diff --git a/.github/workflows/riscv-rt.yaml b/.github/workflows/riscv-rt.yaml
index 8e716b52..0c20eb31 100644
--- a/.github/workflows/riscv-rt.yaml
+++ b/.github/workflows/riscv-rt.yaml
@@ -1,6 +1,6 @@
 on:
   push:
-    branches: [ master ]
+    branches: [ master, riscv-rt-asm ]
   pull_request:
   merge_group:
 
diff --git a/riscv-rt/CHANGELOG.md b/riscv-rt/CHANGELOG.md
index ecbb8a5d..b206df32 100644
--- a/riscv-rt/CHANGELOG.md
+++ b/riscv-rt/CHANGELOG.md
@@ -18,6 +18,7 @@ and this project adheres to [Semantic Versioning](http://semver.org/).
 
 ### Changed
 
+- Removed _start_rust. Now, assembly directly jumps to main
 - Removed U-mode interrupts to align with latest RISC-V specification
 - Changed `Vector` union. Now, it uses `Option<fn>`, which is more idiomatic in Rust
 - Removed riscv-target dependency for build
diff --git a/riscv-rt/macros/src/lib.rs b/riscv-rt/macros/src/lib.rs
index b48cc111..14021902 100644
--- a/riscv-rt/macros/src/lib.rs
+++ b/riscv-rt/macros/src/lib.rs
@@ -212,7 +212,8 @@ pub fn pre_init(args: TokenStream, input: TokenStream) -> TokenStream {
 
 struct AsmLoopArgs {
     asm_template: String,
-    count: usize,
+    count_from: usize,
+    count_to: usize,
 }
 
 impl Parse for AsmLoopArgs {
@@ -220,24 +221,35 @@ impl Parse for AsmLoopArgs {
         let template: LitStr = input.parse().unwrap();
         _ = input.parse::<Token![,]>().unwrap();
         let count: LitInt = input.parse().unwrap();
-
-        Ok(Self {
-            asm_template: template.value(),
-            count: count.base10_parse().unwrap(),
-        })
+        if input.parse::<Token![,]>().is_ok() {
+            let count_to: LitInt = input.parse().unwrap();
+            Ok(Self {
+                asm_template: template.value(),
+                count_from: count.base10_parse().unwrap(),
+                count_to: count_to.base10_parse().unwrap(),
+            })
+        } else {
+            Ok(Self {
+                asm_template: template.value(),
+                count_from: 0,
+                count_to: count.base10_parse().unwrap(),
+            })
+        }
     }
 }
 
 /// Loops an asm expression n times.
 ///
-/// `loop_asm!` takes 2 arguments, the first is a string literal and the second is a number literal
-/// See [the formatting syntax documentation in `std::fmt`](../std/fmt/index.html)
-/// for details.
+/// `loop_asm!` takes 2 or 3 arguments, the first is a string literal and the rest are a number literal
+/// See [the formatting syntax documentation in `std::fmt`](../std/fmt/index.html) for details.
 ///
 /// Argument 1 is an assembly expression, all "{}" in this assembly expression will be replaced with the
 /// current loop index.
 ///
-/// Argument 2 is the number of loops to do with the provided expression.
+/// If 2 arguments are provided, the loop will start at 0 and end at the number provided in argument 2.
+///
+/// If 3 arguments are provided, the loop will start at the number provided in argument 2 and end at
+/// the number provided in argument 3.
 ///
 /// # Examples
 ///
@@ -245,13 +257,14 @@ impl Parse for AsmLoopArgs {
 /// # use riscv_rt_macros::loop_asm;
 /// unsafe {
 ///     loop_asm!("fmv.w.x f{}, x0", 32); // => core::arch::asm!("fmv.w.x f0, x0") ... core::arch::asm!("fmv.w.x f31, x0")
+///     loop_asm!("fmv.w.x f{}, x0", 1, 32); // => core::arch::asm!("fmv.w.x f1, x0") ... core::arch::asm!("fmv.w.x f31, x0")
 /// }
 /// ```
 #[proc_macro]
 pub fn loop_asm(input: TokenStream) -> TokenStream {
     let args = parse_macro_input!(input as AsmLoopArgs);
 
