-rw-r--r-- 100727 lib25519-20240928/crypto_mGnP/ed25519/amd64-51/ge25519_pnielsadd_p1p1.S raw
#include "crypto_asm_hidden.h" // linker define ge25519_pnielsadd_p1p1 // linker use REDMASK51 // linker use 2P0 // linker use 2P1234 # qhasm: int64 rp # qhasm: int64 pp # qhasm: int64 qp # qhasm: input rp # qhasm: input pp # qhasm: input qp # qhasm: int64 caller1 # qhasm: int64 caller2 # qhasm: int64 caller3 # qhasm: int64 caller4 # qhasm: int64 caller5 # qhasm: int64 caller6 # qhasm: int64 caller7 # qhasm: caller caller1 # qhasm: caller caller2 # qhasm: caller caller3 # qhasm: caller caller4 # qhasm: caller caller5 # qhasm: caller caller6 # qhasm: caller caller7 # qhasm: stack64 caller1_stack # qhasm: stack64 caller2_stack # qhasm: stack64 caller3_stack # qhasm: stack64 caller4_stack # qhasm: stack64 caller5_stack # qhasm: stack64 caller6_stack # qhasm: stack64 caller7_stack # qhasm: int64 a0 # qhasm: int64 a1 # qhasm: int64 a2 # qhasm: int64 a3 # qhasm: int64 a4 # qhasm: stack64 a0_stack # qhasm: stack64 a1_stack # qhasm: stack64 a2_stack # qhasm: stack64 a3_stack # qhasm: stack64 a4_stack # qhasm: int64 b0 # qhasm: int64 b1 # qhasm: int64 b2 # qhasm: int64 b3 # qhasm: int64 b4 # qhasm: stack64 b0_stack # qhasm: stack64 b1_stack # qhasm: stack64 b2_stack # qhasm: stack64 b3_stack # qhasm: stack64 b4_stack # qhasm: int64 c0 # qhasm: int64 c1 # qhasm: int64 c2 # qhasm: int64 c3 # qhasm: int64 c4 # qhasm: stack64 c0_stack # qhasm: stack64 c1_stack # qhasm: stack64 c2_stack # qhasm: stack64 c3_stack # qhasm: stack64 c4_stack # qhasm: int64 d0 # qhasm: int64 d1 # qhasm: int64 d2 # qhasm: int64 d3 # qhasm: int64 d4 # qhasm: stack64 d0_stack # qhasm: stack64 d1_stack # qhasm: stack64 d2_stack # qhasm: stack64 d3_stack # qhasm: stack64 d4_stack # qhasm: int64 t10 # qhasm: int64 t11 # qhasm: int64 t12 # qhasm: int64 t13 # qhasm: int64 t14 # qhasm: stack64 t10_stack # qhasm: stack64 t11_stack # qhasm: stack64 t12_stack # qhasm: stack64 t13_stack # qhasm: stack64 t14_stack # qhasm: int64 t20 # qhasm: int64 t21 # qhasm: int64 t22 # qhasm: int64 t23 # qhasm: int64 t24 # qhasm: stack64 t20_stack # qhasm: stack64 t21_stack # qhasm: stack64 t22_stack # qhasm: stack64 t23_stack # qhasm: stack64 t24_stack # qhasm: int64 rx0 # qhasm: int64 rx1 # qhasm: int64 rx2 # qhasm: int64 rx3 # qhasm: int64 rx4 # qhasm: int64 ry0 # qhasm: int64 ry1 # qhasm: int64 ry2 # qhasm: int64 ry3 # qhasm: int64 ry4 # qhasm: int64 rz0 # qhasm: int64 rz1 # qhasm: int64 rz2 # qhasm: int64 rz3 # qhasm: int64 rz4 # qhasm: int64 rt0 # qhasm: int64 rt1 # qhasm: int64 rt2 # qhasm: int64 rt3 # qhasm: int64 rt4 # qhasm: int64 x0 # qhasm: int64 x1 # qhasm: int64 x2 # qhasm: int64 x3 # qhasm: int64 x4 # qhasm: int64 mulr01 # qhasm: int64 mulr11 # qhasm: int64 mulr21 # qhasm: int64 mulr31 # qhasm: int64 mulr41 # qhasm: int64 mulrax # qhasm: int64 mulrdx # qhasm: int64 mult # qhasm: int64 mulredmask # qhasm: stack64 mulx219_stack # qhasm: stack64 mulx319_stack # qhasm: stack64 mulx419_stack # qhasm: enter CRYPTO_SHARED_NAMESPACE(ge25519_pnielsadd_p1p1) .text .p2align 5 ASM_HIDDEN _CRYPTO_SHARED_NAMESPACE(ge25519_pnielsadd_p1p1) .globl _CRYPTO_SHARED_NAMESPACE(ge25519_pnielsadd_p1p1) ASM_HIDDEN CRYPTO_SHARED_NAMESPACE(ge25519_pnielsadd_p1p1) .globl CRYPTO_SHARED_NAMESPACE(ge25519_pnielsadd_p1p1) _CRYPTO_SHARED_NAMESPACE(ge25519_pnielsadd_p1p1): CRYPTO_SHARED_NAMESPACE(ge25519_pnielsadd_p1p1): mov %rsp,%r11 and $31,%r11 add $160,%r11 sub %r11,%rsp # qhasm: caller1_stack = caller1 # asm 1: movq <caller1=int64#9,>caller1_stack=stack64#1 # asm 2: movq <caller1=%r11,>caller1_stack=0(%rsp) movq %r11,0(%rsp) # qhasm: caller2_stack = caller2 # asm 1: movq <caller2=int64#10,>caller2_stack=stack64#2 # asm 2: movq <caller2=%r12,>caller2_stack=8(%rsp) movq %r12,8(%rsp) # qhasm: caller3_stack = caller3 # asm 1: movq <caller3=int64#11,>caller3_stack=stack64#3 # asm 2: movq <caller3=%r13,>caller3_stack=16(%rsp) movq %r13,16(%rsp) # qhasm: caller4_stack = caller4 # asm 1: movq <caller4=int64#12,>caller4_stack=stack64#4 # asm 2: movq <caller4=%r14,>caller4_stack=24(%rsp) movq %r14,24(%rsp) # qhasm: caller5_stack = caller5 # asm 1: movq <caller5=int64#13,>caller5_stack=stack64#5 # asm 2: movq <caller5=%r15,>caller5_stack=32(%rsp) movq %r15,32(%rsp) # qhasm: caller6_stack = caller6 # asm 1: movq <caller6=int64#14,>caller6_stack=stack64#6 # asm 2: movq <caller6=%rbx,>caller6_stack=40(%rsp) movq %rbx,40(%rsp) # qhasm: caller7_stack = caller7 # asm 1: movq <caller7=int64#15,>caller7_stack=stack64#7 # asm 2: movq <caller7=%rbp,>caller7_stack=48(%rsp) movq %rbp,48(%rsp) # qhasm: qp = qp # asm 1: mov <qp=int64#3,>qp=int64#4 # asm 2: mov <qp=%rdx,>qp=%rcx mov %rdx,%rcx # qhasm: a0 = *(uint64 *)(pp + 40) # asm 1: movq 40(<pp=int64#2),>a0=int64#3 # asm 2: movq 40(<pp=%rsi),>a0=%rdx movq 40(%rsi),%rdx # qhasm: a1 = *(uint64 *)(pp + 48) # asm 1: movq 48(<pp=int64#2),>a1=int64#5 # asm 2: movq 48(<pp=%rsi),>a1=%r8 movq 48(%rsi),%r8 # qhasm: a2 = *(uint64 *)(pp + 56) # asm 1: movq 56(<pp=int64#2),>a2=int64#6 # asm 2: movq 56(<pp=%rsi),>a2=%r9 movq 56(%rsi),%r9 # qhasm: a3 = *(uint64 *)(pp + 64) # asm 1: movq 64(<pp=int64#2),>a3=int64#7 # asm 2: movq 64(<pp=%rsi),>a3=%rax movq 64(%rsi),%rax # qhasm: a4 = *(uint64 *)(pp + 72) # asm 1: movq 72(<pp=int64#2),>a4=int64#8 # asm 2: movq 72(<pp=%rsi),>a4=%r10 movq 72(%rsi),%r10 # qhasm: b0 = a0 # asm 1: mov <a0=int64#3,>b0=int64#9 # asm 2: mov <a0=%rdx,>b0=%r11 mov %rdx,%r11 # qhasm: b1 = a1 # asm 1: mov <a1=int64#5,>b1=int64#10 # asm 2: mov <a1=%r8,>b1=%r12 mov %r8,%r12 # qhasm: b2 = a2 # asm 1: mov <a2=int64#6,>b2=int64#11 # asm 2: mov <a2=%r9,>b2=%r13 mov %r9,%r13 # qhasm: b3 = a3 # asm 1: mov <a3=int64#7,>b3=int64#12 # asm 2: mov <a3=%rax,>b3=%r14 mov %rax,%r14 # qhasm: b4 = a4 # asm 1: mov <a4=int64#8,>b4=int64#13 # asm 2: mov <a4=%r10,>b4=%r15 mov %r10,%r15 # qhasm: a0 += *(uint64 *) &CRYPTO_SHARED_NAMESPACE(2P0) # asm 1: add CRYPTO_SHARED_NAMESPACE(2P0),<a0=int64#3 # asm 2: add CRYPTO_SHARED_NAMESPACE(2P0),<a0=%rdx add CRYPTO_SHARED_NAMESPACE(2P0)(%rip),%rdx # qhasm: a1 += *(uint64 *) &CRYPTO_SHARED_NAMESPACE(2P1234) # asm 1: add CRYPTO_SHARED_NAMESPACE(2P1234),<a1=int64#5 # asm 2: add CRYPTO_SHARED_NAMESPACE(2P1234),<a1=%r8 add CRYPTO_SHARED_NAMESPACE(2P1234)(%rip),%r8 # qhasm: a2 += *(uint64 *) &CRYPTO_SHARED_NAMESPACE(2P1234) # asm 1: add CRYPTO_SHARED_NAMESPACE(2P1234),<a2=int64#6 # asm 2: add CRYPTO_SHARED_NAMESPACE(2P1234),<a2=%r9 add CRYPTO_SHARED_NAMESPACE(2P1234)(%rip),%r9 # qhasm: a3 += *(uint64 *) &CRYPTO_SHARED_NAMESPACE(2P1234) # asm 1: add CRYPTO_SHARED_NAMESPACE(2P1234),<a3=int64#7 # asm 2: add CRYPTO_SHARED_NAMESPACE(2P1234),<a3=%rax add CRYPTO_SHARED_NAMESPACE(2P1234)(%rip),%rax # qhasm: a4 += *(uint64 *) &CRYPTO_SHARED_NAMESPACE(2P1234) # asm 1: add CRYPTO_SHARED_NAMESPACE(2P1234),<a4=int64#8 # asm 2: add CRYPTO_SHARED_NAMESPACE(2P1234),<a4=%r10 add CRYPTO_SHARED_NAMESPACE(2P1234)(%rip),%r10 # qhasm: b0 += *(uint64 *) (pp + 0) # asm 1: addq 0(<pp=int64#2),<b0=int64#9 # asm 2: addq 0(<pp=%rsi),<b0=%r11 addq 0(%rsi),%r11 # qhasm: b1 += *(uint64 *) (pp + 8) # asm 1: addq 8(<pp=int64#2),<b1=int64#10 # asm 2: addq 8(<pp=%rsi),<b1=%r12 addq 8(%rsi),%r12 # qhasm: b2 += *(uint64 *) (pp + 16) # asm 1: addq 16(<pp=int64#2),<b2=int64#11 # asm 2: addq 16(<pp=%rsi),<b2=%r13 addq 16(%rsi),%r13 # qhasm: b3 += *(uint64 *) (pp + 24) # asm 1: addq 24(<pp=int64#2),<b3=int64#12 # asm 2: addq 24(<pp=%rsi),<b3=%r14 addq 24(%rsi),%r14 # qhasm: b4 += *(uint64 *) (pp + 32) # asm 1: addq 32(<pp=int64#2),<b4=int64#13 # asm 2: addq 32(<pp=%rsi),<b4=%r15 addq 32(%rsi),%r15 # qhasm: a0 -= *(uint64 *) (pp + 0) # asm 1: subq 0(<pp=int64#2),<a0=int64#3 # asm 2: subq 0(<pp=%rsi),<a0=%rdx subq 0(%rsi),%rdx # qhasm: a1 -= *(uint64 *) (pp + 8) # asm 1: subq 8(<pp=int64#2),<a1=int64#5 # asm 2: subq 8(<pp=%rsi),<a1=%r8 subq 8(%rsi),%r8 # qhasm: a2 -= *(uint64 *) (pp + 16) # asm 1: subq 16(<pp=int64#2),<a2=int64#6 # asm 2: subq 16(<pp=%rsi),<a2=%r9 subq 16(%rsi),%r9 # qhasm: a3 -= *(uint64 *) (pp + 24) # asm 1: subq 24(<pp=int64#2),<a3=int64#7 # asm 2: subq 24(<pp=%rsi),<a3=%rax subq 24(%rsi),%rax # qhasm: a4 -= *(uint64 *) (pp + 32) # asm 1: subq 32(<pp=int64#2),<a4=int64#8 # asm 2: subq 32(<pp=%rsi),<a4=%r10 subq 32(%rsi),%r10 # qhasm: a0_stack = a0 # asm 1: movq <a0=int64#3,>a0_stack=stack64#8 # asm 2: movq <a0=%rdx,>a0_stack=56(%rsp) movq %rdx,56(%rsp) # qhasm: a1_stack = a1 # asm 1: movq <a1=int64#5,>a1_stack=stack64#9 # asm 2: movq <a1=%r8,>a1_stack=64(%rsp) movq %r8,64(%rsp) # qhasm: a2_stack = a2 # asm 1: movq <a2=int64#6,>a2_stack=stack64#10 # asm 2: movq <a2=%r9,>a2_stack=72(%rsp) movq %r9,72(%rsp) # qhasm: a3_stack = a3 # asm 1: movq <a3=int64#7,>a3_stack=stack64#11 # asm 2: movq <a3=%rax,>a3_stack=80(%rsp) movq %rax,80(%rsp) # qhasm: a4_stack = a4 # asm 1: movq <a4=int64#8,>a4_stack=stack64#12 # asm 2: movq <a4=%r10,>a4_stack=88(%rsp) movq %r10,88(%rsp) # qhasm: b0_stack = b0 # asm 1: movq <b0=int64#9,>b0_stack=stack64#13 # asm 2: movq <b0=%r11,>b0_stack=96(%rsp) movq %r11,96(%rsp) # qhasm: b1_stack = b1 # asm 1: movq <b1=int64#10,>b1_stack=stack64#14 # asm 2: movq <b1=%r12,>b1_stack=104(%rsp) movq %r12,104(%rsp) # qhasm: b2_stack = b2 # asm 1: movq <b2=int64#11,>b2_stack=stack64#15 # asm 2: movq <b2=%r13,>b2_stack=112(%rsp) movq %r13,112(%rsp) # qhasm: b3_stack = b3 # asm 1: movq <b3=int64#12,>b3_stack=stack64#16 # asm 2: movq <b3=%r14,>b3_stack=120(%rsp) movq %r14,120(%rsp) # qhasm: b4_stack = b4 # asm 1: movq <b4=int64#13,>b4_stack=stack64#17 # asm 2: movq <b4=%r15,>b4_stack=128(%rsp) movq %r15,128(%rsp) # qhasm: mulrax = a3_stack # asm 1: movq <a3_stack=stack64#11,>mulrax=int64#3 # asm 2: movq <a3_stack=80(%rsp),>mulrax=%rdx movq 80(%rsp),%rdx # qhasm: mulrax *= 19 # asm 1: imulq $19,<mulrax=int64#3,>mulrax=int64#7 # asm 2: imulq $19,<mulrax=%rdx,>mulrax=%rax imulq $19,%rdx,%rax # qhasm: mulx319_stack = mulrax # asm 1: movq <mulrax=int64#7,>mulx319_stack=stack64#18 # asm 2: movq <mulrax=%rax,>mulx319_stack=136(%rsp) movq %rax,136(%rsp) # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 16) # asm 1: mulq 16(<qp=int64#4) # asm 2: mulq 16(<qp=%rcx) mulq 16(%rcx) # qhasm: a0 = mulrax # asm 1: mov <mulrax=int64#7,>a0=int64#5 # asm 2: mov <mulrax=%rax,>a0=%r8 mov %rax,%r8 # qhasm: mulr01 = mulrdx # asm 1: mov <mulrdx=int64#3,>mulr01=int64#6 # asm 2: mov <mulrdx=%rdx,>mulr01=%r9 mov %rdx,%r9 # qhasm: mulrax = a4_stack # asm 1: movq <a4_stack=stack64#12,>mulrax=int64#3 # asm 2: movq <a4_stack=88(%rsp),>mulrax=%rdx movq 88(%rsp),%rdx # qhasm: mulrax *= 19 # asm 1: imulq $19,<mulrax=int64#3,>mulrax=int64#7 # asm 2: imulq $19,<mulrax=%rdx,>mulrax=%rax imulq $19,%rdx,%rax # qhasm: mulx419_stack = mulrax # asm 1: movq <mulrax=int64#7,>mulx419_stack=stack64#19 # asm 2: movq <mulrax=%rax,>mulx419_stack=144(%rsp) movq %rax,144(%rsp) # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 8) # asm 1: mulq 8(<qp=int64#4) # asm 2: mulq 8(<qp=%rcx) mulq 8(%rcx) # qhasm: carry? a0 += mulrax # asm 1: add <mulrax=int64#7,<a0=int64#5 # asm 2: add <mulrax=%rax,<a0=%r8 add %rax,%r8 # qhasm: mulr01 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr01=int64#6 # asm 2: adc <mulrdx=%rdx,<mulr01=%r9 adc %rdx,%r9 # qhasm: mulrax = a0_stack # asm 1: movq <a0_stack=stack64#8,>mulrax=int64#7 # asm 2: movq <a0_stack=56(%rsp),>mulrax=%rax movq 56(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 0) # asm 1: mulq 0(<qp=int64#4) # asm 2: mulq 0(<qp=%rcx) mulq 0(%rcx) # qhasm: carry? a0 += mulrax # asm 1: add <mulrax=int64#7,<a0=int64#5 # asm 2: add <mulrax=%rax,<a0=%r8 add %rax,%r8 # qhasm: mulr01 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr01=int64#6 # asm 2: adc <mulrdx=%rdx,<mulr01=%r9 adc %rdx,%r9 # qhasm: mulrax = a0_stack # asm 1: movq <a0_stack=stack64#8,>mulrax=int64#7 # asm 2: movq <a0_stack=56(%rsp),>mulrax=%rax movq 56(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 8) # asm 1: mulq 8(<qp=int64#4) # asm 2: mulq 8(<qp=%rcx) mulq 8(%rcx) # qhasm: a1 = mulrax # asm 1: mov <mulrax=int64#7,>a1=int64#8 # asm 2: mov <mulrax=%rax,>a1=%r10 