->gs_ngg_scratch.v,
ac_build_gep0(&ctx->ac, ctx->gs_ngg_scratch.t, ctx->gs_ngg_scratch.v,
tid), "");
LLVMValueRef args[] = {
tmp, ngg_get_query_buf(ctx),
@@ -2136,7 +2136,7 @@ void gfx10_ngg_gs_build_end(struct si_shader_context *ctx)
tmp = LLVMBuildSub(builder, tid, LLVMConstInt(ctx->ac.i32, verts_per_prim - i - 1, false), "");
struct ac_llvm_pointer vt = ngg_gs_vertex_ptr(ctx, tmp);
vtxptr[i].t = ac_build_gep0_type(vt.t, ctx->ac.i32_0);
vtxptr[i].v = ac_build_gep0_2(&ctx->ac, vt.t, vt.v, ctx->ac.i32_0);
vtxptr[i].v = ac_build_gep0(&ctx->ac, vt.t, vt.v, ctx->ac.i32_0);
}
for (unsigned i = 0; i < info->num_outputs; i++) {
@@ -2157,7 +2157,7 @@ void gfx10_ngg_gs_build_end(struct si_shader_context *ctx)
continue;
LLVMValueRef idx = LLVMConstInt(ctx->ac.i32, 4 * i + comp, false);
tmp = ac_build_gep0_2(&ctx->ac, vtxptr[vert].t, vtxptr[vert].v, idx);
tmp = ac_build_gep0(&ctx->ac, vtxptr[vert].t, vtxptr[vert].v, idx);
pos[vert][comp] = LLVMBuildLoad2(builder,
ac_build_gep0_type(vtxptr[vert].t, idx),
tmp, "");
@@ -2229,7 +2229,7 @@ void gfx10_ngg_gs_build_end(struct si_shader_context *ctx)
vertlive_scan.enable_reduce = true;
vertlive_scan.enable_exclusive = true;
vertlive_scan.src = vertlive;
vertlive_scan.scratch = ac_build_gep0_2(&ctx->ac, ctx->gs_ngg_scratch.t, ctx->gs_ngg_scratch.v, ctx->ac.i32_0);
vertlive_scan.scratch = ac_build_gep0(&ctx->ac, ctx->gs_ngg_scratch.t, ctx->gs_ngg_scratch.v, ctx->ac.i32_0);
vertlive_scan.waveidx = get_wave_id_in_tg(ctx);
vertlive_scan.numwaves = get_tgsize(ctx);
vertlive_scan.maxwaves = DIV_ROUND_UP(256, ctx->ac.wave_size);