-    let tokens = (0..args.count)
+    let tokens = (args.count_from..args.count_to)
         .map(|i| {
             let i = i.to_string();
             let asm = args.asm_template.replace("{}", &i);
@@ -261,3 +274,41 @@ pub fn loop_asm(input: TokenStream) -> TokenStream {
         .join("\n");
     tokens.parse().unwrap()
 }
+
+/// Loops a global_asm expression n times.
+///
+/// `loop_global_asm!` takes 2 or 3 arguments, the first is a string literal and the rest are a number literal
+/// See [the formatting syntax documentation in `std::fmt`](../std/fmt/index.html) for details.
+///
+/// Argument 1 is an assembly expression, all "{}" in this assembly expression will be replaced with the
+/// current loop index.
+///
+/// If 2 arguments are provided, the loop will start at 0 and end at the number provided in argument 2.
+///
+/// If 3 arguments are provided, the loop will start at the number provided in argument 2 and end at
+/// the number provided in argument 3.
+///
+/// # Examples
+///
+/// ```
+/// # use riscv_rt_macros::loop_global_asm;
+/// unsafe {
+///     loop_global_asm!("fmv.w.x f{}, x0", 32); // => core::arch::global_asm!("fmv.w.x f0, x0") ... core::arch::global_asm!("fmv.w.x f31, x0")
+///     loop_global_asm!("fmv.w.x f{}, x0", 1, 32); // => core::arch::global_asm!("fmv.w.x f1, x0") ... core::arch::global_asm!("fmv.w.x f31, x0")
+/// }
+/// ```
+#[proc_macro]
+pub fn loop_global_asm(input: TokenStream) -> TokenStream {
+    let args = parse_macro_input!(input as AsmLoopArgs);
+
+    let instructions = (args.count_from..args.count_to)
+        .map(|i| {
+            let i = i.to_string();
+            args.asm_template.replace("{}", &i)
+        })
+        .collect::<Vec<String>>()
+        .join("\n");
+
+    let res = format!("core::arch::global_asm!(\n\"{}\"\n);", instructions);
+    res.parse().unwrap()
+}
diff --git a/riscv-rt/src/asm.rs b/riscv-rt/src/asm.rs
index 0eedebe6..b83d304d 100644
--- a/riscv-rt/src/asm.rs
+++ b/riscv-rt/src/asm.rs
@@ -52,54 +52,36 @@ _abs_start:
     #[cfg(not(feature = "s-mode"))]
     "csrw mie, 0
     csrw mip, 0",
-    "li  x1, 0
-    li  x2, 0
-    li  x3, 0
-    li  x4, 0
-    li  x5, 0
-    li  x6, 0
-    li  x7, 0
-    li  x8, 0
-    li  x9, 0
-    // a0..a2 (x10..x12) skipped
-    li  x13, 0
-    li  x14, 0
-    li  x15, 0
-    li  x16, 0
-    li  x17, 0
-    li  x18, 0
-    li  x19, 0
-    li  x20, 0
-    li  x21, 0
-    li  x22, 0
-    li  x23, 0
-    li  x24, 0
-    li  x25, 0
-    li  x26, 0
-    li  x27, 0
-    li  x28, 0
-    li  x29, 0
-    li  x30, 0
-    li  x31, 0
+);
+
+// ZERO OUT GENERAL-PURPOSE REGISTERS
+riscv_rt_macros::loop_global_asm!("    li x{}, 0", 1, 10);
+// a0..a2 (x10..x12) skipped
+riscv_rt_macros::loop_global_asm!("    li x{}, 0", 13, 32);
 