mov %rax,%r10 # qhasm: mulr11 = mulrdx # asm 1: mov <mulrdx=int64#3,>mulr11=int64#9 # asm 2: mov <mulrdx=%rdx,>mulr11=%r11 mov %rdx,%r11 # qhasm: mulrax = a0_stack # asm 1: movq <a0_stack=stack64#8,>mulrax=int64#7 # asm 2: movq <a0_stack=56(%rsp),>mulrax=%rax movq 56(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 16) # asm 1: mulq 16(<qp=int64#4) # asm 2: mulq 16(<qp=%rcx) mulq 16(%rcx) # qhasm: a2 = mulrax # asm 1: mov <mulrax=int64#7,>a2=int64#10 # asm 2: mov <mulrax=%rax,>a2=%r12 mov %rax,%r12 # qhasm: mulr21 = mulrdx # asm 1: mov <mulrdx=int64#3,>mulr21=int64#11 # asm 2: mov <mulrdx=%rdx,>mulr21=%r13 mov %rdx,%r13 # qhasm: mulrax = a0_stack # asm 1: movq <a0_stack=stack64#8,>mulrax=int64#7 # asm 2: movq <a0_stack=56(%rsp),>mulrax=%rax movq 56(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 24) # asm 1: mulq 24(<qp=int64#4) # asm 2: mulq 24(<qp=%rcx) mulq 24(%rcx) # qhasm: a3 = mulrax # asm 1: mov <mulrax=int64#7,>a3=int64#12 # asm 2: mov <mulrax=%rax,>a3=%r14 mov %rax,%r14 # qhasm: mulr31 = mulrdx # asm 1: mov <mulrdx=int64#3,>mulr31=int64#13 # asm 2: mov <mulrdx=%rdx,>mulr31=%r15 mov %rdx,%r15 # qhasm: mulrax = a0_stack # asm 1: movq <a0_stack=stack64#8,>mulrax=int64#7 # asm 2: movq <a0_stack=56(%rsp),>mulrax=%rax movq 56(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 32) # asm 1: mulq 32(<qp=int64#4) # asm 2: mulq 32(<qp=%rcx) mulq 32(%rcx) # qhasm: a4 = mulrax # asm 1: mov <mulrax=int64#7,>a4=int64#14 # asm 2: mov <mulrax=%rax,>a4=%rbx mov %rax,%rbx # qhasm: mulr41 = mulrdx # asm 1: mov <mulrdx=int64#3,>mulr41=int64#15 # asm 2: mov <mulrdx=%rdx,>mulr41=%rbp mov %rdx,%rbp # qhasm: mulrax = a1_stack # asm 1: movq <a1_stack=stack64#9,>mulrax=int64#7 # asm 2: movq <a1_stack=64(%rsp),>mulrax=%rax movq 64(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 0) # asm 1: mulq 0(<qp=int64#4) # asm 2: mulq 0(<qp=%rcx) mulq 0(%rcx) # qhasm: carry? a1 += mulrax # asm 1: add <mulrax=int64#7,<a1=int64#8 # asm 2: add <mulrax=%rax,<a1=%r10 add %rax,%r10 # qhasm: mulr11 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr11=int64#9 # asm 2: adc <mulrdx=%rdx,<mulr11=%r11 adc %rdx,%r11 # qhasm: mulrax = a1_stack # asm 1: movq <a1_stack=stack64#9,>mulrax=int64#7 # asm 2: movq <a1_stack=64(%rsp),>mulrax=%rax movq 64(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 8) # asm 1: mulq 8(<qp=int64#4) # asm 2: mulq 8(<qp=%rcx) mulq 8(%rcx) # qhasm: carry? a2 += mulrax # asm 1: add <mulrax=int64#7,<a2=int64#10 # asm 2: add <mulrax=%rax,<a2=%r12 add %rax,%r12 # qhasm: mulr21 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr21=int64#11 # asm 2: adc <mulrdx=%rdx,<mulr21=%r13 adc %rdx,%r13 # qhasm: mulrax = a1_stack # asm 1: movq <a1_stack=stack64#9,>mulrax=int64#7 # asm 2: movq <a1_stack=64(%rsp),>mulrax=%rax movq 64(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 16) # asm 1: mulq 16(<qp=int64#4) # asm 2: mulq 16(<qp=%rcx) mulq 16(%rcx) # qhasm: carry? a3 += mulrax # asm 1: add <mulrax=int64#7,<a3=int64#12 # asm 2: add <mulrax=%rax,<a3=%r14 add %rax,%r14 # qhasm: mulr31 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr31=int64#13 # asm 2: adc <mulrdx=%rdx,<mulr31=%r15 adc %rdx,%r15 # qhasm: mulrax = a1_stack # asm 1: movq <a1_stack=stack64#9,>mulrax=int64#7 # asm 2: movq <a1_stack=64(%rsp),>mulrax=%rax movq 64(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 24) # asm 1: mulq 24(<qp=int64#4) # asm 2: mulq 24(<qp=%rcx) mulq 24(%rcx) # qhasm: carry? a4 += mulrax # asm 1: add <mulrax=int64#7,<a4=int64#14 # asm 2: add <mulrax=%rax,<a4=%rbx add %rax,%rbx # qhasm: mulr41 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr41=int64#15 # asm 2: adc <mulrdx=%rdx,<mulr41=%rbp adc %rdx,%rbp # qhasm: mulrax = a1_stack # asm 1: movq <a1_stack=stack64#9,>mulrax=int64#3 # asm 2: movq <a1_stack=64(%rsp),>mulrax=%rdx movq 64(%rsp),%rdx # qhasm: mulrax *= 19 # asm 1: imulq $19,<mulrax=int64#3,>mulrax=int64#7 # asm 2: imulq $19,<mulrax=%rdx,>mulrax=%rax imulq $19,%rdx,%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 32) # asm 1: mulq 32(<qp=int64#4) # asm 2: mulq 32(<qp=%rcx) mulq 32(%rcx) # qhasm: carry? a0 += mulrax # asm 1: add <mulrax=int64#7,<a0=int64#5 # asm 2: add <mulrax=%rax,<a0=%r8 add %rax,%r8 # qhasm: mulr01 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr01=int64#6 # asm 2: adc <mulrdx=%rdx,<mulr01=%r9 adc %rdx,%r9 # qhasm: mulrax = a2_stack # asm 1: movq <a2_stack=stack64#10,>mulrax=int64#7 # asm 2: movq <a2_stack=72(%rsp),>mulrax=%rax movq 72(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 0) # asm 1: mulq 0(<qp=int64#4) # asm 2: mulq 0(<qp=%rcx) mulq 0(%rcx) # qhasm: carry? a2 += mulrax # asm 1: add <mulrax=int64#7,<a2=int64#10 # asm 2: add <mulrax=%rax,<a2=%r12 add %rax,%r12 # qhasm: mulr21 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr21=int64#11 # asm 2: adc <mulrdx=%rdx,<mulr21=%r13 adc %rdx,%r13 # qhasm: mulrax = a2_stack # asm 1: movq <a2_stack=stack64#10,>mulrax=int64#7 # asm 2: movq <a2_stack=72(%rsp),>mulrax=%rax movq 72(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 8) # asm 1: mulq 8(<qp=int64#4) # asm 2: mulq 8(<qp=%rcx) mulq 8(%rcx) # qhasm: carry? a3 += mulrax # asm 1: add <mulrax=int64#7,<a3=int64#12 # asm 2: add <mulrax=%rax,<a3=%r14 add %rax,%r14 # qhasm: mulr31 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr31=int64#13 # asm 2: adc <mulrdx=%rdx,<mulr31=%r15 adc %rdx,%r15 # qhasm: mulrax = a2_stack # asm 1: movq <a2_stack=stack64#10,>mulrax=int64#7 # asm 2: movq <a2_stack=72(%rsp),>mulrax=%rax movq 72(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 16) # asm 1: mulq 16(<qp=int64#4) # asm 2: mulq 16(<qp=%rcx) mulq 16(%rcx) # qhasm: carry? a4 += mulrax # asm 1: add <mulrax=int64#7,<a4=int64#14 # asm 2: add <mulrax=%rax,<a4=%rbx add %rax,%rbx # qhasm: mulr41 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr41=int64#15 # asm 2: adc <mulrdx=%rdx,<mulr41=%rbp adc %rdx,%rbp # qhasm: mulrax = a2_stack # asm 1: movq <a2_stack=stack64#10,>mulrax=int64#3 # asm 2: movq <a2_stack=72(%rsp),>mulrax=%rdx movq 72(%rsp),%rdx # qhasm: mulrax *= 19 # asm 1: imulq $19,<mulrax=int64#3,>mulrax=int64#7 # asm 2: imulq $19,<mulrax=%rdx,>mulrax=%rax imulq $19,%rdx,%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 24) # asm 1: mulq 24(<qp=int64#4) # asm 2: mulq 24(<qp=%rcx) mulq 24(%rcx) # qhasm: carry? a0 += mulrax # asm 1: add <mulrax=int64#7,<a0=int64#5 # asm 2: add <mulrax=%rax,<a0=%r8 add %rax,%r8 # qhasm: mulr01 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr01=int64#6 # asm 2: adc <mulrdx=%rdx,<mulr01=%r9 adc %rdx,%r9 # qhasm: mulrax = a2_stack # asm 1: movq <a2_stack=stack64#10,>mulrax=int64#3 # asm 2: movq <a2_stack=72(%rsp),>mulrax=%rdx movq 72(%rsp),%rdx # qhasm: mulrax *= 19 # asm 1: imulq $19,<mulrax=int64#3,>mulrax=int64#7 # asm 2: imulq $19,<mulrax=%rdx,>mulrax=%rax imulq $19,%rdx,%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 32) # asm 1: mulq 32(<qp=int64#4) # asm 2: mulq 32(<qp=%rcx) mulq 32(%rcx) # qhasm: carry? a1 += mulrax # asm 1: add <mulrax=int64#7,<a1=int64#8 # asm 2: add <mulrax=%rax,<a1=%r10 add %rax,%r10 # qhasm: mulr11 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr11=int64#9 # asm 2: adc <mulrdx=%rdx,<mulr11=%r11 adc %rdx,%r11 # qhasm: mulrax = a3_stack # asm 1: movq <a3_stack=stack64#11,>mulrax=int64#7 # asm 2: movq <a3_stack=80(%rsp),>mulrax=%rax movq 80(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 0) # asm 1: mulq 0(<qp=int64#4) # asm 2: mulq 0(<qp=%rcx) mulq 0(%rcx) # qhasm: carry? a3 += mulrax # asm 1: add <mulrax=int64#7,<a3=int64#12 # asm 2: add <mulrax=%rax,<a3=%r14 add %rax,%r14 # qhasm: mulr31 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr31=int64#13 # asm 2: adc <mulrdx=%rdx,<mulr31=%r15 adc %rdx,%r15 # qhasm: mulrax = a3_stack # asm 1: movq <a3_stack=stack64#11,>mulrax=int64#7 # asm 2: movq <a3_stack=80(%rsp),>mulrax=%rax movq 80(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 8) # asm 1: mulq 8(<qp=int64#4) # asm 2: mulq 8(<qp=%rcx) mulq 8(%rcx) # qhasm: carry? a4 += mulrax # asm 1: add <mulrax=int64#7,<a4=int64#14 # asm 2: add <mulrax=%rax,<a4=%rbx add %rax,%rbx # qhasm: mulr41 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr41=int64#15 # asm 2: adc <mulrdx=%rdx,<mulr41=%rbp adc %rdx,%rbp # qhasm: mulrax = mulx319_stack # asm 1: movq <mulx319_stack=stack64#18,>mulrax=int64#7 # asm 2: movq <mulx319_stack=136(%rsp),>mulrax=%rax movq 136(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 24) # asm 1: mulq 24(<qp=int64#4) # asm 2: mulq 24(<qp=%rcx) mulq 24(%rcx) # qhasm: carry? a1 += mulrax # asm 1: add <mulrax=int64#7,<a1=int64#8 # asm 2: add <mulrax=%rax,<a1=%r10 add %rax,%r10 # qhasm: mulr11 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr11=int64#9 # asm 2: adc <mulrdx=%rdx,<mulr11=%r11 adc %rdx,%r11 # qhasm: mulrax = mulx319_stack # asm 1: movq <mulx319_stack=stack64#18,>mulrax=int64#7 # asm 2: movq <mulx319_stack=136(%rsp),>mulrax=%rax movq 136(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 32) # asm 1: mulq 32(<qp=int64#4) # asm 2: mulq 32(<qp=%rcx) mulq 32(%rcx) # qhasm: carry? a2 += mulrax # asm 1: add <mulrax=int64#7,<a2=int64#10 # asm 2: add <mulrax=%rax,<a2=%r12 add %rax,%r12 # qhasm: mulr21 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr21=int64#11 # asm 2: adc <mulrdx=%rdx,<mulr21=%r13 adc %rdx,%r13 # qhasm: mulrax = a4_stack # asm 1: movq <a4_stack=stack64#12,>mulrax=int64#7 # asm 2: movq <a4_stack=88(%rsp),>mulrax=%rax movq 88(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 0) # asm 1: mulq 0(<qp=int64#4) # asm 2: mulq 0(<qp=%rcx) mulq 0(%rcx) # qhasm: carry? a4 += mulrax # asm 1: add <mulrax=int64#7,<a4=int64#14 # asm 2: add <mulrax=%rax,<a4=%rbx add %rax,%rbx # qhasm: mulr41 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr41=int64#15 # asm 2: adc <mulrdx=%rdx,<mulr41=%rbp adc %rdx,%rbp # qhasm: mulrax = mulx419_stack # asm 1: movq <mulx419_stack=stack64#19,>mulrax=int64#7 # asm 2: movq <mulx419_stack=144(%rsp),>mulrax=%rax movq 144(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 16) # asm 1: mulq 16(<qp=int64#4) # asm 2: mulq 16(<qp=%rcx) mulq 16(%rcx) # qhasm: carry? a1 += mulrax # asm 1: add <mulrax=int64#7,<a1=int64#8 # asm 2: add <mulrax=%rax,<a1=%r10 add %rax,%r10 # qhasm: mulr11 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr11=int64#9 # asm 2: adc <mulrdx=%rdx,<mulr11=%r11 adc %rdx,%r11 # qhasm: mulrax = mulx419_stack # asm 1: movq <mulx419_stack=stack64#19,>mulrax=int64#7 # asm 2: movq <mulx419_stack=144(%rsp),>mulrax=%rax movq 144(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 24) # asm 1: mulq 24(<qp=int64#4) # asm 2: mulq 24(<qp=%rcx) mulq 24(%rcx) # qhasm: carry? a2 += mulrax # asm 1: add <mulrax=int64#7,<a2=int64#10 # asm 2: add <mulrax=%rax,<a2=%r12 add %rax,%r12 # qhasm: mulr21 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr21=int64#11 # asm 2: adc <mulrdx=%rdx,<mulr21=%r13 adc %rdx,%r13 # qhasm: mulrax = mulx419_stack # asm 1: movq <mulx419_stack=stack64#19,>mulrax=int64#7 # asm 2: movq <mulx419_stack=144(%rsp),>mulrax=%rax movq 144(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 32) # asm 1: mulq 32(<qp=int64#4) # asm 2: mulq 32(<qp=%rcx) mulq 32(%rcx) # qhasm: carry? a3 += mulrax # asm 1: add <mulrax=int64#7,<a3=int64#12 # asm 2: add <mulrax=%rax,<a3=%r14 add %rax,%r14 # qhasm: mulr31 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr31=int64#13 # asm 2: adc <mulrdx=%rdx,<mulr31=%r15 adc %rdx,%r15 # qhasm: mulredmask = *(uint64 *) &CRYPTO_SHARED_NAMESPACE(REDMASK51) # asm 1: movq CRYPTO_SHARED_NAMESPACE(REDMASK51),>mulredmask=int64#3 # asm 2: movq CRYPTO_SHARED_NAMESPACE(REDMASK51),>mulredmask=%rdx movq CRYPTO_SHARED_NAMESPACE(REDMASK51)(%rip),%rdx # qhasm: mulr01 = (mulr01.a0) << 13 # asm 1: shld $13,<a0=int64#5,<mulr01=int64#6 # asm 2: shld $13,<a0=%r8,<mulr01=%r9 shld $13,%r8,%r9 # qhasm: a0 &= mulredmask # asm 1: and <mulredmask=int64#3,<a0=int64#5 # asm 2: and <mulredmask=%rdx,<a0=%r8 and %rdx,%r8 # qhasm: mulr11 = (mulr11.a1) << 13 # asm 1: shld $13,<a1=int64#8,<mulr11=int64#9 # asm 2: shld $13,<a1=%r10,<mulr11=%r11 shld $13,%r10,%r11 # qhasm: a1 &= mulredmask # asm 1: and <mulredmask=int64#3,<a1=int64#8 # asm 2: and <mulredmask=%rdx,<a1=%r10 and %rdx,%r10 # qhasm: a1 += mulr01 # asm 1: add <mulr01=int64#6,<a1=int64#8 # asm 2: add <mulr01=%r9,<a1=%r10 add %r9,%r10 # qhasm: mulr21 = (mulr21.a2) << 13 # asm 1: shld $13,<a2=int64#10,<mulr21=int64#11 # asm 2: shld $13,<a2=%r12,<mulr21=%r13 shld $13,%r12,%r13 # qhasm: a2 &= mulredmask # asm 1: and <mulredmask=int64#3,<a2=int64#10 # asm 2: and <mulredmask=%rdx,<a2=%r12 and %rdx,%r12 # qhasm: a2 += mulr11 # asm 1: add <mulr11=int64#9,<a2=int64#10 # asm 2: add <mulr11=%r11,<a2=%r12 add %r11,%r12 # qhasm: mulr31 = (mulr31.a3) << 13 # asm 1: shld $13,<a3=int64#12,<mulr31=int64#13 # asm 2: shld $13,<a3=%r14,<mulr31=%r15 shld $13,%r14,%r15 # qhasm: a3 &= mulredmask # asm 1: and <mulredmask=int64#3,<a3=int64#12 # asm 2: and <mulredmask=%rdx,<a3=%r14 and %rdx,%r14 # qhasm: a3 += mulr21 # asm 1: add <mulr21=int64#11,<a3=int64#12 # asm 2: add <mulr21=%r13,<a3=%r14 add %r13,%r14 # qhasm: mulr41 = (mulr41.a4) << 13 # asm 1: shld $13,<a4=int64#14,<mulr41=int64#15 # asm 2: shld $13,<a4=%rbx,<mulr41=%rbp shld $13,%rbx,%rbp # qhasm: a4 &= mulredmask # asm 1: and <mulredmask=int64#3,<a4=int64#14 # asm 2: and <mulredmask=%rdx,<a4=%rbx and %rdx,%rbx # qhasm: a4 += mulr31 # asm 1: add <mulr31=int64#13,<a4=int64#14 # asm 2: add <mulr31=%r15,<a4=%rbx add %r15,%rbx # qhasm: mulr41 = mulr41 * 19 # asm 1: imulq $19,<mulr41=int64#15,>mulr41=int64#6 # asm 2: imulq $19,<mulr41=%rbp,>mulr41=%r9 imulq $19,%rbp,%r9 # qhasm: a0 += mulr41 # asm 1: add <mulr41=int64#6,<a0=int64#5 # asm 2: add <mulr41=%r9,<a0=%r8 add %r9,%r8 # qhasm: mult = a0 # asm 1: mov <a0=int64#5,>mult=int64#6 # asm 2: mov <a0=%r8,>mult=%r9 mov %r8,%r9 # qhasm: (uint64) mult >>= 51 # asm 1: shr $51,<mult=int64#6 # asm 2: shr $51,<mult=%r9 shr $51,%r9 # qhasm: mult += a1 # asm 1: add <a1=int64#8,<mult=int64#6 # asm 2: add <a1=%r10,<mult=%r9 add %r10,%r9 # qhasm: a1 = mult # asm 1: mov <mult=int64#6,>a1=int64#7 # asm 2: mov <mult=%r9,>a1=%rax mov %r9,%rax # qhasm: (uint64) mult >>= 51 # asm 1: shr $51,<mult=int64#6 # asm 2: shr $51,<mult=%r9 shr $51,%r9 # qhasm: a0 &= mulredmask # asm 1: and <mulredmask=int64#3,<a0=int64#5 # asm 2: and <mulredmask=%rdx,<a0=%r8 and %rdx,%r8 # qhasm: mult += a2 # asm 1: add <a2=int64#10,<mult=int64#6 # asm 2: add <a2=%r12,<mult=%r9 add %r12,%r9 # qhasm: a2 = mult # asm 1: mov <mult=int64#6,>a2=int64#8 # asm 2: mov <mult=%r9,>a2=%r10 mov %r9,%r10 # qhasm: (uint64) mult >>= 51 # asm 1: shr $51,<mult=int64#6 # asm 2: shr $51,<mult=%r9 shr $51,%r9 # qhasm: a1 &= mulredmask # asm 1: and <mulredmask=int64#3,<a1=int64#7 # asm 2: and <mulredmask=%rdx,<a1=%rax and %rdx,%rax # qhasm: mult += a3 # asm 1: add <a3=int64#12,<mult=int64#6 # asm 2: add <a3=%r14,<mult=%r9 add %r14,%r9 # qhasm: a3 = mult # asm 1: mov <mult=int64#6,>a3=int64#9 # asm 2: mov <mult=%r9,>a3=%r11 mov %r9,%r11 # qhasm: (uint64) mult >>= 51 # asm 1: shr $51,<mult=int64#6 # asm 2: shr $51,<mult=%r9 shr $51,%r9 # qhasm: a2 &= mulredmask # asm 1: and <mulredmask=int64#3,<a2=int64#8 # asm 2: and <mulredmask=%rdx,<a2=%r10 and %rdx,%r10 # qhasm: mult += a4 # asm 1: add <a4=int64#14,<mult=int64#6 # asm 2: add <a4=%rbx,<mult=%r9 add %rbx,%r9 # qhasm: a4 = mult # asm 1: mov <mult=int64#6,>a4=int64#10 # asm 2: mov <mult=%r9,>a4=%r12 mov %r9,%r12 # qhasm: (uint64) mult >>= 51 # asm 1: shr $51,<mult=int64#6 # asm 2: shr $51,<mult=%r9 shr $51,%r9 # qhasm: a3 &= mulredmask # asm 1: and <mulredmask=int64#3,<a3=int64#9 # asm 2: and <mulredmask=%rdx,<a3=%r11 and %rdx,%r11 # qhasm: mult *= 19 # asm 1: imulq $19,<mult=int64#6,>mult=int64#6 # asm 2: imulq $19,<mult=%r9,>mult=%r9 imulq $19,%r9,%r9 # qhasm: a0 += mult # asm 1: add <mult=int64#6,<a0=int64#5 # asm 2: add <mult=%r9,<a0=%r8 add %r9,%r8 # qhasm: a4 &= mulredmask # asm 1: and <mulredmask=int64#3,<a4=int64#10 # asm 2: and <mulredmask=%rdx,<a4=%r12 and %rdx,%r12 # qhasm: a0_stack = a0 # asm 1: movq <a0=int64#5,>a0_stack=stack64#8 # asm 2: movq <a0=%r8,>a0_stack=56(%rsp) movq %r8,56(%rsp) # qhasm: a1_stack = a1 # asm 1: movq <a1=int64#7,>a1_stack=stack64#9 # asm 2: movq <a1=%rax,>a1_stack=64(%rsp) movq %rax,64(%rsp) # qhasm: a2_stack = a2 # asm 1: movq <a2=int64#8,>a2_stack=stack64#10 # asm 2: movq <a2=%r10,>a2_stack=72(%rsp) movq %r10,72(%rsp) # qhasm: a3_stack = a3 # asm 1: movq <a3=int64#9,>a3_stack=stack64#11 # asm 2: movq <a3=%r11,>a3_stack=80(%rsp) movq %r11,80(%rsp) # qhasm: a4_stack = a4 # asm 1: movq <a4=int64#10,>a4_stack=stack64#12 # asm 2: movq <a4=%r12,>a4_stack=88(%rsp) movq %r12,88(%rsp) # qhasm: mulrax = b3_stack # asm 1: movq <b3_stack=stack64#16,>mulrax=int64#3 # asm 2: movq <b3_stack=120(%rsp),>mulrax=%rdx movq 120(%rsp),%rdx # qhasm: mulrax *= 19 # asm 1: imulq $19,<mulrax=int64#3,>mulrax=int64#7 # asm 2: imulq $19,<mulrax=%rdx,>mulrax=%rax imulq $19,%rdx,%rax # qhasm: mulx319_stack = mulrax # asm 1: movq <mulrax=int64#7,>mulx319_stack=stack64#18 # asm 2: movq <mulrax=%rax,>mulx319_stack=136(%rsp) movq %rax,136(%rsp) # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 56) # asm 1: mulq 56(<qp=int64#4) # asm 2: mulq 56(<qp=%rcx) mulq 56(%rcx) # qhasm: rx0 = mulrax # asm 1: mov <mulrax=int64#7,>rx0=int64#5 # asm 2: mov <mulrax=%rax,>rx0=%r8 mov %rax,%r8 # qhasm: mulr01 = mulrdx # asm 1: mov <mulrdx=int64#3,>mulr01=int64#6 # asm 2: mov <mulrdx=%rdx,>mulr01=%r9 mov %rdx,%r9 # qhasm: mulrax = b4_stack # asm 1: movq <b4_stack=stack64#17,>mulrax=int64#3 # asm 2: movq <b4_stack=128(%rsp),>mulrax=%rdx movq 128(%rsp),%rdx # qhasm: mulrax *= 19 # asm 1: imulq $19,<mulrax=int64#3,>mulrax=int64#7 # asm 2: imulq $19,<mulrax=%rdx,>mulrax=%rax imulq $19,%rdx,%rax # qhasm: mulx419_stack = mulrax # asm 1: movq <mulrax=int64#7,>mulx419_stack=stack64#19 # asm 2: movq <mulrax=%rax,>mulx419_stack=144(%rsp) movq %rax,144(%rsp) # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 48) # asm 1: mulq 48(<qp=int64#4) # asm 2: mulq 48(<qp=%rcx) mulq 48(%rcx) # qhasm: carry? rx0 += mulrax # asm 1: add <mulrax=int64#7,<rx0=int64#5 # asm 2: add <mulrax=%rax,<rx0=%r8 add %rax,%r8 # qhasm: mulr01 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr01=int64#6 # asm 2: adc <mulrdx=%rdx,<mulr01=%r9 adc %rdx,%r9 # qhasm: mulrax = b0_stack # asm 1: movq <b0_stack=stack64#13,>mulrax=int64#7 # asm 2: movq <b0_stack=96(%rsp),>mulrax=%rax movq 96(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 40) # asm 1: mulq 40(<qp=int64#4) # asm 2: mulq 40(<qp=%rcx) mulq 40(%rcx) # qhasm: carry? rx0 += mulrax # asm 1: add <mulrax=int64#7,<rx0=int64#5 # asm 2: add <mulrax=%rax,<rx0=%r8 add %rax,%r8 # qhasm: mulr01 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr01=int64#6 # asm 2: adc <mulrdx=%rdx,<mulr01=%r9 adc %rdx,%r9 # qhasm: mulrax = b0_stack # asm 1: movq <b0_stack=stack64#13,>mulrax=int64#7 # asm 2: movq <b0_stack=96(%rsp),>mulrax=%rax movq 96(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 48) # asm 1: mulq 48(<qp=int64#4) # asm 2: mulq 48(<qp=%rcx) mulq 48(%rcx) # qhasm: rx1 = mulrax # asm 1: mov <mulrax=int64#7,>rx1=int64#8 # asm 2: mov <mulrax=%rax,>rx1=%r10 mov %rax,%r10 # qhasm: mulr11 = mulrdx # asm 1: mov <mulrdx=int64#3,>mulr11=int64#9 # asm 2: mov <mulrdx=%rdx,>mulr11=%r11 mov %rdx,%r11 # qhasm: mulrax = b0_stack # asm 1: movq <b0_stack=stack64#13,>mulrax=int64#7 # asm 2: movq <b0_stack=96(%rsp),>mulrax=%rax movq 96(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 56) # asm 1: mulq 56(<qp=int64#4) # asm 2: mulq 56(<qp=%rcx) mulq 56(%rcx) # qhasm: rx2 = mulrax # asm 1: mov <mulrax=int64#7,>rx2=int64#10 # asm 2: mov <mulrax=%rax,>rx2=%r12 mov %rax,%r12 # qhasm: mulr21 = mulrdx # asm 1: mov <mulrdx=int64#3,>mulr21=int64#11 # asm 2: mov <mulrdx=%rdx,>mulr21=%r13 mov %rdx,%r13 # qhasm: mulrax = b0_stack # asm 1: movq <b0_stack=stack64#13,>mulrax=int64#7 # asm 2: movq <b0_stack=96(%rsp),>mulrax=%rax movq 96(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 64) # asm 1: mulq 64(<qp=int64#4) # asm 2: mulq 64(<qp=%rcx) mulq 64(%rcx) # qhasm: rx3 = mulrax # asm 1: mov <mulrax=int64#7,>rx3=int64#12 # asm 2: mov <mulrax=%rax,>rx3=%r14 mov %rax,%r14 # qhasm: mulr31 = mulrdx # asm 1: mov <mulrdx=int64#3,>mulr31=int64#13 # asm 2: mov <mulrdx=%rdx,>mulr31=%r15 mov %rdx,%r15 # qhasm: mulrax = b0_stack # asm 1: movq <b0_stack=stack64#13,>mulrax=int64#7 # asm 2: movq <b0_stack=96(%rsp),>mulrax=%rax movq 96(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 72) # asm 1: mulq 72(<qp=int64#4) # asm 2: mulq 72(<qp=%rcx) mulq 72(%rcx) # qhasm: rx4 = mulrax # asm 1: mov <mulrax=int64#7,>rx4=int64#14 # asm 2: mov <mulrax=%rax,>rx4=%rbx mov %rax,%rbx # qhasm: mulr41 = mulrdx # asm 1: mov <mulrdx=int64#3,>mulr41=int64#15 # asm 2: mov <mulrdx=%rdx,>mulr41=%rbp mov %rdx,%rbp # qhasm: mulrax = b1_stack # asm 1: movq <b1_stack=stack64#14,>mulrax=int64#7 # asm 2: movq <b1_stack=104(%rsp),>mulrax=%rax movq 104(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 40) # asm 1: mulq 40(<qp=int64#4) # asm 2: mulq 40(<qp=%rcx) mulq 40(%rcx) # qhasm: carry? rx1 += mulrax # asm 1: add <mulrax=int64#7,<rx1=int64#8 # asm 2: add <mulrax=%rax,<rx1=%r10 add %rax,%r10 # qhasm: mulr11 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr11=int64#9 # asm 2: adc <mulrdx=%rdx,<mulr11=%r11 adc %rdx,%r11 # qhasm: mulrax = b1_stack # asm 1: movq <b1_stack=stack64#14,>mulrax=int64#7 # asm 2: movq <b1_stack=104(%rsp),>mulrax=%rax movq 104(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 48) # asm 1: mulq 48(<qp=int64#4) # asm 2: mulq 48(<qp=%rcx) mulq 48(%rcx) # qhasm: carry? rx2 += mulrax # asm 1: add <mulrax=int64#7,<rx2=int64#10 # asm 2: add <mulrax=%rax,<rx2=%r12 add %rax,%r12 # qhasm: mulr21 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr21=int64#11 # asm 2: adc <mulrdx=%rdx,<mulr21=%r13 adc %rdx,%r13 # qhasm: mulrax = b1_stack # asm 1: movq <b1_stack=stack64#14,>mulrax=int64#7 # asm 2: movq <b1_stack=104(%rsp),>mulrax=%rax movq 104(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 56) # asm 1: mulq 56(<qp=int64#4) # asm 2: mulq 56(<qp=%rcx) mulq 56(%rcx) # qhasm: carry? rx3 += mulrax # asm 1: add <mulrax=int64#7,<rx3=int64#12 # asm 2: add <mulrax=%rax,<rx3=%r14 add %rax,%r14 # qhasm: mulr31 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr31=int64#13 # asm 2: adc <mulrdx=%rdx,<mulr31=%r15 adc %rdx,%r15 # qhasm: mulrax = b1_stack # asm 1: movq <b1_stack=stack64#14,>mulrax=int64#7 # asm 2: movq <b1_stack=104(%rsp),>mulrax=%rax movq 104(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 64) # asm 1: mulq 64(<qp=int64#4) # asm 2: mulq 64(<qp=%rcx) mulq 64(%rcx) # qhasm: carry? rx4 += mulrax # asm 1: add <mulrax=int64#7,<rx4=int64#14 # asm 2: add <mulrax=%rax,<rx4=%rbx add %rax,%rbx # qhasm: mulr41 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr41=int64#15 # asm 2: adc <mulrdx=%rdx,<mulr41=%rbp adc %rdx,%rbp # qhasm: mulrax = b1_stack # asm 1: movq <b1_stack=stack64#14,>mulrax=int64#3 # asm 2: movq <b1_stack=104(%rsp),>mulrax=%rdx movq 104(%rsp),%rdx # qhasm: mulrax *= 19 # asm 1: imulq $19,<mulrax=int64#3,>mulrax=int64#7 # asm 2: imulq $19,<mulrax=%rdx,>mulrax=%rax imulq $19,%rdx,%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 72) # asm 1: mulq 72(<qp=int64#4) # asm 2: mulq 72(<qp=%rcx) mulq 72(%rcx) # qhasm: carry? rx0 += mulrax # asm 1: add <mulrax=int64#7,<rx0=int64#5 # asm 2: add <mulrax=%rax,<rx0=%r8 add %rax,%r8 # qhasm: mulr01 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr01=int64#6 # asm 2: adc <mulrdx=%rdx,<mulr01=%r9 adc %rdx,%r9 # qhasm: mulrax = b2_stack # asm 1: movq <b2_stack=stack64#15,>mulrax=int64#7 # asm 2: movq <b2_stack=112(%rsp),>mulrax=%rax movq 112(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 40) # asm 1: mulq 40(<qp=int64#4) # asm 2: mulq 40(<qp=%rcx) mulq 40(%rcx) # qhasm: carry? rx2 += mulrax # asm 1: add <mulrax=int64#7,<rx2=int64#10 # asm 2: add <mulrax=%rax,<rx2=%r12 add %rax,%r12 # qhasm: mulr21 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr21=int64#11 # asm 2: adc <mulrdx=%rdx,<mulr21=%r13 adc %rdx,%r13 # qhasm: mulrax = b2_stack # asm 1: movq <b2_stack=stack64#15,>mulrax=int64#7 # asm 2: movq <b2_stack=112(%rsp),>mulrax=%rax movq 112(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 48) # asm 1: mulq 48(<qp=int64#4) # asm 2: mulq 48(<qp=%rcx) mulq 48(%rcx) # qhasm: carry? rx3 += mulrax # asm 1: add <mulrax=int64#7,<rx3=int64#12 # asm 2: add <mulrax=%rax,<rx3=%r14 add %rax,%r14 # qhasm: mulr31 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr31=int64#13 # asm 2: adc <mulrdx=%rdx,<mulr31=%r15 adc %rdx,%r15 # qhasm: mulrax = b2_stack # asm 1: movq <b2_stack=stack64#15,>mulrax=int64#7 # asm 2: movq <b2_stack=112(%rsp),>mulrax=%rax movq 112(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 56) # asm 1: mulq 56(<qp=int64#4) # asm 2: mulq 56(<qp=%rcx) mulq 56(%rcx) # qhasm: carry? rx4 += mulrax # asm 1: add <mulrax=int64#7,<rx4=int64#14 # asm 2: add <mulrax=%rax,<rx4=%rbx add %rax,%rbx # qhasm: mulr41 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr41=int64#15 # asm 2: adc <mulrdx=%rdx,<mulr41=%rbp adc %rdx,%rbp # qhasm: mulrax = b2_stack # asm 1: movq <b2_stack=stack64#15,>mulrax=int64#3 # asm 2: movq <b2_stack=112(%rsp),>mulrax=%rdx movq 112(%rsp),%rdx # qhasm: mulrax *= 19 # asm 1: imulq $19,<mulrax=int64#3,>mulrax=int64#7 # asm 2: imulq $19,<mulrax=%rdx,>mulrax=%rax imulq $19,%rdx,%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 64) # asm 1: mulq 64(<qp=int64#4) # asm 2: mulq 64(<qp=%rcx) mulq 64(%rcx) # qhasm: carry? rx0 += mulrax # asm 1: add <mulrax=int64#7,<rx0=int64#5 # asm 2: add <mulrax=%rax,<rx0=%r8 add %rax,%r8 # qhasm: mulr01 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr01=int64#6 # asm 2: adc <mulrdx=%rdx,<mulr01=%r9 adc %rdx,%r9 # qhasm: mulrax = b2_stack # asm 1: movq <b2_stack=stack64#15,>mulrax=int64#3 # asm 2: movq <b2_stack=112(%rsp),>mulrax=%rdx movq 112(%rsp),%rdx # qhasm: mulrax *= 19 # asm 1: imulq $19,<mulrax=int64#3,>mulrax=int64#7 # asm 2: imulq $19,<mulrax=%rdx,>mulrax=%rax imulq $19,%rdx,%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 72) # asm 1: mulq 72(<qp=int64#4) # asm 2: mulq 72(<qp=%rcx) mulq 72(%rcx) # qhasm: carry? rx1 += mulrax # asm 1: add <mulrax=int64#7,<rx1=int64#8 # asm 2: add <mulrax=%rax,<rx1=%r10 add %rax,%r10 # qhasm: mulr11 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr11=int64#9 # asm 2: adc <mulrdx=%rdx,<mulr11=%r11 adc %rdx,%r11 # qhasm: mulrax = b3_stack # asm 1: movq <b3_stack=stack64#16,>mulrax=int64#7 # asm 2: movq <b3_stack=120(%rsp),>mulrax=%rax movq 120(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 40) # asm 1: mulq 40(<qp=int64#4) # asm 2: mulq 40(<qp=%rcx) mulq 40(%rcx) # qhasm: carry? rx3 += mulrax # asm 1: add <mulrax=int64#7,<rx3=int64#12 # asm 2: add <mulrax=%rax,<rx3=%r14 add %rax,%r14 # qhasm: mulr31 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr31=int64#13 # asm 2: adc <mulrdx=%rdx,<mulr31=%r15 adc %rdx,%r15 # qhasm: mulrax = b3_stack # asm 1: movq <b3_stack=stack64#16,>mulrax=int64#7 # asm 2: movq <b3_stack=120(%rsp),>mulrax=%rax movq 120(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 48) # asm 1: mulq 48(<qp=int64#4) # asm 2: mulq 48(<qp=%rcx) mulq 48(%rcx) # qhasm: carry? rx4 += mulrax # asm 1: add <mulrax=int64#7,<rx4=int64#14 # asm 2: add <mulrax=%rax,<rx4=%rbx add %rax,%rbx # qhasm: mulr41 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr41=int64#15 # asm 2: adc <mulrdx=%rdx,<mulr41=%rbp adc %rdx,%rbp # qhasm: mulrax = mulx319_stack # asm 1: movq <mulx319_stack=stack64#18,>mulrax=int64#7 # asm 2: movq <mulx319_stack=136(%rsp),>mulrax=%rax movq 136(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 64) # asm 1: mulq 64(<qp=int64#4) # asm 2: mulq 64(<qp=%rcx) mulq 64(%rcx) # qhasm: carry? rx1 += mulrax # asm 1: add <mulrax=int64#7,<rx1=int64#8 # asm 2: add <mulrax=%rax,<rx1=%r10 add %rax,%r10 # qhasm: mulr11 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr11=int64#9 # asm 2: adc <mulrdx=%rdx,<mulr11=%r11 adc %rdx,%r11 # qhasm: mulrax = mulx319_stack # asm 1: movq <mulx319_stack=stack64#18,>mulrax=int64#7 # asm 2: movq <mulx319_stack=136(%rsp),>mulrax=%rax movq 136(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 72) # asm 1: mulq 72(<qp=int64#4) # asm 2: mulq 72(<qp=%rcx) mulq 72(%rcx) # qhasm: carry? rx2 += mulrax # asm 1: add <mulrax=int64#7,<rx2=int64#10 # asm 2: add <mulrax=%rax,<rx2=%r12 add %rax,%r12 # qhasm: mulr21 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr21=int64#11 # asm 2: adc <mulrdx=%rdx,<mulr21=%r13 adc %rdx,%r13 # qhasm: mulrax = b4_stack # asm 1: movq <b4_stack=stack64#17,>mulrax=int64#7 # asm 2: movq <b4_stack=128(%rsp),>mulrax=%rax movq 128(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 40) # asm 1: mulq 40(<qp=int64#4) # asm 2: mulq 40(<qp=%rcx) mulq 40(%rcx) # qhasm: carry? rx4 += mulrax # asm 1: add <mulrax=int64#7,<rx4=int64#14 # asm 2: add <mulrax=%rax,<rx4=%rbx add %rax,%rbx # qhasm: mulr41 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr41=int64#15 # asm 2: adc <mulrdx=%rdx,<mulr41=%rbp adc %rdx,%rbp # qhasm: mulrax = mulx419_stack # asm 1: movq <mulx419_stack=stack64#19,>mulrax=int64#7 # asm 2: movq <mulx419_stack=144(%rsp),>mulrax=%rax movq 144(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 56) # asm 1: mulq 56(<qp=int64#4) # asm 2: mulq 56(<qp=%rcx) mulq 56(%rcx) # qhasm: carry? rx1 += mulrax # asm 1: add <mulrax=int64#7,<rx1=int64#8 # asm 2: add <mulrax=%rax,<rx1=%r10 add %rax,%r10 # qhasm: mulr11 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr11=int64#9 # asm 2: adc <mulrdx=%rdx,<mulr11=%r11 adc %rdx,%r11 # qhasm: mulrax = mulx419_stack # asm 1: movq <mulx419_stack=stack64#19,>mulrax=int64#7 # asm 2: movq <mulx419_stack=144(%rsp),>mulrax=%rax movq 144(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 64) # asm 1: mulq 64(<qp=int64#4) # asm 2: mulq 64(<qp=%rcx) mulq 64(%rcx) # qhasm: carry? rx2 += mulrax # asm 1: add <mulrax=int64#7,<rx2=int64#10 # asm 2: add <mulrax=%rax,<rx2=%r12 add %rax,%r12 # qhasm: mulr21 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr21=int64#11 # asm 2: adc <mulrdx=%rdx,<mulr21=%r13 adc %rdx,%r13 # qhasm: mulrax = mulx419_stack # asm 1: movq <mulx419_stack=stack64#19,>mulrax=int64#7 # asm 2: movq <mulx419_stack=144(%rsp),>mulrax=%rax movq 144(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 72) # asm 1: mulq 72(<qp=int64#4) # asm 2: mulq 72(<qp=%rcx) mulq 72(%rcx) # qhasm: carry? rx3 += mulrax # asm 1: add <mulrax=int64#7,<rx3=int64#12 # asm 2: add <mulrax=%rax,<rx3=%r14 add %rax,%r14 # qhasm: mulr31 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr31=int64#13 # asm 2: adc <mulrdx=%rdx,<mulr31=%r15 adc %rdx,%r15 # qhasm: mulredmask = *(uint64 *) &CRYPTO_SHARED_NAMESPACE(REDMASK51) # asm 1: movq CRYPTO_SHARED_NAMESPACE(REDMASK51),>mulredmask=int64#3 # asm 2: movq CRYPTO_SHARED_NAMESPACE(REDMASK51),>mulredmask=%rdx movq CRYPTO_SHARED_NAMESPACE(REDMASK51)(%rip),%rdx # qhasm: mulr01 = (mulr01.rx0) << 13 # asm 1: shld $13,<rx0=int64#5,<mulr01=int64#6 # asm 2: shld $13,<rx0=%r8,<mulr01=%r9 shld $13,%r8,%r9 # qhasm: rx0 &= mulredmask # asm 1: and <mulredmask=int64#3,<rx0=int64#5 # asm 2: and <mulredmask=%rdx,<rx0=%r8 and %rdx,%r8 # qhasm: mulr11 = (mulr11.rx1) << 13 # asm 1: shld $13,<rx1=int64#8,<mulr11=int64#9 # asm 2: shld $13,<rx1=%r10,<mulr11=%r11 shld $13,%r10,%r11 # qhasm: rx1 &= mulredmask # asm 1: and <mulredmask=int64#3,<rx1=int64#8 # asm 2: and <mulredmask=%rdx,<rx1=%r10 and %rdx,%r10 # qhasm: rx1 += mulr01 # asm 1: add <mulr01=int64#6,<rx1=int64#8 # asm 2: add <mulr01=%r9,<rx1=%r10 add %r9,%r10 # qhasm: mulr21 = (mulr21.rx2) << 13 # asm 1: shld $13,<rx2=int64#10,<mulr21=int64#11 # asm 2: shld $13,<rx2=%r12,<mulr21=%r13 shld $13,%r12,%r13 # qhasm: rx2 &= mulredmask # asm 1: and <mulredmask=int64#3,<rx2=int64#10 # asm 2: and <mulredmask=%rdx,<rx2=%r12 and %rdx,%r12 # qhasm: rx2 += mulr11 # asm 1: add <mulr11=int64#9,<rx2=int64#10 # asm 2: add <mulr11=%r11,<rx2=%r12 add %r11,%r12 # qhasm: mulr31 = (mulr31.rx3) << 13 # asm 1: shld $13,<rx3=int64#12,<mulr31=int64#13 # asm 2: shld $13,<rx3=%r14,<mulr31=%r15 shld $13,%r14,%r15 # qhasm: rx3 &= mulredmask # asm 1: and <mulredmask=int64#3,<rx3=int64#12 # asm 2: and <mulredmask=%rdx,<rx3=%r14 and %rdx,%r14 # qhasm: rx3 += mulr21 # asm 1: add <mulr21=int64#11,<rx3=int64#12 # asm 2: add <mulr21=%r13,<rx3=%r14 add %r13,%r14 # qhasm: mulr41 = (mulr41.rx4) << 13 # asm 1: shld $13,<rx4=int64#14,<mulr41=int64#15 # asm 2: shld $13,<rx4=%rbx,<mulr41=%rbp shld $13,%rbx,%rbp # qhasm: rx4 &= mulredmask # asm 1: and <mulredmask=int64#3,<rx4=int64#14 # asm 2: and <mulredmask=%rdx,<rx4=%rbx and %rdx,%rbx # qhasm: rx4 += mulr31 # asm 1: add <mulr31=int64#13,<rx4=int64#14 # asm 2: add <mulr31=%r15,<rx4=%rbx add %r15,%rbx # qhasm: mulr41 = mulr41 * 19 # asm 1: imulq $19,<mulr41=int64#15,>mulr41=int64#6 # asm 2: imulq $19,<mulr41=%rbp,>mulr41=%r9 imulq $19,%rbp,%r9 # qhasm: rx0 += mulr41 # asm 1: add <mulr41=int64#6,<rx0=int64#5 # asm 2: add <mulr41=%r9,<rx0=%r8 add %r9,%r8 # qhasm: mult = rx0 # asm 1: mov <rx0=int64#5,>mult=int64#6 # asm 2: mov <rx0=%r8,>mult=%r9 mov %r8,%r9 # qhasm: (uint64) mult >>= 51 # asm 1: shr $51,<mult=int64#6 # asm 2: shr $51,<mult=%r9 shr $51,%r9 # qhasm: mult += rx1 # asm 1: add <rx1=int64#8,<mult=int64#6 # asm 2: add <rx1=%r10,<mult=%r9 add %r10,%r9 # qhasm: rx1 = mult # asm 1: mov <mult=int64#6,>rx1=int64#7 # asm 2: mov <mult=%r9,>rx1=%rax mov %r9,%rax # qhasm: (uint64) mult >>= 51 # asm 1: shr $51,<mult=int64#6 # asm 2: shr $51,<mult=%r9 shr $51,%r9 # qhasm: rx0 &= mulredmask # asm 1: and <mulredmask=int64#3,<rx0=int64#5 # asm 2: and <mulredmask=%rdx,<rx0=%r8 and %rdx,%r8 # qhasm: mult += rx2 # asm 1: add <rx2=int64#10,<mult=int64#6 # asm 2: add <rx2=%r12,<mult=%r9 add %r12,%r9 # qhasm: rx2 = mult # asm 1: mov <mult=int64#6,>rx2=int64#8 # asm 2: mov <mult=%r9,>rx2=%r10 mov %r9,%r10 # qhasm: (uint64) mult >>= 51 # asm 1: shr $51,<mult=int64#6 # asm 2: shr $51,<mult=%r9 shr $51,%r9 # qhasm: rx1 &= mulredmask # asm 1: and <mulredmask=int64#3,<rx1=int64#7 # asm 2: and <mulredmask=%rdx,<rx1=%rax and %rdx,%rax # qhasm: mult += rx3 # asm 1: add <rx3=int64#12,<mult=int64#6 # asm 2: add <rx3=%r14,<mult=%r9 add %r14,%r9 # qhasm: rx3 = mult # asm 1: mov <mult=int64#6,>rx3=int64#9 # asm 2: mov <mult=%r9,>rx3=%r11 mov %r9,%r11 # qhasm: (uint64) mult >>= 51 # asm 1: shr $51,<mult=int64#6 # asm 2: shr $51,<mult=%r9 shr $51,%r9 # qhasm: rx2 &= mulredmask # asm 1: and <mulredmask=int64#3,<rx2=int64#8 # asm 2: and <mulredmask=%rdx,<rx2=%r10 and %rdx,%r10 # qhasm: mult += rx4 # asm 1: add <rx4=int64#14,<mult=int64#6 # asm 2: add <rx4=%rbx,<mult=%r9 add %rbx,%r9 # qhasm: rx4 = mult # asm 1: mov <mult=int64#6,>rx4=int64#10 # asm 2: mov <mult=%r9,>rx4=%r12 mov %r9,%r12 # qhasm: (uint64) mult >>= 51 # asm 1: shr $51,<mult=int64#6 # asm 2: shr $51,<mult=%r9 shr $51,%r9 # qhasm: rx3 &= mulredmask # asm 1: and <mulredmask=int64#3,<rx3=int64#9 # asm 2: and <mulredmask=%rdx,<rx3=%r11 and %rdx,%r11 # qhasm: mult *= 19 # asm 1: imulq $19,<mult=int64#6,>mult=int64#6 # asm 2: imulq $19,<mult=%r9,>mult=%r9 imulq $19,%r9,%r9 # qhasm: rx0 += mult # asm 1: add <mult=int64#6,<rx0=int64#5 # asm 2: add <mult=%r9,<rx0=%r8 add %r9,%r8 # qhasm: rx4 &= mulredmask # asm 1: and <mulredmask=int64#3,<rx4=int64#10 # asm 2: and <mulredmask=%rdx,<rx4=%r12 and %rdx,%r12 # qhasm: ry0 = rx0 # asm 1: mov <rx0=int64#5,>ry0=int64#3 # asm 2: mov <rx0=%r8,>ry0=%rdx mov %r8,%rdx # qhasm: ry1 = rx1 # asm 1: mov <rx1=int64#7,>ry1=int64#6 # asm 2: mov <rx1=%rax,>ry1=%r9 mov %rax,%r9 # qhasm: ry2 = rx2 # asm 1: mov <rx2=int64#8,>ry2=int64#11 # asm 2: mov <rx2=%r10,>ry2=%r13 mov %r10,%r13 # qhasm: ry3 = rx3 # asm 1: mov <rx3=int64#9,>ry3=int64#12 # asm 2: mov <rx3=%r11,>ry3=%r14 mov %r11,%r14 # qhasm: ry4 = rx4 # asm 1: mov <rx4=int64#10,>ry4=int64#13 # asm 2: mov <rx4=%r12,>ry4=%r15 mov %r12,%r15 # qhasm: rx0 += *(uint64 *)&CRYPTO_SHARED_NAMESPACE(2P0) # asm 1: add CRYPTO_SHARED_NAMESPACE(2P0),<rx0=int64#5 # asm 2: add CRYPTO_SHARED_NAMESPACE(2P0),<rx0=%r8 add CRYPTO_SHARED_NAMESPACE(2P0)(%rip),%r8 # qhasm: rx1 += *(uint64 *)&CRYPTO_SHARED_NAMESPACE(2P1234) # asm 1: add CRYPTO_SHARED_NAMESPACE(2P1234),<rx1=int64#7 # asm 2: add CRYPTO_SHARED_NAMESPACE(2P1234),<rx1=%rax add CRYPTO_SHARED_NAMESPACE(2P1234)(%rip),%rax # qhasm: rx2 += *(uint64 *)&CRYPTO_SHARED_NAMESPACE(2P1234) # asm 1: add CRYPTO_SHARED_NAMESPACE(2P1234),<rx2=int64#8 # asm 2: add CRYPTO_SHARED_NAMESPACE(2P1234),<rx2=%r10 add CRYPTO_SHARED_NAMESPACE(2P1234)(%rip),%r10 # qhasm: rx3 += *(uint64 *)&CRYPTO_SHARED_NAMESPACE(2P1234) # asm 1: add CRYPTO_SHARED_NAMESPACE(2P1234),<rx3=int64#9 # asm 2: add CRYPTO_SHARED_NAMESPACE(2P1234),<rx3=%r11 add CRYPTO_SHARED_NAMESPACE(2P1234)(%rip),%r11 # qhasm: rx4 += *(uint64 *)&CRYPTO_SHARED_NAMESPACE(2P1234) # asm 1: add CRYPTO_SHARED_NAMESPACE(2P1234),<rx4=int64#10 # asm 2: add CRYPTO_SHARED_NAMESPACE(2P1234),<rx4=%r12 add CRYPTO_SHARED_NAMESPACE(2P1234)(%rip),%r12 # qhasm: ry0 += a0_stack # asm 1: addq <a0_stack=stack64#8,<ry0=int64#3 # asm 2: addq <a0_stack=56(%rsp),<ry0=%rdx addq 56(%rsp),%rdx # qhasm: ry1 += a1_stack # asm 1: addq <a1_stack=stack64#9,<ry1=int64#6 # asm 2: addq <a1_stack=64(%rsp),<ry1=%r9 addq 64(%rsp),%r9 # qhasm: ry2 += a2_stack # asm 1: addq <a2_stack=stack64#10,<ry2=int64#11 # asm 2: addq <a2_stack=72(%rsp),<ry2=%r13 addq 72(%rsp),%r13 # qhasm: ry3 += a3_stack # asm 1: addq <a3_stack=stack64#11,<ry3=int64#12 # asm 2: addq <a3_stack=80(%rsp),<ry3=%r14 addq 80(%rsp),%r14 # qhasm: ry4 += a4_stack # asm 1: addq <a4_stack=stack64#12,<ry4=int64#13 # asm 2: addq <a4_stack=88(%rsp),<ry4=%r15 addq 88(%rsp),%r15 # qhasm: rx0 -= a0_stack # asm 1: subq <a0_stack=stack64#8,<rx0=int64#5 # asm 2: subq <a0_stack=56(%rsp),<rx0=%r8 subq 56(%rsp),%r8 # qhasm: rx1 -= a1_stack # asm 1: subq <a1_stack=stack64#9,<rx1=int64#7 # asm 2: subq <a1_stack=64(%rsp),<rx1=%rax subq 64(%rsp),%rax # qhasm: rx2 -= a2_stack # asm 1: subq <a2_stack=stack64#10,<rx2=int64#8 # asm 2: subq <a2_stack=72(%rsp),<rx2=%r10 subq 72(%rsp),%r10 # qhasm: rx3 -= a3_stack # asm 1: subq <a3_stack=stack64#11,<rx3=int64#9 # asm 2: subq <a3_stack=80(%rsp),<rx3=%r11 subq 80(%rsp),%r11 # qhasm: rx4 -= a4_stack # asm 1: subq <a4_stack=stack64#12,<rx4=int64#10 # asm 2: subq <a4_stack=88(%rsp),<rx4=%r12 subq 88(%rsp),%r12 # qhasm: *(uint64 *) (rp + 0) = rx0 # asm 1: movq <rx0=int64#5,0(<rp=int64#1) # asm 2: movq <rx0=%r8,0(<rp=%rdi) movq %r8,0(%rdi) # qhasm: *(uint64 *) (rp + 8) = rx1 # asm 1: movq <rx1=int64#7,8(<rp=int64#1) # asm 2: movq <rx1=%rax,8(<rp=%rdi) movq %rax,8(%rdi) # qhasm: *(uint64 *) (rp + 16) = rx2 # asm 1: movq <rx2=int64#8,16(<rp=int64#1) # asm 2: movq <rx2=%r10,16(<rp=%rdi) movq %r10,16(%rdi) # qhasm: *(uint64 *) (rp + 24) = rx3 # asm 1: movq <rx3=int64#9,24(<rp=int64#1) # asm 2: movq <rx3=%r11,24(<rp=%rdi) movq %r11,24(%rdi) # qhasm: *(uint64 *) (rp + 32) = rx4 # asm 1: movq <rx4=int64#10,32(<rp=int64#1) # asm 2: movq <rx4=%r12,32(<rp=%rdi) movq %r12,32(%rdi) # qhasm: *(uint64 *) (rp + 80) = ry0 # asm 1: movq <ry0=int64#3,80(<rp=int64#1) # asm 2: movq <ry0=%rdx,80(<rp=%rdi) movq %rdx,80(%rdi) # qhasm: *(uint64 *) (rp + 88) = ry1 # asm 1: movq <ry1=int64#6,88(<rp=int64#1) # asm 2: movq <ry1=%r9,88(<rp=%rdi) movq %r9,88(%rdi) # qhasm: *(uint64 *) (rp + 96) = ry2 # asm 1: movq <ry2=int64#11,96(<rp=int64#1) # asm 2: movq <ry2=%r13,96(<rp=%rdi) movq %r13,96(%rdi) # qhasm: *(uint64 *) (rp + 104) = ry3 # asm 1: movq <ry3=int64#12,104(<rp=int64#1) # asm 2: movq <ry3=%r14,104(<rp=%rdi) movq %r14,104(%rdi) # qhasm: *(uint64 *) (rp + 112) = ry4 # asm 1: movq <ry4=int64#13,112(<rp=int64#1) # asm 2: movq <ry4=%r15,112(<rp=%rdi) movq %r15,112(%rdi) # qhasm: mulrax = *(uint64 *)(pp + 144) # asm 1: movq 144(<pp=int64#2),>mulrax=int64#3 # asm 2: movq 144(<pp=%rsi),>mulrax=%rdx movq 144(%rsi),%rdx # qhasm: mulrax *= 19 # asm 1: imulq $19,<mulrax=int64#3,>mulrax=int64#7 # asm 2: imulq $19,<mulrax=%rdx,>mulrax=%rax imulq $19,%rdx,%rax # qhasm: mulx319_stack = mulrax # asm 1: movq <mulrax=int64#7,>mulx319_stack=stack64#8 # asm 2: movq <mulrax=%rax,>mulx319_stack=56(%rsp) movq %rax,56(%rsp) # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 136) # asm 1: mulq 136(<qp=int64#4) # asm 2: mulq 136(<qp=%rcx) mulq 136(%rcx) # qhasm: c0 = mulrax # asm 1: mov <mulrax=int64#7,>c0=int64#5 # asm 2: mov <mulrax=%rax,>c0=%r8 mov %rax,%r8 # qhasm: mulr01 = mulrdx # asm 1: mov <mulrdx=int64#3,>mulr01=int64#6 # asm 2: mov <mulrdx=%rdx,>mulr01=%r9 mov %rdx,%r9 # qhasm: mulrax = *(uint64 *)(pp + 152) # asm 1: movq 152(<pp=int64#2),>mulrax=int64#3 # asm 2: movq 152(<pp=%rsi),>mulrax=%rdx movq 152(%rsi),%rdx # qhasm: mulrax *= 19 # asm 1: imulq $19,<mulrax=int64#3,>mulrax=int64#7 # asm 2: imulq $19,<mulrax=%rdx,>mulrax=%rax imulq $19,%rdx,%rax # qhasm: mulx419_stack = mulrax # asm 1: movq <mulrax=int64#7,>mulx419_stack=stack64#9 # asm 2: movq <mulrax=%rax,>mulx419_stack=64(%rsp) movq %rax,64(%rsp) # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 128) # asm 1: mulq 128(<qp=int64#4) # asm 2: mulq 128(<qp=%rcx) mulq 128(%rcx) # qhasm: carry? c0 += mulrax # asm 1: add <mulrax=int64#7,<c0=int64#5 # asm 2: add <mulrax=%rax,<c0=%r8 add %rax,%r8 # qhasm: mulr01 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr01=int64#6 # asm 2: adc <mulrdx=%rdx,<mulr01=%r9 adc %rdx,%r9 # qhasm: mulrax = *(uint64 *)(pp + 120) # asm 1: movq 120(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 120(<pp=%rsi),>mulrax=%rax movq 120(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 120) # asm 1: mulq 120(<qp=int64#4) # asm 2: mulq 120(<qp=%rcx) mulq 120(%rcx) # qhasm: carry? c0 += mulrax # asm 1: add <mulrax=int64#7,<c0=int64#5 # asm 2: add <mulrax=%rax,<c0=%r8 add %rax,%r8 # qhasm: mulr01 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr01=int64#6 # asm 2: adc <mulrdx=%rdx,<mulr01=%r9 adc %rdx,%r9 # qhasm: mulrax = *(uint64 *)(pp + 120) # asm 1: movq 120(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 120(<pp=%rsi),>mulrax=%rax movq 120(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 128) # asm 1: mulq 128(<qp=int64#4) # asm 2: mulq 128(<qp=%rcx) mulq 128(%rcx) # qhasm: c1 = mulrax # asm 1: mov <mulrax=int64#7,>c1=int64#8 # asm 2: mov <mulrax=%rax,>c1=%r10 mov %rax,%r10 # qhasm: mulr11 = mulrdx # asm 1: mov <mulrdx=int64#3,>mulr11=int64#9 # asm 2: mov <mulrdx=%rdx,>mulr11=%r11 mov %rdx,%r11 # qhasm: mulrax = *(uint64 *)(pp + 120) # asm 1: movq 120(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 120(<pp=%rsi),>mulrax=%rax movq 120(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 136) # asm 1: mulq 136(<qp=int64#4) # asm 2: mulq 136(<qp=%rcx) mulq 136(%rcx) # qhasm: c2 = mulrax # asm 1: mov <mulrax=int64#7,>c2=int64#10 # asm 2: mov <mulrax=%rax,>c2=%r12 mov %rax,%r12 # qhasm: mulr21 = mulrdx # asm 1: mov <mulrdx=int64#3,>mulr21=int64#11 # asm 2: mov <mulrdx=%rdx,>mulr21=%r13 mov %rdx,%r13 # qhasm: mulrax = *(uint64 *)(pp + 120) # asm 1: movq 120(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 120(<pp=%rsi),>mulrax=%rax movq 120(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 144) # asm 1: mulq 144(<qp=int64#4) # asm 2: mulq 144(<qp=%rcx) mulq 144(%rcx) # qhasm: c3 = mulrax # asm 1: mov <mulrax=int64#7,>c3=int64#12 # asm 2: mov <mulrax=%rax,>c3=%r14 mov %rax,%r14 # qhasm: mulr31 = mulrdx # asm 1: mov <mulrdx=int64#3,>mulr31=int64#13 # asm 2: mov <mulrdx=%rdx,>mulr31=%r15 mov %rdx,%r15 # qhasm: mulrax = *(uint64 *)(pp + 120) # asm 1: movq 120(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 120(<pp=%rsi),>mulrax=%rax movq 120(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 152) # asm 1: mulq 152(<qp=int64#4) # asm 2: mulq 152(<qp=%rcx) mulq 152(%rcx) # qhasm: c4 = mulrax # asm 1: mov <mulrax=int64#7,>c4=int64#14 # asm 2: mov <mulrax=%rax,>c4=%rbx mov %rax,%rbx # qhasm: mulr41 = mulrdx # asm 1: mov <mulrdx=int64#3,>mulr41=int64#15 # asm 2: mov <mulrdx=%rdx,>mulr41=%rbp mov %rdx,%rbp # qhasm: mulrax = *(uint64 *)(pp + 128) # asm 1: movq 128(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 128(<pp=%rsi),>mulrax=%rax movq 128(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 120) # asm 1: mulq 120(<qp=int64#4) # asm 2: mulq 120(<qp=%rcx) mulq 120(%rcx) # qhasm: carry? c1 += mulrax # asm 1: add <mulrax=int64#7,<c1=int64#8 # asm 2: add <mulrax=%rax,<c1=%r10 add %rax,%r10 # qhasm: mulr11 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr11=int64#9 # asm 2: adc <mulrdx=%rdx,<mulr11=%r11 adc %rdx,%r11 # qhasm: mulrax = *(uint64 *)(pp + 128) # asm 1: movq 128(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 128(<pp=%rsi),>mulrax=%rax movq 128(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 128) # asm 1: mulq 128(<qp=int64#4) # asm 2: mulq 128(<qp=%rcx) mulq 128(%rcx) # qhasm: carry? c2 += mulrax # asm 1: add <mulrax=int64#7,<c2=int64#10 # asm 2: add <mulrax=%rax,<c2=%r12 add %rax,%r12 # qhasm: mulr21 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr21=int64#11 # asm 2: adc <mulrdx=%rdx,<mulr21=%r13 adc %rdx,%r13 # qhasm: mulrax = *(uint64 *)(pp + 128) # asm 1: movq 128(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 128(<pp=%rsi),>mulrax=%rax movq 128(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 136) # asm 1: mulq 136(<qp=int64#4) # asm 2: mulq 136(<qp=%rcx) mulq 136(%rcx) # qhasm: carry? c3 += mulrax # asm 1: add <mulrax=int64#7,<c3=int64#12 # asm 2: add <mulrax=%rax,<c3=%r14 add %rax,%r14 # qhasm: mulr31 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr31=int64#13 # asm 2: adc <mulrdx=%rdx,<mulr31=%r15 adc %rdx,%r15 # qhasm: mulrax = *(uint64 *)(pp + 128) # asm 1: movq 128(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 128(<pp=%rsi),>mulrax=%rax movq 128(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 144) # asm 1: mulq 144(<qp=int64#4) # asm 2: mulq 144(<qp=%rcx) mulq 144(%rcx) # qhasm: carry? c4 += mulrax # asm 1: add <mulrax=int64#7,<c4=int64#14 # asm 2: add <mulrax=%rax,<c4=%rbx add %rax,%rbx # qhasm: mulr41 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr41=int64#15 # asm 2: adc <mulrdx=%rdx,<mulr41=%rbp adc %rdx,%rbp # qhasm: mulrax = *(uint64 *)(pp + 128) # asm 1: movq 128(<pp=int64#2),>mulrax=int64#3 # asm 2: movq 128(<pp=%rsi),>mulrax=%rdx movq 128(%rsi),%rdx # qhasm: mulrax *= 19 # asm 1: imulq $19,<mulrax=int64#3,>mulrax=int64#7 # asm 2: imulq $19,<mulrax=%rdx,>mulrax=%rax imulq $19,%rdx,%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 152) # asm 1: mulq 152(<qp=int64#4) # asm 2: mulq 152(<qp=%rcx) mulq 152(%rcx) # qhasm: carry? c0 += mulrax # asm 1: add <mulrax=int64#7,<c0=int64#5 # asm 2: add <mulrax=%rax,<c0=%r8 add %rax,%r8 # qhasm: mulr01 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr01=int64#6 # asm 2: adc <mulrdx=%rdx,<mulr01=%r9 adc %rdx,%r9 # qhasm: mulrax = *(uint64 *)(pp + 136) # asm 1: movq 136(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 136(<pp=%rsi),>mulrax=%rax movq 136(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 120) # asm 1: mulq 120(<qp=int64#4) # asm 2: mulq 120(<qp=%rcx) mulq 120(%rcx) # qhasm: carry? c2 += mulrax # asm 1: add <mulrax=int64#7,<c2=int64#10 # asm 2: add <mulrax=%rax,<c2=%r12 add %rax,%r12 # qhasm: mulr21 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr21=int64#11 # asm 2: adc <mulrdx=%rdx,<mulr21=%r13 adc %rdx,%r13 # qhasm: mulrax = *(uint64 *)(pp + 136) # asm 1: movq 136(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 136(<pp=%rsi),>mulrax=%rax movq 136(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 128) # asm 1: mulq 128(<qp=int64#4) # asm 2: mulq 128(<qp=%rcx) mulq 128(%rcx) # qhasm: carry? c3 += mulrax # asm 1: add <mulrax=int64#7,<c3=int64#12 # asm 2: add <mulrax=%rax,<c3=%r14 add %rax,%r14 # qhasm: mulr31 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr31=int64#13 # asm 2: adc <mulrdx=%rdx,<mulr31=%r15 adc %rdx,%r15 # qhasm: mulrax = *(uint64 *)(pp + 136) # asm 1: movq 136(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 136(<pp=%rsi),>mulrax=%rax movq 136(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 136) # asm 1: mulq 136(<qp=int64#4) # asm 2: mulq 136(<qp=%rcx) mulq 136(%rcx) # qhasm: carry? c4 += mulrax # asm 1: add <mulrax=int64#7,<c4=int64#14 # asm 2: add <mulrax=%rax,<c4=%rbx add %rax,%rbx # qhasm: mulr41 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr41=int64#15 # asm 2: adc <mulrdx=%rdx,<mulr41=%rbp adc %rdx,%rbp # qhasm: mulrax = *(uint64 *)(pp + 136) # asm 1: movq 136(<pp=int64#2),>mulrax=int64#3 # asm 2: movq 136(<pp=%rsi),>mulrax=%rdx movq 136(%rsi),%rdx # qhasm: mulrax *= 19 # asm 1: imulq $19,<mulrax=int64#3,>mulrax=int64#7 # asm 2: imulq $19,<mulrax=%rdx,>mulrax=%rax imulq $19,%rdx,%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 144) # asm 1: mulq 144(<qp=int64#4) # asm 2: mulq 144(<qp=%rcx) mulq 144(%rcx) # qhasm: carry? c0 += mulrax # asm 1: add <mulrax=int64#7,<c0=int64#5 # asm 2: add <mulrax=%rax,<c0=%r8 add %rax,%r8 # qhasm: mulr01 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr01=int64#6 # asm 2: adc <mulrdx=%rdx,<mulr01=%r9 adc %rdx,%r9 # qhasm: mulrax = *(uint64 *)(pp + 136) # asm 1: movq 136(<pp=int64#2),>mulrax=int64#3 # asm 2: movq 136(<pp=%rsi),>mulrax=%rdx movq 136(%rsi),%rdx # qhasm: mulrax *= 19 # asm 1: imulq $19,<mulrax=int64#3,>mulrax=int64#7 # asm 2: imulq $19,<mulrax=%rdx,>mulrax=%rax imulq $19,%rdx,%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 152) # asm 1: mulq 152(<qp=int64#4) # asm 2: mulq 152(<qp=%rcx) mulq 152(%rcx) # qhasm: carry? c1 += mulrax # asm 1: add <mulrax=int64#7,<c1=int64#8 # asm 2: add <mulrax=%rax,<c1=%r10 add %rax,%r10 # qhasm: mulr11 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr11=int64#9 # asm 2: adc <mulrdx=%rdx,<mulr11=%r11 adc %rdx,%r11 # qhasm: mulrax = *(uint64 *)(pp + 144) # asm 1: movq 144(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 144(<pp=%rsi),>mulrax=%rax movq 144(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 120) # asm 1: mulq 120(<qp=int64#4) # asm 2: mulq 120(<qp=%rcx) mulq 120(%rcx) # qhasm: carry? c3 += mulrax # asm 1: add <mulrax=int64#7,<c3=int64#12 # asm 2: add <mulrax=%rax,<c3=%r14 add %rax,%r14 # qhasm: mulr31 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr31=int64#13 # asm 2: adc <mulrdx=%rdx,<mulr31=%r15 adc %rdx,%r15 # qhasm: mulrax = *(uint64 *)(pp + 144) # asm 1: movq 144(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 144(<pp=%rsi),>mulrax=%rax movq 144(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 128) # asm 1: mulq 128(<qp=int64#4) # asm 2: mulq 128(<qp=%rcx) mulq 128(%rcx) # qhasm: carry? c4 += mulrax # asm 1: add <mulrax=int64#7,<c4=int64#14 # asm 2: add <mulrax=%rax,<c4=%rbx add %rax,%rbx # qhasm: mulr41 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr41=int64#15 # asm 2: adc <mulrdx=%rdx,<mulr41=%rbp adc %rdx,%rbp # qhasm: mulrax = mulx319_stack # asm 1: movq <mulx319_stack=stack64#8,>mulrax=int64#7 # asm 2: movq <mulx319_stack=56(%rsp),>mulrax=%rax movq 56(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 144) # asm 1: mulq 144(<qp=int64#4) # asm 2: mulq 144(<qp=%rcx) mulq 144(%rcx) # qhasm: carry? c1 += mulrax # asm 1: add <mulrax=int64#7,<c1=int64#8 # asm 2: add <mulrax=%rax,<c1=%r10 add %rax,%r10 # qhasm: mulr11 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr11=int64#9 # asm 2: adc <mulrdx=%rdx,<mulr11=%r11 adc %rdx,%r11 # qhasm: mulrax = mulx319_stack # asm 1: movq <mulx319_stack=stack64#8,>mulrax=int64#7 # asm 2: movq <mulx319_stack=56(%rsp),>mulrax=%rax movq 56(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 152) # asm 1: mulq 152(<qp=int64#4) # asm 2: mulq 152(<qp=%rcx) mulq 152(%rcx) # qhasm: carry? c2 += mulrax # asm 1: add <mulrax=int64#7,<c2=int64#10 # asm 2: add <mulrax=%rax,<c2=%r12 add %rax,%r12 # qhasm: mulr21 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr21=int64#11 # asm 2: adc <mulrdx=%rdx,<mulr21=%r13 adc %rdx,%r13 # qhasm: mulrax = *(uint64 *)(pp + 152) # asm 1: movq 152(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 152(<pp=%rsi),>mulrax=%rax movq 152(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 120) # asm 1: mulq 120(<qp=int64#4) # asm 2: mulq 120(<qp=%rcx) mulq 120(%rcx) # qhasm: carry? c4 += mulrax # asm 1: add <mulrax=int64#7,<c4=int64#14 # asm 2: add <mulrax=%rax,<c4=%rbx add %rax,%rbx # qhasm: mulr41 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr41=int64#15 # asm 2: adc <mulrdx=%rdx,<mulr41=%rbp adc %rdx,%rbp # qhasm: mulrax = mulx419_stack # asm 1: movq <mulx419_stack=stack64#9,>mulrax=int64#7 # asm 2: movq <mulx419_stack=64(%rsp),>mulrax=%rax movq 64(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 136) # asm 1: mulq 136(<qp=int64#4) # asm 2: mulq 136(<qp=%rcx) mulq 136(%rcx) # qhasm: carry? c1 += mulrax # asm 1: add <mulrax=int64#7,<c1=int64#8 # asm 2: add <mulrax=%rax,<c1=%r10 add %rax,%r10 # qhasm: mulr11 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr11=int64#9 # asm 2: adc <mulrdx=%rdx,<mulr11=%r11 adc %rdx,%r11 # qhasm: mulrax = mulx419_stack # asm 1: movq <mulx419_stack=stack64#9,>mulrax=int64#7 # asm 2: movq <mulx419_stack=64(%rsp),>mulrax=%rax movq 64(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 144) # asm 1: mulq 144(<qp=int64#4) # asm 2: mulq 144(<qp=%rcx) mulq 144(%rcx) # qhasm: carry? c2 += mulrax # asm 1: add <mulrax=int64#7,<c2=int64#10 # asm 2: add <mulrax=%rax,<c2=%r12 add %rax,%r12 # qhasm: mulr21 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr21=int64#11 # asm 2: adc <mulrdx=%rdx,<mulr21=%r13 adc %rdx,%r13 # qhasm: mulrax = mulx419_stack # asm 1: movq <mulx419_stack=stack64#9,>mulrax=int64#7 # asm 2: movq <mulx419_stack=64(%rsp),>mulrax=%rax movq 64(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 152) # asm 1: mulq 152(<qp=int64#4) # asm 2: mulq 152(<qp=%rcx) mulq 152(%rcx) # qhasm: carry? c3 += mulrax # asm 1: add <mulrax=int64#7,<c3=int64#12 # asm 2: add <mulrax=%rax,<c3=%r14 add %rax,%r14 # qhasm: mulr31 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr31=int64#13 # asm 2: adc <mulrdx=%rdx,<mulr31=%r15 adc %rdx,%r15 # qhasm: mulredmask = *(uint64 *) &CRYPTO_SHARED_NAMESPACE(REDMASK51) # asm 1: movq CRYPTO_SHARED_NAMESPACE(REDMASK51),>mulredmask=int64#3 # asm 2: movq CRYPTO_SHARED_NAMESPACE(REDMASK51),>mulredmask=%rdx movq CRYPTO_SHARED_NAMESPACE(REDMASK51)(%rip),%rdx # qhasm: mulr01 = (mulr01.c0) << 13 # asm 1: shld $13,<c0=int64#5,<mulr01=int64#6 # asm 2: shld $13,<c0=%r8,<mulr01=%r9 shld $13,%r8,%r9 # qhasm: c0 &= mulredmask # asm 1: and <mulredmask=int64#3,<c0=int64#5 # asm 2: and <mulredmask=%rdx,<c0=%r8 and %rdx,%r8 # qhasm: mulr11 = (mulr11.c1) << 13 # asm 1: shld $13,<c1=int64#8,<mulr11=int64#9 # asm 2: shld $13,<c1=%r10,<mulr11=%r11 shld $13,%r10,%r11 # qhasm: c1 &= mulredmask # asm 1: and <mulredmask=int64#3,<c1=int64#8 # asm 2: and <mulredmask=%rdx,<c1=%r10 and %rdx,%r10 # qhasm: c1 += mulr01 # asm 1: add <mulr01=int64#6,<c1=int64#8 # asm 2: add <mulr01=%r9,<c1=%r10 add %r9,%r10 # qhasm: mulr21 = (mulr21.c2) << 13 # asm 1: shld $13,<c2=int64#10,<mulr21=int64#11 # asm 2: shld $13,<c2=%r12,<mulr21=%r13 shld $13,%r12,%r13 # qhasm: c2 &= mulredmask # asm 1: and <mulredmask=int64#3,<c2=int64#10 # asm 2: and <mulredmask=%rdx,<c2=%r12 and %rdx,%r12 # qhasm: c2 += mulr11 # asm 1: add <mulr11=int64#9,<c2=int64#10 # asm 2: add <mulr11=%r11,<c2=%r12 add %r11,%r12 # qhasm: mulr31 = (mulr31.c3) << 13 # asm 1: shld $13,<c3=int64#12,<mulr31=int64#13 # asm 2: shld $13,<c3=%r14,<mulr31=%r15 shld $13,%r14,%r15 # qhasm: c3 &= mulredmask # asm 1: and <mulredmask=int64#3,<c3=int64#12 # asm 2: and <mulredmask=%rdx,<c3=%r14 and %rdx,%r14 # qhasm: c3 += mulr21 # asm 1: add <mulr21=int64#11,<c3=int64#12 # asm 2: add <mulr21=%r13,<c3=%r14 add %r13,%r14 # qhasm: mulr41 = (mulr41.c4) << 13 # asm 1: shld $13,<c4=int64#14,<mulr41=int64#15 # asm 2: shld $13,<c4=%rbx,<mulr41=%rbp shld $13,%rbx,%rbp # qhasm: c4 &= mulredmask # asm 1: and <mulredmask=int64#3,<c4=int64#14 # asm 2: and <mulredmask=%rdx,<c4=%rbx and %rdx,%rbx # qhasm: c4 += mulr31 # asm 1: add <mulr31=int64#13,<c4=int64#14 # asm 2: add <mulr31=%r15,<c4=%rbx add %r15,%rbx # qhasm: mulr41 = mulr41 * 19 # asm 1: imulq $19,<mulr41=int64#15,>mulr41=int64#6 # asm 2: imulq $19,<mulr41=%rbp,>mulr41=%r9 imulq $19,%rbp,%r9 # qhasm: c0 += mulr41 # asm 1: add <mulr41=int64#6,<c0=int64#5 # asm 2: add <mulr41=%r9,<c0=%r8 add %r9,%r8 # qhasm: mult = c0 # asm 1: mov <c0=int64#5,>mult=int64#6 # asm 2: mov <c0=%r8,>mult=%r9 mov %r8,%r9 # qhasm: (uint64) mult >>= 51 # asm 1: shr $51,<mult=int64#6 # asm 2: shr $51,<mult=%r9 shr $51,%r9 # qhasm: mult += c1 # asm 1: add <c1=int64#8,<mult=int64#6 # asm 2: add <c1=%r10,<mult=%r9 add %r10,%r9 # qhasm: c1 = mult # asm 1: mov <mult=int64#6,>c1=int64#7 # asm 2: mov <mult=%r9,>c1=%rax mov %r9,%rax # qhasm: (uint64) mult >>= 51 # asm 1: shr $51,<mult=int64#6 # asm 2: shr $51,<mult=%r9 shr $51,%r9 # qhasm: c0 &= mulredmask # asm 1: and <mulredmask=int64#3,<c0=int64#5 # asm 2: and <mulredmask=%rdx,<c0=%r8 and %rdx,%r8 # qhasm: mult += c2 # asm 1: add <c2=int64#10,<mult=int64#6 # asm 2: add <c2=%r12,<mult=%r9 add %r12,%r9 # qhasm: c2 = mult # asm 1: mov <mult=int64#6,>c2=int64#8 # asm 2: mov <mult=%r9,>c2=%r10 mov %r9,%r10 # qhasm: (uint64) mult >>= 51 # asm 1: shr $51,<mult=int64#6 # asm 2: shr $51,<mult=%r9 shr $51,%r9 # qhasm: c1 &= mulredmask # asm 1: and <mulredmask=int64#3,<c1=int64#7 # asm 2: and <mulredmask=%rdx,<c1=%rax and %rdx,%rax # qhasm: mult += c3 # asm 1: add <c3=int64#12,<mult=int64#6 # asm 2: add <c3=%r14,<mult=%r9 add %r14,%r9 # qhasm: c3 = mult # asm 1: mov <mult=int64#6,>c3=int64#9 # asm 2: mov <mult=%r9,>c3=%r11 mov %r9,%r11 # qhasm: (uint64) mult >>= 51 # asm 1: shr $51,<mult=int64#6 # asm 2: shr $51,<mult=%r9 shr $51,%r9 # qhasm: c2 &= mulredmask # asm 1: and <mulredmask=int64#3,<c2=int64#8 # asm 2: and <mulredmask=%rdx,<c2=%r10 and %rdx,%r10 # qhasm: mult += c4 # asm 1: add <c4=int64#14,<mult=int64#6 # asm 2: add <c4=%rbx,<mult=%r9 add %rbx,%r9 # qhasm: c4 = mult # asm 1: mov <mult=int64#6,>c4=int64#10 # asm 2: mov <mult=%r9,>c4=%r12 mov %r9,%r12 # qhasm: (uint64) mult >>= 51 # asm 1: shr $51,<mult=int64#6 # asm 2: shr $51,<mult=%r9 shr $51,%r9 # qhasm: c3 &= mulredmask # asm 1: and <mulredmask=int64#3,<c3=int64#9 # asm 2: and <mulredmask=%rdx,<c3=%r11 and %rdx,%r11 # qhasm: mult *= 19 # asm 1: imulq $19,<mult=int64#6,>mult=int64#6 # asm 2: imulq $19,<mult=%r9,>mult=%r9 imulq $19,%r9,%r9 # qhasm: c0 += mult # asm 1: add <mult=int64#6,<c0=int64#5 # asm 2: add <mult=%r9,<c0=%r8 add %r9,%r8 # qhasm: c4 &= mulredmask # asm 1: and <mulredmask=int64#3,<c4=int64#10 # asm 2: and <mulredmask=%rdx,<c4=%r12 and %rdx,%r12 # qhasm: c0_stack = c0 # asm 1: movq <c0=int64#5,>c0_stack=stack64#8 # asm 2: movq <c0=%r8,>c0_stack=56(%rsp) movq %r8,56(%rsp) # qhasm: c1_stack = c1 # asm 1: movq <c1=int64#7,>c1_stack=stack64#9 # asm 2: movq <c1=%rax,>c1_stack=64(%rsp) movq %rax,64(%rsp) # qhasm: c2_stack = c2 # asm 1: movq <c2=int64#8,>c2_stack=stack64#10 # asm 2: movq <c2=%r10,>c2_stack=72(%rsp) movq %r10,72(%rsp) # qhasm: c3_stack = c3 # asm 1: movq <c3=int64#9,>c3_stack=stack64#11 # asm 2: movq <c3=%r11,>c3_stack=80(%rsp) movq %r11,80(%rsp) # qhasm: c4_stack = c4 # asm 1: movq <c4=int64#10,>c4_stack=stack64#12 # asm 2: movq <c4=%r12,>c4_stack=88(%rsp) movq %r12,88(%rsp) # qhasm: mulrax = *(uint64 *)(pp + 104) # asm 1: movq 104(<pp=int64#2),>mulrax=int64#3 # asm 2: movq 104(<pp=%rsi),>mulrax=%rdx movq 104(%rsi),%rdx # qhasm: mulrax *= 19 # asm 1: imulq $19,<mulrax=int64#3,>mulrax=int64#7 # asm 2: imulq $19,<mulrax=%rdx,>mulrax=%rax imulq $19,%rdx,%rax # qhasm: mulx319_stack = mulrax # asm 1: movq <mulrax=int64#7,>mulx319_stack=stack64#13 # asm 2: movq <mulrax=%rax,>mulx319_stack=96(%rsp) movq %rax,96(%rsp) # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 96) # asm 1: mulq 96(<qp=int64#4) # asm 2: mulq 96(<qp=%rcx) mulq 96(%rcx) # qhasm: rt0 = mulrax # asm 1: mov <mulrax=int64#7,>rt0=int64#5 # asm 2: mov <mulrax=%rax,>rt0=%r8 mov %rax,%r8 # qhasm: mulr01 = mulrdx # asm 1: mov <mulrdx=int64#3,>mulr01=int64#6 # asm 2: mov <mulrdx=%rdx,>mulr01=%r9 mov %rdx,%r9 # qhasm: mulrax = *(uint64 *)(pp + 112) # asm 1: movq 112(<pp=int64#2),>mulrax=int64#3 # asm 2: movq 112(<pp=%rsi),>mulrax=%rdx movq 112(%rsi),%rdx # qhasm: mulrax *= 19 # asm 1: imulq $19,<mulrax=int64#3,>mulrax=int64#7 # asm 2: imulq $19,<mulrax=%rdx,>mulrax=%rax imulq $19,%rdx,%rax # qhasm: mulx419_stack = mulrax # asm 1: movq <mulrax=int64#7,>mulx419_stack=stack64#14 # asm 2: movq <mulrax=%rax,>mulx419_stack=104(%rsp) movq %rax,104(%rsp) # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 88) # asm 1: mulq 88(<qp=int64#4) # asm 2: mulq 88(<qp=%rcx) mulq 88(%rcx) # qhasm: carry? rt0 += mulrax # asm 1: add <mulrax=int64#7,<rt0=int64#5 # asm 2: add <mulrax=%rax,<rt0=%r8 add %rax,%r8 # qhasm: mulr01 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr01=int64#6 # asm 2: adc <mulrdx=%rdx,<mulr01=%r9 adc %rdx,%r9 # qhasm: mulrax = *(uint64 *)(pp + 80) # asm 1: movq 80(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 80(<pp=%rsi),>mulrax=%rax movq 80(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 80) # asm 1: mulq 80(<qp=int64#4) # asm 2: mulq 80(<qp=%rcx) mulq 80(%rcx) # qhasm: carry? rt0 += mulrax # asm 1: add <mulrax=int64#7,<rt0=int64#5 # asm 2: add <mulrax=%rax,<rt0=%r8 add %rax,%r8 # qhasm: mulr01 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr01=int64#6 # asm 2: adc <mulrdx=%rdx,<mulr01=%r9 adc %rdx,%r9 # qhasm: mulrax = *(uint64 *)(pp + 80) # asm 1: movq 80(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 80(<pp=%rsi),>mulrax=%rax movq 80(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 88) # asm 1: mulq 88(<qp=int64#4) # asm 2: mulq 88(<qp=%rcx) mulq 88(%rcx) # qhasm: rt1 = mulrax # asm 1: mov <mulrax=int64#7,>rt1=int64#8 # asm 2: mov <mulrax=%rax,>rt1=%r10 mov %rax,%r10 # qhasm: mulr11 = mulrdx # asm 1: mov <mulrdx=int64#3,>mulr11=int64#9 # asm 2: mov <mulrdx=%rdx,>mulr11=%r11 mov %rdx,%r11 # qhasm: mulrax = *(uint64 *)(pp + 80) # asm 1: movq 80(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 80(<pp=%rsi),>mulrax=%rax movq 80(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 96) # asm 1: mulq 96(<qp=int64#4) # asm 2: mulq 96(<qp=%rcx) mulq 96(%rcx) # qhasm: rt2 = mulrax # asm 1: mov <mulrax=int64#7,>rt2=int64#10 # asm 2: mov <mulrax=%rax,>rt2=%r12 mov %rax,%r12 # qhasm: mulr21 = mulrdx # asm 1: mov <mulrdx=int64#3,>mulr21=int64#11 # asm 2: mov <mulrdx=%rdx,>mulr21=%r13 mov %rdx,%r13 # qhasm: mulrax = *(uint64 *)(pp + 80) # asm 1: movq 80(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 80(<pp=%rsi),>mulrax=%rax movq 80(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 104) # asm 1: mulq 104(<qp=int64#4) # asm 2: mulq 104(<qp=%rcx) mulq 104(%rcx) # qhasm: rt3 = mulrax # asm 1: mov <mulrax=int64#7,>rt3=int64#12 # asm 2: mov <mulrax=%rax,>rt3=%r14 mov %rax,%r14 # qhasm: mulr31 = mulrdx # asm 1: mov <mulrdx=int64#3,>mulr31=int64#13 # asm 2: mov <mulrdx=%rdx,>mulr31=%r15 mov %rdx,%r15 # qhasm: mulrax = *(uint64 *)(pp + 80) # asm 1: movq 80(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 80(<pp=%rsi),>mulrax=%rax movq 80(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 112) # asm 1: mulq 112(<qp=int64#4) # asm 2: mulq 112(<qp=%rcx) mulq 112(%rcx) # qhasm: rt4 = mulrax # asm 1: mov <mulrax=int64#7,>rt4=int64#14 # asm 2: mov <mulrax=%rax,>rt4=%rbx mov %rax,%rbx # qhasm: mulr41 = mulrdx # asm 1: mov <mulrdx=int64#3,>mulr41=int64#15 # asm 2: mov <mulrdx=%rdx,>mulr41=%rbp mov %rdx,%rbp # qhasm: mulrax = *(uint64 *)(pp + 88) # asm 1: movq 88(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 88(<pp=%rsi),>mulrax=%rax movq 88(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 80) # asm 1: mulq 80(<qp=int64#4) # asm 2: mulq 80(<qp=%rcx) mulq 80(%rcx) # qhasm: carry? rt1 += mulrax # asm 1: add <mulrax=int64#7,<rt1=int64#8 # asm 2: add <mulrax=%rax,<rt1=%r10 add %rax,%r10 # qhasm: mulr11 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr11=int64#9 # asm 2: adc <mulrdx=%rdx,<mulr11=%r11 adc %rdx,%r11 # qhasm: mulrax = *(uint64 *)(pp + 88) # asm 1: movq 88(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 88(<pp=%rsi),>mulrax=%rax movq 88(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 88) # asm 1: mulq 88(<qp=int64#4) # asm 2: mulq 88(<qp=%rcx) mulq 88(%rcx) # qhasm: carry? rt2 += mulrax # asm 1: add <mulrax=int64#7,<rt2=int64#10 # asm 2: add <mulrax=%rax,<rt2=%r12 add %rax,%r12 # qhasm: mulr21 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr21=int64#11 # asm 2: adc <mulrdx=%rdx,<mulr21=%r13 adc %rdx,%r13 # qhasm: mulrax = *(uint64 *)(pp + 88) # asm 1: movq 88(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 88(<pp=%rsi),>mulrax=%rax movq 88(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 96) # asm 1: mulq 96(<qp=int64#4) # asm 2: mulq 96(<qp=%rcx) mulq 96(%rcx) # qhasm: carry? rt3 += mulrax # asm 1: add <mulrax=int64#7,<rt3=int64#12 # asm 2: add <mulrax=%rax,<rt3=%r14 add %rax,%r14 # qhasm: mulr31 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr31=int64#13 # asm 2: adc <mulrdx=%rdx,<mulr31=%r15 adc %rdx,%r15 # qhasm: mulrax = *(uint64 *)(pp + 88) # asm 1: movq 88(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 88(<pp=%rsi),>mulrax=%rax movq 88(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 104) # asm 1: mulq 104(<qp=int64#4) # asm 2: mulq 104(<qp=%rcx) mulq 104(%rcx) # qhasm: carry? rt4 += mulrax # asm 1: add <mulrax=int64#7,<rt4=int64#14 # asm 2: add <mulrax=%rax,<rt4=%rbx add %rax,%rbx # qhasm: mulr41 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr41=int64#15 # asm 2: adc <mulrdx=%rdx,<mulr41=%rbp adc %rdx,%rbp # qhasm: mulrax = *(uint64 *)(pp + 88) # asm 1: movq 88(<pp=int64#2),>mulrax=int64#3 # asm 2: movq 88(<pp=%rsi),>mulrax=%rdx movq 88(%rsi),%rdx # qhasm: mulrax *= 19 # asm 1: imulq $19,<mulrax=int64#3,>mulrax=int64#7 # asm 2: imulq $19,<mulrax=%rdx,>mulrax=%rax imulq $19,%rdx,%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 112) # asm 1: mulq 112(<qp=int64#4) # asm 2: mulq 112(<qp=%rcx) mulq 112(%rcx) # qhasm: carry? rt0 += mulrax # asm 1: add <mulrax=int64#7,<rt0=int64#5 # asm 2: add <mulrax=%rax,<rt0=%r8 add %rax,%r8 # qhasm: mulr01 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr01=int64#6 # asm 2: adc <mulrdx=%rdx,<mulr01=%r9 adc %rdx,%r9 # qhasm: mulrax = *(uint64 *)(pp + 96) # asm 1: movq 96(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 96(<pp=%rsi),>mulrax=%rax movq 96(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 80) # asm 1: mulq 80(<qp=int64#4) # asm 2: mulq 80(<qp=%rcx) mulq 80(%rcx) # qhasm: carry? rt2 += mulrax # asm 1: add <mulrax=int64#7,<rt2=int64#10 # asm 2: add <mulrax=%rax,<rt2=%r12 add %rax,%r12 # qhasm: mulr21 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr21=int64#11 # asm 2: adc <mulrdx=%rdx,<mulr21=%r13 adc %rdx,%r13 # qhasm: mulrax = *(uint64 *)(pp + 96) # asm 1: movq 96(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 96(<pp=%rsi),>mulrax=%rax movq 96(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 88) # asm 1: mulq 88(<qp=int64#4) # asm 2: mulq 88(<qp=%rcx) mulq 88(%rcx) # qhasm: carry? rt3 += mulrax # asm 1: add <mulrax=int64#7,<rt3=int64#12 # asm 2: add <mulrax=%rax,<rt3=%r14 add %rax,%r14 # qhasm: mulr31 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr31=int64#13 # asm 2: adc <mulrdx=%rdx,<mulr31=%r15 adc %rdx,%r15 # qhasm: mulrax = *(uint64 *)(pp + 96) # asm 1: movq 96(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 96(<pp=%rsi),>mulrax=%rax movq 96(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 96) # asm 1: mulq 96(<qp=int64#4) # asm 2: mulq 96(<qp=%rcx) mulq 96(%rcx) # qhasm: carry? rt4 += mulrax # asm 1: add <mulrax=int64#7,<rt4=int64#14 # asm 2: add <mulrax=%rax,<rt4=%rbx add %rax,%rbx # qhasm: mulr41 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr41=int64#15 # asm 2: adc <mulrdx=%rdx,<mulr41=%rbp adc %rdx,%rbp # qhasm: mulrax = *(uint64 *)(pp + 96) # asm 1: movq 96(<pp=int64#2),>mulrax=int64#3 # asm 2: movq 96(<pp=%rsi),>mulrax=%rdx movq 96(%rsi),%rdx # qhasm: mulrax *= 19 # asm 1: imulq $19,<mulrax=int64#3,>mulrax=int64#7 # asm 2: imulq $19,<mulrax=%rdx,>mulrax=%rax imulq $19,%rdx,%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 104) # asm 1: mulq 104(<qp=int64#4) # asm 2: mulq 104(<qp=%rcx) mulq 104(%rcx) # qhasm: carry? rt0 += mulrax # asm 1: add <mulrax=int64#7,<rt0=int64#5 # asm 2: add <mulrax=%rax,<rt0=%r8 add %rax,%r8 # qhasm: mulr01 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr01=int64#6 # asm 2: adc <mulrdx=%rdx,<mulr01=%r9 adc %rdx,%r9 # qhasm: mulrax = *(uint64 *)(pp + 96) # asm 1: movq 96(<pp=int64#2),>mulrax=int64#3 # asm 2: movq 96(<pp=%rsi),>mulrax=%rdx movq 96(%rsi),%rdx # qhasm: mulrax *= 19 # asm 1: imulq $19,<mulrax=int64#3,>mulrax=int64#7 # asm 2: imulq $19,<mulrax=%rdx,>mulrax=%rax imulq $19,%rdx,%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 112) # asm 1: mulq 112(<qp=int64#4) # asm 2: mulq 112(<qp=%rcx) mulq 112(%rcx) # qhasm: carry? rt1 += mulrax # asm 1: add <mulrax=int64#7,<rt1=int64#8 # asm 2: add <mulrax=%rax,<rt1=%r10 add %rax,%r10 # qhasm: mulr11 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr11=int64#9 # asm 2: adc <mulrdx=%rdx,<mulr11=%r11 adc %rdx,%r11 # qhasm: mulrax = *(uint64 *)(pp + 104) # asm 1: movq 104(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 104(<pp=%rsi),>mulrax=%rax movq 104(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 80) # asm 1: mulq 80(<qp=int64#4) # asm 2: mulq 80(<qp=%rcx) mulq 80(%rcx) # qhasm: carry? rt3 += mulrax # asm 1: add <mulrax=int64#7,<rt3=int64#12 # asm 2: add <mulrax=%rax,<rt3=%r14 add %rax,%r14 # qhasm: mulr31 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr31=int64#13 # asm 2: adc <mulrdx=%rdx,<mulr31=%r15 adc %rdx,%r15 # qhasm: mulrax = *(uint64 *)(pp + 104) # asm 1: movq 104(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 104(<pp=%rsi),>mulrax=%rax movq 104(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 88) # asm 1: mulq 88(<qp=int64#4) # asm 2: mulq 88(<qp=%rcx) mulq 88(%rcx) # qhasm: carry? rt4 += mulrax # asm 1: add <mulrax=int64#7,<rt4=int64#14 # asm 2: add <mulrax=%rax,<rt4=%rbx add %rax,%rbx # qhasm: mulr41 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr41=int64#15 # asm 2: adc <mulrdx=%rdx,<mulr41=%rbp adc %rdx,%rbp # qhasm: mulrax = mulx319_stack # asm 1: movq <mulx319_stack=stack64#13,>mulrax=int64#7 # asm 2: movq <mulx319_stack=96(%rsp),>mulrax=%rax movq 96(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 104) # asm 1: mulq 104(<qp=int64#4) # asm 2: mulq 104(<qp=%rcx) mulq 104(%rcx) # qhasm: carry? rt1 += mulrax # asm 1: add <mulrax=int64#7,<rt1=int64#8 # asm 2: add <mulrax=%rax,<rt1=%r10 add %rax,%r10 # qhasm: mulr11 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr11=int64#9 # asm 2: adc <mulrdx=%rdx,<mulr11=%r11 adc %rdx,%r11 # qhasm: mulrax = mulx319_stack # asm 1: movq <mulx319_stack=stack64#13,>mulrax=int64#7 # asm 2: movq <mulx319_stack=96(%rsp),>mulrax=%rax movq 96(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 112) # asm 1: mulq 112(<qp=int64#4) # asm 2: mulq 112(<qp=%rcx) mulq 112(%rcx) # qhasm: carry? rt2 += mulrax # asm 1: add <mulrax=int64#7,<rt2=int64#10 # asm 2: add <mulrax=%rax,<rt2=%r12 add %rax,%r12 # qhasm: mulr21 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr21=int64#11 # asm 2: adc <mulrdx=%rdx,<mulr21=%r13 adc %rdx,%r13 # qhasm: mulrax = *(uint64 *)(pp + 112) # asm 1: movq 112(<pp=int64#2),>mulrax=int64#7 # asm 2: movq 112(<pp=%rsi),>mulrax=%rax movq 112(%rsi),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 80) # asm 1: mulq 80(<qp=int64#4) # asm 2: mulq 80(<qp=%rcx) mulq 80(%rcx) # qhasm: carry? rt4 += mulrax # asm 1: add <mulrax=int64#7,<rt4=int64#14 # asm 2: add <mulrax=%rax,<rt4=%rbx add %rax,%rbx # qhasm: mulr41 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr41=int64#15 # asm 2: adc <mulrdx=%rdx,<mulr41=%rbp adc %rdx,%rbp # qhasm: mulrax = mulx419_stack # asm 1: movq <mulx419_stack=stack64#14,>mulrax=int64#7 # asm 2: movq <mulx419_stack=104(%rsp),>mulrax=%rax movq 104(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 96) # asm 1: mulq 96(<qp=int64#4) # asm 2: mulq 96(<qp=%rcx) mulq 96(%rcx) # qhasm: carry? rt1 += mulrax # asm 1: add <mulrax=int64#7,<rt1=int64#8 # asm 2: add <mulrax=%rax,<rt1=%r10 add %rax,%r10 # qhasm: mulr11 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr11=int64#9 # asm 2: adc <mulrdx=%rdx,<mulr11=%r11 adc %rdx,%r11 # qhasm: mulrax = mulx419_stack # asm 1: movq <mulx419_stack=stack64#14,>mulrax=int64#7 # asm 2: movq <mulx419_stack=104(%rsp),>mulrax=%rax movq 104(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 104) # asm 1: mulq 104(<qp=int64#4) # asm 2: mulq 104(<qp=%rcx) mulq 104(%rcx) # qhasm: carry? rt2 += mulrax # asm 1: add <mulrax=int64#7,<rt2=int64#10 # asm 2: add <mulrax=%rax,<rt2=%r12 add %rax,%r12 # qhasm: mulr21 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr21=int64#11 # asm 2: adc <mulrdx=%rdx,<mulr21=%r13 adc %rdx,%r13 # qhasm: mulrax = mulx419_stack # asm 1: movq <mulx419_stack=stack64#14,>mulrax=int64#7 # asm 2: movq <mulx419_stack=104(%rsp),>mulrax=%rax movq 104(%rsp),%rax # qhasm: (uint128) mulrdx mulrax = mulrax * *(uint64 *)(qp + 112) # asm 1: mulq 112(<qp=int64#4) # asm 2: mulq 112(<qp=%rcx) mulq 112(%rcx) # qhasm: carry? rt3 += mulrax # asm 1: add <mulrax=int64#7,<rt3=int64#12 # asm 2: add <mulrax=%rax,<rt3=%r14 add %rax,%r14 # qhasm: mulr31 += mulrdx + carry # asm 1: adc <mulrdx=int64#3,<mulr31=int64#13 # asm 2: adc <mulrdx=%rdx,<mulr31=%r15 adc %rdx,%r15 # qhasm: mulredmask = *(uint64 *) &CRYPTO_SHARED_NAMESPACE(REDMASK51) # asm 1: movq CRYPTO_SHARED_NAMESPACE(REDMASK51),>mulredmask=int64#2 # asm 2: movq CRYPTO_SHARED_NAMESPACE(REDMASK51),>mulredmask=%rsi movq CRYPTO_SHARED_NAMESPACE(REDMASK51)(%rip),%rsi # qhasm: mulr01 = (mulr01.rt0) << 13 # asm 1: shld $13,<rt0=int64#5,<mulr01=int64#6 # asm 2: shld $13,<rt0=%r8,<mulr01=%r9 shld $13,%r8,%r9 # qhasm: rt0 &= mulredmask # asm 1: and <mulredmask=int64#2,<rt0=int64#5 # asm 2: and <mulredmask=%rsi,<rt0=%r8 and %rsi,%r8 # qhasm: mulr11 = (mulr11.rt1) << 13 # asm 1: shld $13,<rt1=int64#8,<mulr11=int64#9 # asm 2: shld $13,<rt1=%r10,<mulr11=%r11 shld $13,%r10,%r11 # qhasm: rt1 &= mulredmask # asm 1: and <mulredmask=int64#2,<rt1=int64#8 # asm 2: and <mulredmask=%rsi,<rt1=%r10 and %rsi,%r10 # qhasm: rt1 += mulr01 # asm 1: add <mulr01=int64#6,<rt1=int64#8 # asm 2: add <mulr01=%r9,<rt1=%r10 add %r9,%r10 # qhasm: mulr21 = (mulr21.rt2) << 13 # asm 1: shld $13,<rt2=int64#10,<mulr21=int64#11 # asm 2: shld $13,<rt2=%r12,<mulr21=%r13 shld $13,%r12,%r13 # qhasm: rt2 &= mulredmask # asm 1: and <mulredmask=int64#2,<rt2=int64#10 # asm 2: and <mulredmask=%rsi,<rt2=%r12 and %rsi,%r12 # qhasm: rt2 += mulr11 # asm 1: add <mulr11=int64#9,<rt2=int64#10 # asm 2: add <mulr11=%r11,<rt2=%r12 add %r11,%r12 # qhasm: mulr31 = (mulr31.rt3) << 13 # asm 1: shld $13,<rt3=int64#12,<mulr31=int64#13 # asm 2: shld $13,<rt3=%r14,<mulr31=%r15 shld $13,%r14,%r15 # qhasm: rt3 &= mulredmask # asm 1: and <mulredmask=int64#2,<rt3=int64#12 # asm 2: and <mulredmask=%rsi,<rt3=%r14 and %rsi,%r14 # qhasm: rt3 += mulr21 # asm 1: add <mulr21=int64#11,<rt3=int64#12 # asm 2: add <mulr21=%r13,<rt3=%r14 add %r13,%r14 # qhasm: mulr41 = (mulr41.rt4) << 13 # asm 1: shld $13,<rt4=int64#14,<mulr41=int64#15 # asm 2: shld $13,<rt4=%rbx,<mulr41=%rbp shld $13,%rbx,%rbp # qhasm: rt4 &= mulredmask # asm 1: and <mulredmask=int64#2,<rt4=int64#14 # asm 2: and <mulredmask=%rsi,<rt4=%rbx and %rsi,%rbx # qhasm: rt4 += mulr31 # asm 1: add <mulr31=int64#13,<rt4=int64#14 # asm 2: add <mulr31=%r15,<rt4=%rbx add %r15,%rbx # qhasm: mulr41 = mulr41 * 19 # asm 1: imulq $19,<mulr41=int64#15,>mulr41=int64#3 # asm 2: imulq $19,<mulr41=%rbp,>mulr41=%rdx imulq $19,%rbp,%rdx # qhasm: rt0 += mulr41 # asm 1: add <mulr41=int64#3,<rt0=int64#5 # asm 2: add <mulr41=%rdx,<rt0=%r8 add %rdx,%r8 # qhasm: mult = rt0 # asm 1: mov <rt0=int64#5,>mult=int64#3 # asm 2: mov <rt0=%r8,>mult=%rdx mov %r8,%rdx # qhasm: (uint64) mult >>= 51 # asm 1: shr $51,<mult=int64#3 # asm 2: shr $51,<mult=%rdx shr $51,%rdx # qhasm: mult += rt1 # asm 1: add <rt1=int64#8,<mult=int64#3 # asm 2: add <rt1=%r10,<mult=%rdx add %r10,%rdx # qhasm: rt1 = mult # asm 1: mov <mult=int64#3,>rt1=int64#4 # asm 2: mov <mult=%rdx,>rt1=%rcx mov %rdx,%rcx # qhasm: (uint64) mult >>= 51 # asm 1: shr $51,<mult=int64#3 # asm 2: shr $51,<mult=%rdx shr $51,%rdx # qhasm: rt0 &= mulredmask # asm 1: and <mulredmask=int64#2,<rt0=int64#5 # asm 2: and <mulredmask=%rsi,<rt0=%r8 and %rsi,%r8 # qhasm: mult += rt2 # asm 1: add <rt2=int64#10,<mult=int64#3 # asm 2: add <rt2=%r12,<mult=%rdx add %r12,%rdx # qhasm: rt2 = mult # asm 1: mov <mult=int64#3,>rt2=int64#6 # asm 2: mov <mult=%rdx,>rt2=%r9 mov %rdx,%r9 # qhasm: (uint64) mult >>= 51 # asm 1: shr $51,<mult=int64#3 # asm 2: shr $51,<mult=%rdx shr $51,%rdx # qhasm: rt1 &= mulredmask # asm 1: and <mulredmask=int64#2,<rt1=int64#4 # asm 2: and <mulredmask=%rsi,<rt1=%rcx and %rsi,%rcx # qhasm: mult += rt3 # asm 1: add <rt3=int64#12,<mult=int64#3 # asm 2: add <rt3=%r14,<mult=%rdx add %r14,%rdx # qhasm: rt3 = mult # asm 1: mov <mult=int64#3,>rt3=int64#7 # asm 2: mov <mult=%rdx,>rt3=%rax mov %rdx,%rax # qhasm: (uint64) mult >>= 51 # asm 1: shr $51,<mult=int64#3 # asm 2: shr $51,<mult=%rdx shr $51,%rdx # qhasm: rt2 &= mulredmask # asm 1: and <mulredmask=int64#2,<rt2=int64#6 # asm 2: and <mulredmask=%rsi,<rt2=%r9 and %rsi,%r9 # qhasm: mult += rt4 # asm 1: add <rt4=int64#14,<mult=int64#3 # asm 2: add <rt4=%rbx,<mult=%rdx add %rbx,%rdx # qhasm: rt4 = mult # asm 1: mov <mult=int64#3,>rt4=int64#8 # asm 2: mov <mult=%rdx,>rt4=%r10 mov %rdx,%r10 # qhasm: (uint64) mult >>= 51 # asm 1: shr $51,<mult=int64#3 # asm 2: shr $51,<mult=%rdx shr $51,%rdx # qhasm: rt3 &= mulredmask # asm 1: and <mulredmask=int64#2,<rt3=int64#7 # asm 2: and <mulredmask=%rsi,<rt3=%rax and %rsi,%rax # qhasm: mult *= 19 # asm 1: imulq $19,<mult=int64#3,>mult=int64#3 # asm 2: imulq $19,<mult=%rdx,>mult=%rdx imulq $19,%rdx,%rdx # qhasm: rt0 += mult # asm 1: add <mult=int64#3,<rt0=int64#5 # asm 2: add <mult=%rdx,<rt0=%r8 add %rdx,%r8 # qhasm: rt4 &= mulredmask # asm 1: and <mulredmask=int64#2,<rt4=int64#8 # asm 2: and <mulredmask=%rsi,<rt4=%r10 and %rsi,%r10 # qhasm: rt0 += rt0 # asm 1: add <rt0=int64#5,<rt0=int64#5 # asm 2: add <rt0=%r8,<rt0=%r8 add %r8,%r8 # qhasm: rt1 += rt1 # asm 1: add <rt1=int64#4,<rt1=int64#4 # asm 2: add <rt1=%rcx,<rt1=%rcx add %rcx,%rcx # qhasm: rt2 += rt2 # asm 1: add <rt2=int64#6,<rt2=int64#6 # asm 2: add <rt2=%r9,<rt2=%r9 add %r9,%r9 # qhasm: rt3 += rt3 # asm 1: add <rt3=int64#7,<rt3=int64#7 # asm 2: add <rt3=%rax,<rt3=%rax add %rax,%rax # qhasm: rt4 += rt4 # asm 1: add <rt4=int64#8,<rt4=int64#8 # asm 2: add <rt4=%r10,<rt4=%r10 add %r10,%r10 # qhasm: rz0 = rt0 # asm 1: mov <rt0=int64#5,>rz0=int64#2 # asm 2: mov <rt0=%r8,>rz0=%rsi mov %r8,%rsi # qhasm: rz1 = rt1 # asm 1: mov <rt1=int64#4,>rz1=int64#3 # asm 2: mov <rt1=%rcx,>rz1=%rdx mov %rcx,%rdx # qhasm: rz2 = rt2 # asm 1: mov <rt2=int64#6,>rz2=int64#9 # asm 2: mov <rt2=%r9,>rz2=%r11 mov %r9,%r11 # qhasm: rz3 = rt3 # asm 1: mov <rt3=int64#7,>rz3=int64#10 # asm 2: mov <rt3=%rax,>rz3=%r12 mov %rax,%r12 # qhasm: rz4 = rt4 # asm 1: mov <rt4=int64#8,>rz4=int64#11 # asm 2: mov <rt4=%r10,>rz4=%r13 mov %r10,%r13 # qhasm: rt0 += *(uint64 *)&CRYPTO_SHARED_NAMESPACE(2P0) # asm 1: add CRYPTO_SHARED_NAMESPACE(2P0),<rt0=int64#5 # asm 2: add CRYPTO_SHARED_NAMESPACE(2P0),<rt0=%r8 add CRYPTO_SHARED_NAMESPACE(2P0)(%rip),%r8 # qhasm: rt1 += *(uint64 *)&CRYPTO_SHARED_NAMESPACE(2P1234) # asm 1: add CRYPTO_SHARED_NAMESPACE(2P1234),<rt1=int64#4 # asm 2: add CRYPTO_SHARED_NAMESPACE(2P1234),<rt1=%rcx add CRYPTO_SHARED_NAMESPACE(2P1234)(%rip),%rcx # qhasm: rt2 += *(uint64 *)&CRYPTO_SHARED_NAMESPACE(2P1234) # asm 1: add CRYPTO_SHARED_NAMESPACE(2P1234),<rt2=int64#6 # asm 2: add CRYPTO_SHARED_NAMESPACE(2P1234),<rt2=%r9 add CRYPTO_SHARED_NAMESPACE(2P1234)(%rip),%r9 # qhasm: rt3 += *(uint64 *)&CRYPTO_SHARED_NAMESPACE(2P1234) # asm 1: add CRYPTO_SHARED_NAMESPACE(2P1234),<rt3=int64#7 # asm 2: add CRYPTO_SHARED_NAMESPACE(2P1234),<rt3=%rax add CRYPTO_SHARED_NAMESPACE(2P1234)(%rip),%rax # qhasm: rt4 += *(uint64 *)&CRYPTO_SHARED_NAMESPACE(2P1234) # asm 1: add CRYPTO_SHARED_NAMESPACE(2P1234),<rt4=int64#8 # asm 2: add CRYPTO_SHARED_NAMESPACE(2P1234),<rt4=%r10 add CRYPTO_SHARED_NAMESPACE(2P1234)(%rip),%r10 # qhasm: rz0 += c0_stack # asm 1: addq <c0_stack=stack64#8,<rz0=int64#2 # asm 2: addq <c0_stack=56(%rsp),<rz0=%rsi addq 56(%rsp),%rsi # qhasm: rz1 += c1_stack # asm 1: addq <c1_stack=stack64#9,<rz1=int64#3 # asm 2: addq <c1_stack=64(%rsp),<rz1=%rdx addq 64(%rsp),%rdx # qhasm: rz2 += c2_stack # asm 1: addq <c2_stack=stack64#10,<rz2=int64#9 # asm 2: addq <c2_stack=72(%rsp),<rz2=%r11 addq 72(%rsp),%r11 # qhasm: rz3 += c3_stack # asm 1: addq <c3_stack=stack64#11,<rz3=int64#10 # asm 2: addq <c3_stack=80(%rsp),<rz3=%r12 addq 80(%rsp),%r12 # qhasm: rz4 += c4_stack # asm 1: addq <c4_stack=stack64#12,<rz4=int64#11 # asm 2: addq <c4_stack=88(%rsp),<rz4=%r13 addq 88(%rsp),%r13 # qhasm: rt0 -= c0_stack # asm 1: subq <c0_stack=stack64#8,<rt0=int64#5 # asm 2: subq <c0_stack=56(%rsp),<rt0=%r8 subq 56(%rsp),%r8 # qhasm: rt1 -= c1_stack # asm 1: subq <c1_stack=stack64#9,<rt1=int64#4 # asm 2: subq <c1_stack=64(%rsp),<rt1=%rcx subq 64(%rsp),%rcx # qhasm: rt2 -= c2_stack # asm 1: subq <c2_stack=stack64#10,<rt2=int64#6 # asm 2: subq <c2_stack=72(%rsp),<rt2=%r9 subq 72(%rsp),%r9 # qhasm: rt3 -= c3_stack # asm 1: subq <c3_stack=stack64#11,<rt3=int64#7 # asm 2: subq <c3_stack=80(%rsp),<rt3=%rax subq 80(%rsp),%rax # qhasm: rt4 -= c4_stack # asm 1: subq <c4_stack=stack64#12,<rt4=int64#8 # asm 2: subq <c4_stack=88(%rsp),<rt4=%r10 subq 88(%rsp),%r10 # qhasm: *(uint64 *)(rp + 40) = rz0 # asm 1: movq <rz0=int64#2,40(<rp=int64#1) # asm 2: movq <rz0=%rsi,40(<rp=%rdi) movq %rsi,40(%rdi) # qhasm: *(uint64 *)(rp + 48) = rz1 # asm 1: movq <rz1=int64#3,48(<rp=int64#1) # asm 2: movq <rz1=%rdx,48(<rp=%rdi) movq %rdx,48(%rdi) # qhasm: *(uint64 *)(rp + 56) = rz2 # asm 1: movq <rz2=int64#9,56(<rp=int64#1) # asm 2: movq <rz2=%r11,56(<rp=%rdi) movq %r11,56(%rdi) # qhasm: *(uint64 *)(rp + 64) = rz3 # asm 1: movq <rz3=int64#10,64(<rp=int64#1) # asm 2: movq <rz3=%r12,64(<rp=%rdi) movq %r12,64(%rdi) # qhasm: *(uint64 *)(rp + 72) = rz4 # asm 1: movq <rz4=int64#11,72(<rp=int64#1) # asm 2: movq <rz4=%r13,72(<rp=%rdi) movq %r13,72(%rdi) # qhasm: *(uint64 *)(rp + 120) = rt0 # asm 1: movq <rt0=int64#5,120(<rp=int64#1) # asm 2: movq <rt0=%r8,120(<rp=%rdi) movq %r8,120(%rdi) # qhasm: *(uint64 *)(rp + 128) = rt1 # asm 1: movq <rt1=int64#4,128(<rp=int64#1) # asm 2: movq <rt1=%rcx,128(<rp=%rdi) movq %rcx,128(%rdi) # qhasm: *(uint64 *)(rp + 136) = rt2 # asm 1: movq <rt2=int64#6,136(<rp=int64#1) # asm 2: movq <rt2=%r9,136(<rp=%rdi) movq %r9,136(%rdi) # qhasm: *(uint64 *)(rp + 144) = rt3 # asm 1: movq <rt3=int64#7,144(<rp=int64#1) # asm 2: movq <rt3=%rax,144(<rp=%rdi) movq %rax,144(%rdi) # qhasm: *(uint64 *)(rp + 152) = rt4 # asm 1: movq <rt4=int64#8,152(<rp=int64#1) # asm 2: movq <rt4=%r10,152(<rp=%rdi) movq %r10,152(%rdi) # qhasm: caller1 = caller1_stack # asm 1: movq <caller1_stack=stack64#1,>caller1=int64#9 # asm 2: movq <caller1_stack=0(%rsp),>caller1=%r11 movq 0(%rsp),%r11 # qhasm: caller2 = caller2_stack # asm 1: movq <caller2_stack=stack64#2,>caller2=int64#10 # asm 2: movq <caller2_stack=8(%rsp),>caller2=%r12 movq 8(%rsp),%r12 # qhasm: caller3 = caller3_stack # asm 1: movq <caller3_stack=stack64#3,>caller3=int64#11 # asm 2: movq <caller3_stack=16(%rsp),>caller3=%r13 movq 16(%rsp),%r13 # qhasm: caller4 = caller4_stack # asm 1: movq <caller4_stack=stack64#4,>caller4=int64#12 # asm 2: movq <caller4_stack=24(%rsp),>caller4=%r14 movq 24(%rsp),%r14 # qhasm: caller5 = caller5_stack # asm 1: movq <caller5_stack=stack64#5,>caller5=int64#13 # asm 2: movq <caller5_stack=32(%rsp),>caller5=%r15 movq 32(%rsp),%r15 # qhasm: caller6 = caller6_stack # asm 1: movq <caller6_stack=stack64#6,>caller6=int64#14 # asm 2: movq <caller6_stack=40(%rsp),>caller6=%rbx movq 40(%rsp),%rbx # qhasm: caller7 = caller7_stack # asm 1: movq <caller7_stack=stack64#7,>caller7=int64#15 # asm 2: movq <caller7_stack=48(%rsp),>caller7=%rbp movq 48(%rsp),%rbp # qhasm: leave add %r11,%rsp mov %rdi,%rax mov %rsi,%rdx ret .section .note.GNU-stack,"",@progbits