Cortex-M4F: Assembly instruction SMLAxy (and some others) gives wrong result

Hello ARM Support team.

I am having trouble with MSP432E401Y (Cortex-M4F - core). I use evaluation board MSP-EXP432E401Y, Rev. 1.0.

I write my code in Code Composer Studio 12.3.0 in assemble code.

Some instructions do not act correctly. Please, help me to find my mistakes.

Previously I have posted my request to Texas Instruments support forum, but they answer me: “Please consult with Arm support forum for guidance”.

Let us take instruction SMLAxy for example. Description of this instruction (from ARM DUI0473J, pg. 10-479): SMLAxy multiplies the 16-bit signed integers from the selected halves of Rn and Rm, adds the 32-bit result to the 32-bit value in Ra, and places the result in Rd.

I understand it in this way. If we use SMLATT (for example) we have:

Rd := Top_Halfword{Rn} * Top_Halfword{Rm} + Ra

First of all, look at an example without overflow. My code:

MOV R1,  #0x1244

MOVT R1,  #0x0229

MOV R2,  #0x3769

MOVT R2,  #0x0455

MOV R0,  #0000

MOVT R0,  #0001

SMLATT    R3, R1, R2, R0 ; After executing the instruction R3 = 0x000A5B9D

Compare the result with my own calculations:

Rd <= R3

Rn <= R1 := 0x02291244

Rm <= R2 := 0x04553769

Ra <= R0 := 0x00010000

Rd = THW[Rn] * THW[Rm] + Ra = 0x0229 * 0x0455 + 0x00010000 = 0x00095B9D + 0x00010000 = 0x000A5B9D, the result matches.

But if we take other operands, interesting results come out:

MOV R0,             #0x0001

MOVT R0,             #0xAAAA

MOV R1,             #0x000A

MOVT R1,             #0xAAAA

MOV R2,             #0x0001

MOVT R2,             #0x0060

SMLATT     R3, R1, R0, R2; After executing we have R3 = #0x1CD238E5

Compare with my calculation:

Rd = THW[Rn] * THW[Rm] + Ra = 0xAAAA * 0xAAAA + 0x00600001 = 0x71C638E4 + 0x00600001 = 0x722638E5, the result does not match.

The bottom halfword of the result matches, but the top halfword of the result does not match. I have no idea where this result comes from.

MOV R0,             #0x0001

MOVT R0,           #0xAAAA

MOV R1,             #0x000A

MOVT R1,           #0xAAAA

MOV R2,             #0 ; I want to check the multiplication only

SMLATT     R3, R1, R0, R2 ;    After executing the instruction R3 = #0x1C7238E4

Compare with my calculation:

Rd = THW[Rn] * THW[Rm] + Ra = 0xAAAA * 0xAAAA + 0 = 0x71C638E4, the result does not match.

I faced with this problem in other instructions too: SMLAD, SMLAL, SMLALxy, SMLALD, SMLAWx, SMLSLD. I have examples for all instructions.

I think, in all these instructions I do not understand some point. Please explain me, how the multiplication in these instructions acts. I would really appreciate your help.

Best regards.