gallivm: fix SCALED -> NORM conversions
Such conversions (which are most likely rather pointless in practice) were resulting in shifts with negative shift counts and shifts with counts the same as the bit width. This was always undefined in llvm, the code generated was rather horrendous but happened to work. So make sure such shifts are filtered out and replaced with something that works (the generated code is still just as horrendous as before). This fixes lp_test_format, https://bugs.freedesktop.org/show_bug.cgi?id=73846. v2: prettify by using build context shift helpers. Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
This commit is contained in:
parent
7928b946ad
commit
56335b4441
|
@ -792,29 +792,23 @@ lp_build_conv(struct gallivm_state *gallivm,
|
||||||
unsigned dst_shift = lp_const_shift(dst_type);
|
unsigned dst_shift = lp_const_shift(dst_type);
|
||||||
unsigned src_offset = lp_const_offset(src_type);
|
unsigned src_offset = lp_const_offset(src_type);
|
||||||
unsigned dst_offset = lp_const_offset(dst_type);
|
unsigned dst_offset = lp_const_offset(dst_type);
|
||||||
|
struct lp_build_context bld;
|
||||||
|
lp_build_context_init(&bld, gallivm, tmp_type);
|
||||||
|
|
||||||
/* Compensate for different offsets */
|
/* Compensate for different offsets */
|
||||||
if (dst_offset > src_offset && src_type.width > dst_type.width) {
|
/* sscaled -> unorm and similar would cause negative shift count, skip */
|
||||||
|
if (dst_offset > src_offset && src_type.width > dst_type.width && src_shift > 0) {
|
||||||
for (i = 0; i < num_tmps; ++i) {
|
for (i = 0; i < num_tmps; ++i) {
|
||||||
LLVMValueRef shifted;
|
LLVMValueRef shifted;
|
||||||
LLVMValueRef shift = lp_build_const_int_vec(gallivm, tmp_type, src_shift - 1);
|
|
||||||
if(src_type.sign)
|
|
||||||
shifted = LLVMBuildAShr(builder, tmp[i], shift, "");
|
|
||||||
else
|
|
||||||
shifted = LLVMBuildLShr(builder, tmp[i], shift, "");
|
|
||||||
|
|
||||||
|
shifted = lp_build_shr_imm(&bld, tmp[i], src_shift - 1);
|
||||||
tmp[i] = LLVMBuildSub(builder, tmp[i], shifted, "");
|
tmp[i] = LLVMBuildSub(builder, tmp[i], shifted, "");
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
if(src_shift > dst_shift) {
|
if(src_shift > dst_shift) {
|
||||||
LLVMValueRef shift = lp_build_const_int_vec(gallivm, tmp_type,
|
|
||||||
src_shift - dst_shift);
|
|
||||||
for(i = 0; i < num_tmps; ++i)
|
for(i = 0; i < num_tmps; ++i)
|
||||||
if(src_type.sign)
|
tmp[i] = lp_build_shr_imm(&bld, tmp[i], src_shift - dst_shift);
|
||||||
tmp[i] = LLVMBuildAShr(builder, tmp[i], shift, "");
|
|
||||||
else
|
|
||||||
tmp[i] = LLVMBuildLShr(builder, tmp[i], shift, "");
|
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
@ -900,14 +894,27 @@ lp_build_conv(struct gallivm_state *gallivm,
|
||||||
unsigned dst_shift = lp_const_shift(dst_type);
|
unsigned dst_shift = lp_const_shift(dst_type);
|
||||||
unsigned src_offset = lp_const_offset(src_type);
|
unsigned src_offset = lp_const_offset(src_type);
|
||||||
unsigned dst_offset = lp_const_offset(dst_type);
|
unsigned dst_offset = lp_const_offset(dst_type);
|
||||||
|
struct lp_build_context bld;
|
||||||
|
lp_build_context_init(&bld, gallivm, tmp_type);
|
||||||
|
|
||||||
if (src_shift < dst_shift) {
|
if (src_shift < dst_shift) {
|
||||||
LLVMValueRef pre_shift[LP_MAX_VECTOR_LENGTH];
|
LLVMValueRef pre_shift[LP_MAX_VECTOR_LENGTH];
|
||||||
LLVMValueRef shift = lp_build_const_int_vec(gallivm, tmp_type, dst_shift - src_shift);
|
|
||||||
|
|
||||||
for (i = 0; i < num_tmps; ++i) {
|
if (dst_shift - src_shift < dst_type.width) {
|
||||||
pre_shift[i] = tmp[i];
|
for (i = 0; i < num_tmps; ++i) {
|
||||||
tmp[i] = LLVMBuildShl(builder, tmp[i], shift, "");
|
pre_shift[i] = tmp[i];
|
||||||
|
tmp[i] = lp_build_shl_imm(&bld, tmp[i], dst_shift - src_shift);
|
||||||
|
}
|
||||||
|
}
|
||||||
|
else {
|
||||||
|
/*
|
||||||
|
* This happens for things like sscaled -> unorm conversions. Shift
|
||||||
|
* counts equal to bit width cause undefined results, so hack around it.
|
||||||
|
*/
|
||||||
|
for (i = 0; i < num_tmps; ++i) {
|
||||||
|
pre_shift[i] = tmp[i];
|
||||||
|
tmp[i] = lp_build_zero(gallivm, dst_type);
|
||||||
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
/* Compensate for different offsets */
|
/* Compensate for different offsets */
|
||||||
|
|
Loading…
Reference in New Issue