<div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">On Mon, Apr 30, 2018 at 3:53 PM, Chema Casanova <span dir="ltr"><<a href="mailto:jmcasanova@igalia.com" target="_blank">jmcasanova@igalia.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><span class=""><br>
<br>
On 30/04/18 23:12, Jason Ekstrand wrote:<br>
> On Mon, Apr 30, 2018 at 7:18 AM, Iago Toral Quiroga <<a href="mailto:itoral@igalia.com">itoral@igalia.com</a><br>
</span><span class="">> <mailto:<a href="mailto:itoral@igalia.com">itoral@igalia.com</a>>> wrote:<br>
> <br>
> From: Jose Maria Casanova Crespo <<a href="mailto:jmcasanova@igalia.com">jmcasanova@igalia.com</a><br>
</span>> <mailto:<a href="mailto:jmcasanova@igalia.com">jmcasanova@igalia.com</a>><wbr>><br>
<div><div class="h5">> <br>
> 16-bit immediates need to replicate the 16-bit immediate value<br>
> in both words of the 32-bit value. This needs to be careful<br>
> to avoid sign-extension, which the previous implementation was<br>
> not handling properly.<br>
> <br>
> For example, with the previous implementation, storing the value<br>
> -3 would generate imm.d = 0xfffffffd due to signed integer sign<br>
> extension, which is not correct. Instead, we should cast to<br>
> unsigned, which gives us the correct result: imm.ud = 0xfffdfffd.<br>
> <br>
> We only had a couple of cases hitting this path in the driver<br>
> until now, one with value -1, which would work since all bits are<br>
> one in this case, and another with value -2 in brw_clip_tri(),<br>
> which would hit the aforementioned issue (this case only affects<br>
> gen4 although we are not aware of whether this was causing an<br>
> actual bug somewhere).<br>
> ---<br>
> src/intel/compiler/brw_reg.h | 2 +-<br>
> 1 file changed, 1 insertion(+), 1 deletion(-)<br>
> <br>
> diff --git a/src/intel/compiler/brw_reg.h b/src/intel/compiler/brw_reg.h<br>
> index dff9b970b2..0084a78af6 100644<br>
> --- a/src/intel/compiler/brw_reg.h<br>
> +++ b/src/intel/compiler/brw_reg.h<br>
> @@ -705,7 +705,7 @@ static inline struct brw_reg<br>
> brw_imm_w(int16_t w)<br>
> {<br>
> struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_<wbr>W);<br>
> - imm.d = w | (w << 16);<br>
> + imm.ud = (uint16_t)w | ((uint16_t)w << 16);<br>
<br>
> Uh... Is this cast right? Doing a << 16 on a 16-bit data type should<br>
> yield undefined results. I think you want a (uint32_t) cast.<br>
<br>
</div></div>In my test code it was working at least with GCC, I think it is because<br>
at the end we have an integer promotion for any type with lower rank<br>
than int.<br>
<br>
"Formally, the rule says (C11 6.3.1.1):<br>
<br>
If an int can represent all values of the original type (as<br>
restricted by the width, for a bit-field), the value is converted to an<br>
int; otherwise, it is converted to an unsigned int. These are called the<br>
integer promotions."<br>
<br>
But I agree that is clearer if we just use (uint32_t).<br>
I can change also the brw_imm_uw case that has the same issue.<br></blockquote><div><br></div><div>Yeah, best to make it clear. :-) <br></div></div></div></div>