summaryrefslogtreecommitdiffstats
path: root/src
diff options
context:
space:
mode:
authorMarek Olšák <[email protected]>2018-09-21 21:30:09 -0400
committerMarek Olšák <[email protected]>2018-10-06 21:50:09 -0400
commit77903c8cfbaa609d132ea5e5fbca062e0b2f0808 (patch)
tree0e56f3dcaf89f5cd1135cbf325fc3fb0d7222264 /src
parentfa023f293eb48d1d2e3c6efb89f4192c4a565fed (diff)
ac: add ac_build_round
Diffstat (limited to 'src')
-rw-r--r--src/amd/common/ac_llvm_build.c19
-rw-r--r--src/amd/common/ac_llvm_build.h1
-rw-r--r--src/amd/common/ac_nir_to_llvm.c2
-rw-r--r--src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c4
4 files changed, 20 insertions, 6 deletions
diff --git a/src/amd/common/ac_llvm_build.c b/src/amd/common/ac_llvm_build.c
index 81158940b8d..d11c12fcd13 100644
--- a/src/amd/common/ac_llvm_build.c
+++ b/src/amd/common/ac_llvm_build.c
@@ -558,6 +558,22 @@ LLVMValueRef ac_build_expand_to_vec4(struct ac_llvm_context *ctx,
return ac_build_gather_values(ctx, chan, 4);
}
+LLVMValueRef ac_build_round(struct ac_llvm_context *ctx, LLVMValueRef value)
+{
+ unsigned type_size = ac_get_type_size(LLVMTypeOf(value));
+ const char *name;
+
+ if (type_size == 2)
+ name = "llvm.rint.f16";
+ else if (type_size == 4)
+ name = "llvm.rint.f32";
+ else
+ name = "llvm.rint.f64";
+
+ return ac_build_intrinsic(ctx, name, LLVMTypeOf(value), &value, 1,
+ AC_FUNC_ATTR_READNONE);
+}
+
LLVMValueRef
ac_build_fdiv(struct ac_llvm_context *ctx,
LLVMValueRef num,
@@ -675,8 +691,7 @@ ac_prepare_cube_coords(struct ac_llvm_context *ctx,
LLVMValueRef invma;
if (is_array && !is_lod) {
- LLVMValueRef tmp = coords_arg[3];
- tmp = ac_build_intrinsic(ctx, "llvm.rint.f32", ctx->f32, &tmp, 1, 0);
+ LLVMValueRef tmp = ac_build_round(ctx, coords_arg[3]);
/* Section 8.9 (Texture Functions) of the GLSL 4.50 spec says:
*
diff --git a/src/amd/common/ac_llvm_build.h b/src/amd/common/ac_llvm_build.h
index 73f8a6b70cd..65235fb7352 100644
--- a/src/amd/common/ac_llvm_build.h
+++ b/src/amd/common/ac_llvm_build.h
@@ -175,6 +175,7 @@ ac_build_gather_values(struct ac_llvm_context *ctx,
LLVMValueRef ac_build_expand_to_vec4(struct ac_llvm_context *ctx,
LLVMValueRef value,
unsigned num_channels);
+LLVMValueRef ac_build_round(struct ac_llvm_context *ctx, LLVMValueRef value);
LLVMValueRef
ac_build_fdiv(struct ac_llvm_context *ctx,
diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
index 312383db36c..ffc64a79d95 100644
--- a/src/amd/common/ac_nir_to_llvm.c
+++ b/src/amd/common/ac_nir_to_llvm.c
@@ -3311,7 +3311,7 @@ static LLVMValueRef apply_round_slice(struct ac_llvm_context *ctx,
LLVMValueRef coord)
{
coord = ac_to_float(ctx, coord);
- coord = ac_build_intrinsic(ctx, "llvm.rint.f32", ctx->f32, &coord, 1, 0);
+ coord = ac_build_round(ctx, coord);
coord = ac_to_integer(ctx, coord);
return coord;
}
diff --git a/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c b/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c
index cabc448a082..8c44831bccb 100644
--- a/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c
+++ b/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c
@@ -1446,9 +1446,7 @@ static void build_tex_intrinsic(const struct lp_build_tgsi_action *action,
opcode != TGSI_OPCODE_TXF_LZ &&
ctx->screen->info.chip_class <= VI) {
unsigned array_coord = target == TGSI_TEXTURE_1D_ARRAY ? 1 : 2;
- args.coords[array_coord] =
- ac_build_intrinsic(&ctx->ac, "llvm.rint.f32", ctx->f32,
- &args.coords[array_coord], 1, 0);
+ args.coords[array_coord] = ac_build_round(&ctx->ac, args.coords[array_coord]);
}
/* 1D textures are allocated and used as 2D on GFX9. */