12
src/gallium/drivers/radeonsi/si_shader_llvm.c
View File

@@ -767,7 +767,7 @@ static LLVMValueRef si_llvm_load_intrinsic(struct ac_shader_abi *abi, nir_intrin
LLVMValueRef slot = LLVMConstInt(ctx->ac.i32, SI_HS_CONST_DEFAULT_TESS_LEVELS, 0);
LLVMValueRef buf = ac_get_arg(&ctx->ac, ctx->internal_bindings);
LLVMTypeRef type = ac_get_arg_pointee_type(&ctx->ac, &ctx->args, ctx->internal_bindings);
buf = ac_build_load_to_sgpr2(&ctx->ac, type, buf, slot);
buf = ac_build_load_to_sgpr(&ctx->ac, type, buf, slot);
int offset = op == nir_intrinsic_load_tess_level_inner_default ? 4 : 0;
LLVMValueRef val[4];
@@ -816,7 +816,7 @@ static LLVMValueRef si_llvm_load_intrinsic(struct ac_shader_abi *abi, nir_intrin
case nir_intrinsic_load_clip_half_line_width_amd: {
LLVMValueRef ptr = ac_get_arg(&ctx->ac, ctx->small_prim_cull_info);
return ac_build_load_to_sgpr2(&ctx->ac, ctx->ac.v2f32, ptr, LLVMConstInt(ctx->ac.i32, 4, 0));
return ac_build_load_to_sgpr(&ctx->ac, ctx->ac.v2f32, ptr, LLVMConstInt(ctx->ac.i32, 4, 0));
}
case nir_intrinsic_load_viewport_xy_scale_and_offset: {
@@ -824,7 +824,7 @@ static LLVMValueRef si_llvm_load_intrinsic(struct ac_shader_abi *abi, nir_intrin
LLVMValueRef ptr = ac_get_arg(&ctx->ac, ctx->small_prim_cull_info);
LLVMTypeRef type = ac_get_arg_pointee_type(&ctx->ac, &ctx->args, ctx->small_prim_cull_info);
LLVMValueRef terms =
ac_build_load_to_sgpr2(&ctx->ac, type, ptr, prim_is_lines ? ctx->ac.i32_1 : ctx->ac.i32_0);
ac_build_load_to_sgpr(&ctx->ac, type, ptr, prim_is_lines ? ctx->ac.i32_1 : ctx->ac.i32_0);
return LLVMBuildBitCast(ctx->ac.builder, terms, ctx->ac.v4f32, "");
}
@@ -881,7 +881,7 @@ static LLVMValueRef si_llvm_load_user_clip_plane(struct ac_shader_abi *abi, unsi
LLVMValueRef ptr = ac_get_arg(&ctx->ac, ctx->internal_bindings);
LLVMTypeRef type = ac_get_arg_pointee_type(&ctx->ac, &ctx->args, ctx->internal_bindings);
LLVMValueRef constbuf_index = LLVMConstInt(ctx->ac.i32, SI_VS_CONST_CLIP_PLANES, 0);
LLVMValueRef const_resource = ac_build_load_to_sgpr2(&ctx->ac, type, ptr, constbuf_index);
LLVMValueRef const_resource = ac_build_load_to_sgpr(&ctx->ac, type, ptr, constbuf_index);
LLVMValueRef addr = LLVMConstInt(ctx->ac.i32, ucp_id * 16, 0);
return ac_build_buffer_load(&ctx->ac, const_resource, 4, NULL, addr, NULL,
ctx->ac.f32, 0, true, true);
@@ -893,7 +893,7 @@ static LLVMValueRef si_llvm_load_streamout_buffer(struct ac_shader_abi *abi, uns
LLVMValueRef buf_ptr = ac_get_arg(&ctx->ac, ctx->internal_bindings);
LLVMTypeRef type = ac_get_arg_pointee_type(&ctx->ac, &ctx->args, ctx->internal_bindings);
return ac_build_load_to_sgpr2(
return ac_build_load_to_sgpr(
&ctx->ac, type, buf_ptr, LLVMConstInt(ctx->ac.i32, SI_VS_STREAMOUT_BUF0 + buffer, false));
}
@@ -933,7 +933,7 @@ bool si_llvm_translate_nir(struct si_shader_context *ctx, struct si_shader *shad
LLVMValueRef buf = ac_get_arg(&ctx->ac, ctx->internal_bindings);
LLVMTypeRef type = ac_get_arg_pointee_type(&ctx->ac, &ctx->args, ctx->internal_bindings);
ctx->instance_divisor_constbuf =
ac_build_load_to_sgpr2(
ac_build_load_to_sgpr(
&ctx->ac, type, buf, LLVMConstInt(ctx->ac.i32, SI_VS_CONST_INSTANCE_DIVISORS, 0));
}
break;

