Unfortunately, cbnz/cbz has a 6-bit:'0' (7-bit aligned to 2) immediate offset that is *zero extended*, i.e. it is only for forward branches.

Chung-Lin

On 2010/11/29 下午 04:14, Revital1 Eres wrote:

Hello,

I have a question about cbnz/cbz thumb-2 instruction implementation in thumb2.md file:

I have an example where we jump to a label which appears before the branch; for example:

     L4
     ...
     cmp     r3, 0
     bne     .L4

It seems that cbnz instruction should be applied in this loop; replacing cmp+bne; however, cbnz fails to be applied as diff = ADDRESS (L4) - ADDRESS (bne .L4) is negative and according to thumb2_cbz in thumb2.md it should be 2<=diff<=128 (please see snippet below taken from thumb2_cbz).

So I want to double check if the current implementation of thumb2_cbnz in thumb2.md needs to be changed to enable it.

The following is from thumb2_cbnz in thumb2.md:

[(set (attr "length") (if_then_else (and (ge (minus (match_dup 1) (pc)) (const_int 2)) (le (minus (match_dup 1) (pc)) (const_int 128)) (eq (symbol_ref ("which_alternative")) (const_int 0))) (const_int 2) (const_int 8)))]

Thanks, Revital

---

linaro-toolchain mailing list linaro-toolchain@lists.linaro.org http://lists.linaro.org/mailman/listinfo/linaro-toolchain

In the ARM architecture: Under ARM mode, the PC(r15) register value is "address of current insn + 8", while in Thumb mode (including Thumb-2), it is "address of current insn + 4"

With the way the cbz/cbnz immediate is defined, which is 0--126 (even), the branch range is then: +4 to +130 :)

On the issue of the doloop_end pattern for ARM/Thumb-2 in the context of enabling SMS, I actually have tried it in the past. IIRC, I used a "subs+bne" assembly sequence then.

Chung-Lin

On 2010/11/29 下午 04:37, Revital1 Eres wrote:

OK, thanks for your answer. Is there other instruction for backward jumps that can be used to replace cmp to zero+bne?

btw, in the following link of thumb-2 instructions set I understood that the label can appear before the branch so it was confusing...

CB{N}Z Rn,<label> If Rn {== or !=} 0 then PC := label. label is (this instruction + 4-130).

http://infocenter.arm.com/help/topic/com.arm.doc.qrc0001l/QRC0001_UAL.pdf

Thanks, Revital

From: Chung-Lin Tangcltang@codesourcery.com To: Revital1 Eres/Haifa/IBM@IBMIL Cc: linaro-toolchain@lists.linaro.org Date: 29/11/2010 10:21 AM Subject: Re: A question about thumb2 cbnz/cbz implementation in thumb2.md

Unfortunately, cbnz/cbz has a 6-bit:'0' (7-bit aligned to 2) immediate offset that is *zero extended*, i.e. it is only for forward branches.

Chung-Lin

On 2010/11/29 下午 04:14, Revital1 Eres wrote:

...

Hello,

I have a question about cbnz/cbz thumb-2 instruction implementation in thumb2.md file:

I have an example where we jump to a label which appears before the

branch;

...

for example:

      L4
      ...
      cmp     r3, 0
      bne     .L4

It seems that cbnz instruction should be applied in this loop; replacing cmp+bne; however, cbnz fails to be applied as diff = ADDRESS (L4) -

ADDRESS

...

(bne .L4) is negative and according to thumb2_cbz in thumb2.md it should be 2<=diff<=128 (please see snippet below taken from thumb2_cbz).

So I want to double check if the current implementation of thumb2_cbnz in thumb2.md needs to be changed to enable it.

The following is from thumb2_cbnz in thumb2.md:

[(set (attr "length")
      (if_then_else
          (and (ge (minus (match_dup 1) (pc)) (const_int 2))
               (le (minus (match_dup 1) (pc)) (const_int 128))
               (eq (symbol_ref ("which_alternative")) (const_int 0)))
          (const_int 2)
          (const_int 8)))]

Thanks, Revital

---

linaro-toolchain mailing list linaro-toolchain@lists.linaro.org http://lists.linaro.org/mailman/listinfo/linaro-toolchain