amd/common: round cube array slice in ac_prepare_cube_coords

The NIR-to-LLVM pass already does this; now the same fix covers
radeonsi as well.

Fixes various tests of
dEQP-GLES31.functional.texture.filtering.cube_array.combinations.*

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>

diff --git a/src/amd/common/ac_llvm_build.c b/src/amd/common/ac_llvm_build.c
index a831baed3f5c250e54b625e2fa08e898d1eb95a9..7193b80be59ab05064ace3f9a2f2fde973929379 100644 (file)
--- a/src/amd/common/ac_llvm_build.c
+++ b/src/amd/common/ac_llvm_build.c
@@ -484,7 +484,7 @@ static void build_cube_select(LLVMBuilderRef builder,
 
 void
 ac_prepare_cube_coords(struct ac_llvm_context *ctx,
-                      bool is_deriv, bool is_array,
+                      bool is_deriv, bool is_array, bool is_lod,
                       LLVMValueRef *coords_arg,
                       LLVMValueRef *derivs_arg)
 {
@@ -494,6 +494,11 @@ ac_prepare_cube_coords(struct ac_llvm_context *ctx,
        LLVMValueRef coords[3];
        LLVMValueRef invma;
 
+       if (is_array && !is_lod) {
+               coords_arg[3] = ac_build_intrinsic(ctx, "llvm.rint.f32", ctx->f32,
+                                                  &coords_arg[3], 1, 0);
+       }
+
        build_cube_intrinsic(ctx, coords_arg, &selcoords);
 
        invma = ac_build_intrinsic(ctx, "llvm.fabs.f32",

diff --git a/src/amd/common/ac_llvm_build.h b/src/amd/common/ac_llvm_build.h
index dae32e45752b258372d70186a42f7ae85e95ed0c..14ec03f5c84c6a641cb8e392ae87c451b103cadd 100644 (file)
--- a/src/amd/common/ac_llvm_build.h
+++ b/src/amd/common/ac_llvm_build.h
@@ -110,7 +110,7 @@ ac_build_fdiv(struct ac_llvm_context *ctx,
 
 void
 ac_prepare_cube_coords(struct ac_llvm_context *ctx,
-                      bool is_deriv, bool is_array,
+                      bool is_deriv, bool is_array, bool is_lod,
                       LLVMValueRef *coords_arg,
                       LLVMValueRef *derivs_arg);
 

diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
index 42398d1230b3f33283a93eba52ff305ff7358f0a..ba98cb2bde0939c9e92f2b630303e689da230441 100644 (file)
--- a/src/amd/common/ac_nir_to_llvm.c
+++ b/src/amd/common/ac_nir_to_llvm.c
@@ -4550,15 +4550,13 @@ static void visit_tex(struct ac_nir_context *ctx, nir_tex_instr *instr)
        }
 
        if (instr->sampler_dim == GLSL_SAMPLER_DIM_CUBE && coord) {
-               if (instr->is_array && instr->op != nir_texop_lod)
-                       coords[3] = apply_round_slice(&ctx->ac, coords[3]);
                for (chan = 0; chan < instr->coord_components; chan++)
                        coords[chan] = ac_to_float(&ctx->ac, coords[chan]);
                if (instr->coord_components == 3)
                        coords[3] = LLVMGetUndef(ctx->ac.f32);
                ac_prepare_cube_coords(&ctx->ac,
                        instr->op == nir_texop_txd, instr->is_array,
-                       coords, derivs);
+                       instr->op == nir_texop_lod, coords, derivs);
                if (num_deriv_comp)
                        num_deriv_comp--;
        }

diff --git a/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c b/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c
index a37fe586d6f8bc4202b7f2566f4232afd51fbc2c..d0969a7dd024a9caedc4c421efb63c9ba071797f 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c
+++ b/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c
@@ -1483,6 +1483,7 @@ static void tex_fetch_args(
                                       opcode == TGSI_OPCODE_TXD,
                                       target == TGSI_TEXTURE_CUBE_ARRAY ||
                                       target == TGSI_TEXTURE_SHADOWCUBE_ARRAY,
+                                      opcode == TGSI_OPCODE_LODQ,
                                       coords, derivs);
 
        if (opcode == TGSI_OPCODE_TXD)