10
src/gallium/drivers/radeonsi/si_shader_llvm_gs.c
View File

@@ -101,8 +101,8 @@ static LLVMValueRef ngg_get_emulated_counters_buf(struct si_shader_context *ctx)
LLVMValueRef buf_ptr = ac_get_arg(&ctx->ac, ctx->internal_bindings);
LLVMTypeRef type = ac_get_arg_pointee_type(&ctx->ac, &ctx->args, ctx->internal_bindings);
return ac_build_load_to_sgpr2(&ctx->ac, type, buf_ptr,
LLVMConstInt(ctx->ac.i32, SI_GS_QUERY_EMULATED_COUNTERS_BUF, false));
return ac_build_load_to_sgpr(&ctx->ac, type, buf_ptr,
LLVMConstInt(ctx->ac.i32, SI_GS_QUERY_EMULATED_COUNTERS_BUF, false));
}
void si_llvm_gs_build_end(struct si_shader_context *ctx)
@@ -266,7 +266,7 @@ void si_preload_esgs_ring(struct si_shader_context *ctx)
LLVMValueRef buf_ptr = ac_get_arg(&ctx->ac, ctx->internal_bindings);
LLVMTypeRef type = ac_get_arg_pointee_type(&ctx->ac, &ctx->args, ctx->internal_bindings);
ctx->esgs_ring = ac_build_load_to_sgpr2(&ctx->ac, type, buf_ptr, offset);
ctx->esgs_ring = ac_build_load_to_sgpr(&ctx->ac, type, buf_ptr, offset);
if (ctx->stage != MESA_SHADER_GEOMETRY) {
LLVMValueRef desc1 = LLVMBuildExtractElement(builder, ctx->esgs_ring, ctx->ac.i32_1, "");
@@ -309,7 +309,7 @@ void si_preload_gs_rings(struct si_shader_context *ctx)
LLVMValueRef offset = LLVMConstInt(ctx->ac.i32, SI_RING_GSVS, 0);
LLVMValueRef buf_ptr = ac_get_arg(&ctx->ac, ctx->internal_bindings);
LLVMTypeRef type = ac_get_arg_pointee_type(&ctx->ac, &ctx->args, ctx->internal_bindings);
LLVMValueRef base_ring = ac_build_load_to_sgpr2(&ctx->ac, type, buf_ptr, offset);
LLVMValueRef base_ring = ac_build_load_to_sgpr(&ctx->ac, type, buf_ptr, offset);
/* The conceptual layout of the GSVS ring is
* v0c0 .. vLv0 v0c1 .. vLc1 ..
@@ -438,7 +438,7 @@ struct si_shader *si_generate_gs_copy_shader(struct si_screen *sscreen,
LLVMValueRef buf_ptr = ac_get_arg(&ctx.ac, ctx.internal_bindings);
LLVMTypeRef type = ac_get_arg_pointee_type(&ctx.ac, &ctx.args, ctx.internal_bindings);
ctx.gsvs_ring[0] =
ac_build_load_to_sgpr2(&ctx.ac, type, buf_ptr, LLVMConstInt(ctx.ac.i32, SI_RING_GSVS, 0));
ac_build_load_to_sgpr(&ctx.ac, type, buf_ptr, LLVMConstInt(ctx.ac.i32, SI_RING_GSVS, 0));
LLVMValueRef voffset =
LLVMBuildMul(ctx.ac.builder, ctx.abi.vertex_id, LLVMConstInt(ctx.ac.i32, 4, 0), "");

8
src/gallium/drivers/radeonsi/si_shader_llvm_ps.c
View File

@@ -37,7 +37,7 @@ static LLVMValueRef load_sample_position(struct ac_shader_abi *abi, LLVMValueRef
LLVMValueRef desc = ac_get_arg(&ctx->ac, ctx->internal_bindings);
LLVMValueRef buf_index = LLVMConstInt(ctx->ac.i32, SI_PS_CONST_SAMPLE_POSITIONS, 0);
LLVMTypeRef type = ac_get_arg_pointee_type(&ctx->ac, &ctx->args, ctx->internal_bindings);
LLVMValueRef resource = ac_build_load_to_sgpr2(&ctx->ac, type, desc, buf_index);
LLVMValueRef resource = ac_build_load_to_sgpr(&ctx->ac, type, desc, buf_index);
/* offset = sample_id * 8 (8 = 2 floats containing samplepos.xy) */
LLVMValueRef offset0 =
@@ -70,7 +70,7 @@ static LLVMValueRef si_nir_emit_fbfetch(struct ac_shader_abi *abi)
ptr =
LLVMBuildPointerCast(ctx->ac.builder, ptr, ac_array_in_const32_addr_space(ctx->ac.v8i32), "");
image =
ac_build_load_to_sgpr2(&ctx->ac, ctx->ac.v8i32, ptr, LLVMConstInt(ctx->ac.i32, SI_PS_IMAGE_COLORBUF0 / 2, 0));
ac_build_load_to_sgpr(&ctx->ac, ctx->ac.v8i32, ptr, LLVMConstInt(ctx->ac.i32, SI_PS_IMAGE_COLORBUF0 / 2, 0));
unsigned chan = 0;
@@ -90,7 +90,7 @@ static LLVMValueRef si_nir_emit_fbfetch(struct ac_shader_abi *abi)
ctx->shader->key.ps.mono.fbfetch_msaa &&
!(ctx->screen->debug_flags & DBG(NO_FMASK))) {
fmask = ac_build_load_to_sgpr2(&ctx->ac, ctx->ac.v8i32, ptr,
fmask = ac_build_load_to_sgpr(&ctx->ac, ctx->ac.v8i32, ptr,
LLVMConstInt(ctx->ac.i32, SI_PS_IMAGE_COLORBUF0_FMASK / 2, 0));
ac_apply_fmask_to_sample(&ctx->ac, fmask, args.coords,
@@ -550,7 +550,7 @@ static void si_llvm_emit_polygon_stipple(struct si_shader_context *ctx,
/* Load the buffer descriptor. */
slot = LLVMConstInt(ctx->ac.i32, SI_PS_CONST_POLY_STIPPLE, 0);
desc = ac_build_load_to_sgpr2(&ctx->ac, ctx->ac.v4i32, param_internal_bindings, slot);
desc = ac_build_load_to_sgpr(&ctx->ac, ctx->ac.v4i32, param_internal_bindings, slot);
/* The stipple pattern is 32x32, each row has 32 bits. */
offset = LLVMBuildMul(builder, address[1], LLVMConstInt(ctx->ac.i32, 4, 0), "");

