The code to handle mat multipication by a scalar tries to pick either imul or fmul depending on whether the matrix is float or integer. However it was doing this by checking whether the base type is float. This was making it choose the int path for doubles (and presumably float16s).
--- This was discovered from running the arb_gpu_shader_fp64 tests through the GL_ARB_gl_spirv branch. For example: arb_gpu_shader_fp64@execution@built-in-functions@vs-op-div-dmat4-double. Note however these tests are hidden behind a glslang bug: https://github.com/KhronosGroup/glslang/issues/1278 The fix in that issue along with this patch makes them all pass. There’s also a little test case for this with VkRunner here: https://github.com/Igalia/vkrunner/blob/tests/examples/dmat-mul-scalar.shader_test src/compiler/spirv/vtn_alu.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/src/compiler/spirv/vtn_alu.c b/src/compiler/spirv/vtn_alu.c index d0c9e316935..110fcec2a60 100644 --- a/src/compiler/spirv/vtn_alu.c +++ b/src/compiler/spirv/vtn_alu.c @@ -142,10 +142,10 @@ mat_times_scalar(struct vtn_builder *b, { struct vtn_ssa_value *dest = vtn_create_ssa_value(b, mat->type); for (unsigned i = 0; i < glsl_get_matrix_columns(mat->type); i++) { - if (glsl_get_base_type(mat->type) == GLSL_TYPE_FLOAT) - dest->elems[i]->def = nir_fmul(&b->nb, mat->elems[i]->def, scalar); - else + if (glsl_base_type_is_integer(glsl_get_base_type(mat->type))) dest->elems[i]->def = nir_imul(&b->nb, mat->elems[i]->def, scalar); + else + dest->elems[i]->def = nir_fmul(&b->nb, mat->elems[i]->def, scalar); } return dest; -- 2.14.3 _______________________________________________ mesa-dev mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/mesa-dev
