<div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">On Thu, Mar 8, 2018 at 8:07 AM, Neil Roberts <<a href="mailto:nroberts@igalia.com" target="_blank">nroberts@igalia.com</a>> wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">The implementation is inspired by lower_instructions_visitor::dfrexp_sig_to_arith. This has been tested against the arb_gpu_shader_fp64/fs-frexp-dvec4 test using the ARB_gl_spirv branch. --- Please also see this related patch which I probably should have bundled in this series: <a href="https://patchwork.freedesktop.org/patch/208702/" rel="noreferrer" target="_blank">https://patchwork.freedesktop.org/patch/208702/</a> src/compiler/spirv/vtn_glsl450.c | 60 +++++++++++++++++++++++++++++++++++++--- 1 file changed, 56 insertions(+), 4 deletions(-) diff --git a/src/compiler/spirv/vtn_glsl450.c b/src/compiler/spirv/vtn_glsl450.c index 46ef40f5e3f..51c4cd271bc 100644 --- a/src/compiler/spirv/vtn_glsl450.c +++ b/src/compiler/spirv/vtn_glsl450.c @@ -380,7 +380,7 @@ build_atan2(nir_builder *b, nir_ssa_def *y, nir_ssa_def *x) } static nir_ssa_def * -build_frexp(nir_builder *b, nir_ssa_def *x, nir_ssa_def **exponent) +build_frexp32(nir_builder *b, nir_ssa_def *x, nir_ssa_def **exponent) { nir_ssa_def *abs_x = nir_fabs(b, x); nir_ssa_def *zero = nir_imm_float(b, 0.0f); @@ -412,6 +412,51 @@ build_frexp(nir_builder *b, nir_ssa_def *x, nir_ssa_def **exponent) nir_bcsel(b, is_not_zero, exponent_value, zero)); } +static nir_ssa_def * +build_frexp64(nir_builder *b, nir_ssa_def *x, nir_ssa_def **exponent) +{ + nir_ssa_def *abs_x = nir_fabs(b, x); + nir_ssa_def *zero = nir_imm_double(b, 0.0); + nir_ssa_def *zero32 = nir_imm_float(b, 0.0f); + + /* Double-precision floating-point values are stored as + * 1 sign bit; + * 11 exponent bits; + * 52 mantissa bits. + * + * We only need to deal with the exponent so first we extract the upper 32 + * bits using nir_unpack_64_2x32_split_y. + */ + nir_ssa_def *upper_x = nir_unpack_64_2x32_split_y(b, x); </blockquote><div> </div><div>It'd be nice to have a lower_x as well </div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> + nir_ssa_def *abs_upper_x = nir_unpack_64_2x32_split_y(b, abs_x); + + /* An exponent shift of 20 will shift the remaining mantissa bits out, + * leaving only the exponent and sign bit (which itself may be zero, if the + * absolute value was taken before the bitcast and shift. + */ + nir_ssa_def *exponent_shift = nir_imm_int(b, 20); + nir_ssa_def *exponent_bias = nir_imm_int(b, -1022); + + nir_ssa_def *sign_mantissa_mask = nir_imm_int(b, 0x800fffffu); + + /* Exponent of floating-point values in the range [0.5, 1.0). */ + nir_ssa_def *exponent_value = nir_imm_int(b, 0x3fe00000u); + + nir_ssa_def *is_not_zero = nir_fne(b, abs_x, zero); + + *exponent = + nir_iadd(b, nir_ushr(b, abs_upper_x, exponent_shift), + nir_bcsel(b, is_not_zero, exponent_bias, zero32)); + + nir_ssa_def *new_upper = + nir_ior(b, nir_iand(b, upper_x, sign_mantissa_mask), + nir_bcsel(b, is_not_zero, exponent_value, zero32)); + + return nir_pack_64_2x32_split(b, + nir_unpack_64_2x32_split_x(b, x), </blockquote><div> </div><div>It would make this more clear. With that, </div><div>Reviewed-by: Jason Ekstrand <<a href="mailto:jason@jlekstrand.net">jason@jlekstrand.net</a>> </div><div>We'll need to add a 16-bit version as well but we have no way to test that right now. </div><div>--Jason </div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> + new_upper); +} + static nir_op vtn_nir_alu_op_for_spirv_glsl_opcode(struct vtn_builder *b, enum GLSLstd450 opcode) @@ -685,15 +730,22 @@ handle_glsl450_alu(struct vtn_builder *b, enum GLSLstd450 entrypoint, case GLSLstd450Frexp: { nir_ssa_def *exponent; - val->ssa->def = build_frexp(nb, src[0], &exponent); + if (src[0]->bit_size == 64) + val->ssa->def = build_frexp64(nb, src[0], &exponent); + else + val->ssa->def = build_frexp32(nb, src[0], &exponent); nir_store_deref_var(nb, vtn_nir_deref(b, w[6]), exponent, 0xf); return; } case GLSLstd450FrexpStruct: { vtn_assert(glsl_type_is_struct(val->ssa->type)); - val->ssa->elems[0]->def = build_frexp(nb, src[0], - &val->ssa->elems[1]->def); + if (src[0]->bit_size == 64) + val->ssa->elems[0]->def = build_frexp64(nb, src[0], + &val->ssa->elems[1]->def); + else + val->ssa->elems[0]->def = build_frexp32(nb, src[0], + &val->ssa->elems[1]->def); return; } -- 2.14.3 _______________________________________________ mesa-dev mailing list <a href="mailto:mesa-dev@lists.freedesktop.org">mesa-dev@lists.freedesktop.org</a> <a href="https://lists.freedesktop.org/mailman/listinfo/mesa-dev" rel="noreferrer" target="_blank">https://lists.freedesktop.org/mailman/listinfo/mesa-dev</a> </blockquote></div> </div></div>