20
src/gallium/drivers/radeonsi/si_shader_llvm_resources.c
View File

@@ -104,10 +104,10 @@ static LLVMValueRef load_ubo(struct ac_shader_abi *abi, LLVMValueRef index)
index =
LLVMBuildAdd(ctx->ac.builder, index, LLVMConstInt(ctx->ac.i32, SI_NUM_SHADER_BUFFERS, 0), "");
return ac_build_load_to_sgpr2(&ctx->ac,
ac_get_arg_pointee_type(&ctx->ac, &ctx->args, ctx->const_and_shader_buffers),
ptr,
index);
return ac_build_load_to_sgpr(&ctx->ac,
ac_get_arg_pointee_type(&ctx->ac, &ctx->args, ctx->const_and_shader_buffers),
ptr,
index);
}
static LLVMValueRef load_ssbo(struct ac_shader_abi *abi, LLVMValueRef index, bool write, bool non_uniform)
@@ -124,9 +124,9 @@ static LLVMValueRef load_ssbo(struct ac_shader_abi *abi, LLVMValueRef index, boo
index = LLVMBuildSub(ctx->ac.builder, LLVMConstInt(ctx->ac.i32, SI_NUM_SHADER_BUFFERS - 1, 0),
index, "");
return ac_build_load_to_sgpr2(&ctx->ac,
ac_get_arg_pointee_type(&ctx->ac, &ctx->args, ctx->const_and_shader_buffers),
rsrc_ptr, index);
return ac_build_load_to_sgpr(&ctx->ac,
ac_get_arg_pointee_type(&ctx->ac, &ctx->args, ctx->const_and_shader_buffers),
rsrc_ptr, index);
}
/**
@@ -197,9 +197,9 @@ static LLVMValueRef si_load_image_desc(struct si_shader_context *ctx, LLVMTypeRe
}
if (bindless)
rsrc = ac_build_load_to_sgpr_uint_wraparound2(&ctx->ac, type, list, index);
rsrc = ac_build_load_to_sgpr_uint_wraparound(&ctx->ac, type, list, index);
else
rsrc = ac_build_load_to_sgpr2(&ctx->ac, type, list, index);
rsrc = ac_build_load_to_sgpr(&ctx->ac, type, list, index);
if (desc_type == AC_DESC_IMAGE)
rsrc = fixup_image_desc(ctx, rsrc, uses_store);
@@ -248,7 +248,7 @@ static LLVMValueRef si_load_sampler_desc(struct si_shader_context *ctx, LLVMType
unreachable("Plane descriptor requested in radeonsi.");
}
return ac_build_load_to_sgpr2(&ctx->ac, list_type, list, index);
return ac_build_load_to_sgpr(&ctx->ac, list_type, list, index);
}
static LLVMValueRef si_nir_load_sampler_desc(struct ac_shader_abi *abi, unsigned descriptor_set,

14
src/gallium/drivers/radeonsi/si_shader_llvm_vs.c
View File

@@ -169,8 +169,8 @@ static void load_input_vs(struct si_shader_context *ctx, unsigned input_index, L
} else {
unsigned index = input_index - num_vbos_in_user_sgprs;
LLVMTypeRef type = ac_get_arg_pointee_type(&ctx->ac, &ctx->args, ctx->args.vertex_buffers);
vb_desc = ac_build_load_to_sgpr2(&ctx->ac, type, ac_get_arg(&ctx->ac, ctx->args.vertex_buffers),
LLVMConstInt(ctx->ac.i32, index, 0));
vb_desc = ac_build_load_to_sgpr(&ctx->ac, type, ac_get_arg(&ctx->ac, ctx->args.vertex_buffers),
LLVMConstInt(ctx->ac.i32, index, 0));
}
if (ctx->abi.vertex_id_replaced) {
@@ -418,7 +418,7 @@ void si_llvm_emit_streamout(struct si_shader_context *ctx, struct si_shader_outp
LLVMValueRef offset = LLVMConstInt(ctx->ac.i32, SI_VS_STREAMOUT_BUF0 + i, 0);
so_buffers[i] = ac_build_load_to_sgpr2(&ctx->ac, type, buf_ptr, offset);
so_buffers[i] = ac_build_load_to_sgpr(&ctx->ac, type, buf_ptr, offset);
LLVMValueRef so_offset = ac_get_arg(&ctx->ac, ctx->args.streamout_offset[i]);
so_offset = LLVMBuildMul(builder, so_offset, LLVMConstInt(ctx->ac.i32, 4, 0), "");
@@ -454,7 +454,7 @@ void si_llvm_clipvertex_to_clipdist(struct si_shader_context *ctx,
LLVMValueRef ptr = ac_get_arg(&ctx->ac, ctx->internal_bindings);
LLVMTypeRef type = ac_get_arg_pointee_type(&ctx->ac, &ctx->args, ctx->internal_bindings);
LLVMValueRef constbuf_index = LLVMConstInt(ctx->ac.i32, SI_VS_CONST_CLIP_PLANES, 0);
LLVMValueRef const_resource = ac_build_load_to_sgpr2(&ctx->ac, type, ptr, constbuf_index);
LLVMValueRef const_resource = ac_build_load_to_sgpr(&ctx->ac, type, ptr, constbuf_index);
unsigned clipdist_mask = ctx->shader->selector->info.clipdist_mask &
~ctx->shader->key.ge.opt.kill_clip_distances;
@@ -819,8 +819,8 @@ void si_llvm_build_vs_exports(struct si_shader_context *ctx, LLVMValueRef num_ex
LLVMBuildPointerCast(ctx->ac.builder,
ac_get_arg(&ctx->ac, ctx->internal_bindings),
LLVMPointerType(ctx->ac.i32, AC_ADDR_SPACE_CONST_32BIT), "");
attr_address = ac_build_load_to_sgpr2(&ctx->ac, ctx->ac.i32, ptr,
LLVMConstInt(ctx->ac.i32, SI_GS_ATTRIBUTE_RING * 4, 0));
attr_address = ac_build_load_to_sgpr(&ctx->ac, ctx->ac.i32, ptr,
LLVMConstInt(ctx->ac.i32, SI_GS_ATTRIBUTE_RING * 4, 0));
} else {
attr_address = ac_get_arg(&ctx->ac, ctx->gs_attr_address);
}
@@ -1039,7 +1039,7 @@ void si_llvm_build_vs_prolog(struct si_shader_context *ctx, union si_shader_part
if (key->vs_prolog.states.instance_divisor_is_fetched) {
LLVMValueRef list = si_prolog_get_internal_bindings(ctx);
LLVMValueRef buf_index = LLVMConstInt(ctx->ac.i32, SI_VS_CONST_INSTANCE_DIVISORS, 0);
instance_divisor_constbuf = ac_build_load_to_sgpr2(&ctx->ac, ctx->ac.v4i32, list, buf_index);
instance_divisor_constbuf = ac_build_load_to_sgpr(&ctx->ac, ctx->ac.v4i32, list, buf_index);
}
for (i = 0; i < key->vs_prolog.num_inputs; i++) {