-    .option push
+// INITIALIZE GLOBAL POINTER
+cfg_global_asm!(
+    ".option push
     .option norelax
     la gp, __global_pointer$
-    .option pop
-    // Allocate stacks",
-    #[cfg(all(not(feature = "single-hart"), feature = "s-mode"))]
+    .option pop",
+);
+
+// INITIALIZE STACK POINTER AND FRAME POINTER
+#[cfg(not(feature = "single-hart"))]
+cfg_global_asm!(
+    #[cfg(feature = "s-mode")]
     "mv t2, a0 // the hartid is passed as parameter by SMODE",
-    #[cfg(all(not(feature = "single-hart"), not(feature = "s-mode")))]
+    #[cfg(not(feature = "s-mode"))]
     "csrr t2, mhartid",
-    #[cfg(not(feature = "single-hart"))]
     "lui t0, %hi(_max_hart_id)
     add t0, t0, %lo(_max_hart_id)
     bgtu t2, t0, abort
     lui t0, %hi(_hart_stack_size)
     add t0, t0, %lo(_hart_stack_size)",
-    #[cfg(all(not(feature = "single-hart"), riscvm))]
+    #[cfg(riscvm)]
     "mul t0, t2, t0",
-    #[cfg(all(not(feature = "single-hart"), not(riscvm)))]
+    #[cfg(not(riscvm))]
     "beqz t2, 2f  // Jump if single-hart
     mv t1, t2
     mv t3, t0
@@ -109,14 +91,109 @@ _abs_start:
     bnez t1, 1b
 2:  ",
     "la t1, _stack_start",
-    #[cfg(not(feature = "single-hart"))]
     "sub t1, t1, t0",
-    "andi sp, t1, -16 // Force 16-byte alignment
-    // Set frame pointer
-    add s0, sp, zero
+);
+cfg_global_asm!(
+    "andi sp, t1, -16 // align stack to 16-bytes
+    add s0, sp, zero",
+);
+
+// STORE A0..A2 IN THE STACK, AS THEY WILL BE NEEDED LATER BY main
+cfg_global_asm!(
+    #[cfg(riscv32)]
+    "addi sp, sp, -4 * 3
+    sw a0, 4 * 0(sp)
+    sw a1, 4 * 1(sp)
+    sw a2, 4 * 2(sp)",
+    #[cfg(riscv64)]
+    "addi sp, sp, -8 * 3
+    sd a0, 8 * 0(sp)
+    sd a1, 8 * 1(sp)
+    sd a2, 8 * 2(sp)",
+);
 
-    jal zero, _start_rust
+// SKIP RAM INITIALIZATION IF CURRENT HART IS NOT THE BOOT HART
+#[cfg(not(feature = "single-hart"))]
+cfg_global_asm!(
+    #[cfg(not(feature = "s-mode"))]
+    "csrr a0, mhartid",
+    "call _mp_hook
+    mv t0, a0
+
+    beqz a0, 4f",
+);
+// IF CURRENT HART IS THE BOOT HART CALL __pre_init AND INITIALIZE RAM
+cfg_global_asm!(
+    "call __pre_init
+    // Copy .data from flash to RAM
+    la t0, _sdata
+    la t2, _edata
+    la t1, _sidata
+    bgeu t0, t2, 2f
+1:  ",
+    #[cfg(target_arch = "riscv32")]
+    "lw t3, 0(t1)
+    addi t1, t1, 4
+    sw t3, 0(t0)
+    addi t0, t0, 4
+    bltu t0, t2, 1b",
+    #[cfg(target_arch = "riscv64")]
+    "ld t3, 0(t1)
+    addi t1, t1, 8
+    sd t3, 0(t0)
+    addi t0, t0, 8
+    bltu t0, t2, 1b",
+    "
+2:  // Zero out .bss
+    la t0, _sbss
+    la t2, _ebss
+    bgeu  t0, t2, 4f
+3:  ",
+    #[cfg(target_arch = "riscv32")]
+    "sw  zero, 0(t0)
+    addi t0, t0, 4
+    bltu t0, t2, 3b",
+    #[cfg(target_arch = "riscv64")]
+    "sd zero, 0(t0)
+    addi t0, t0, 8
+    bltu t0, t2, 3b",
+    "
+4: // RAM initilized",
+);
 
+// INITIALIZE FLOATING POINT UNIT
+#[cfg(any(riscvf, riscvd))]
+cfg_global_asm!(
+    #[cfg(feature = "s-mode")]
+    "csrrc x0, sstatus, 0x4000
+    csrrs x0, sstatus, 0x2000",
+    #[cfg(not(feature = "s-mode"))]
+    "csrrc x0, mstatus, 0x4000
+    csrrs x0, mstatus, 0x2000",
+    "fscsr x0",
+);
+// ZERO OUT FLOATING POINT REGISTERS
+#[cfg(all(riscv32, riscvd))]
+riscv_rt_macros::loop_global_asm!("    fcvt.d.w f{}, x0", 32);
+#[cfg(all(riscv64, riscvd))]
+riscv_rt_macros::loop_global_asm!("    fmv.d.x f{}, x0", 32);
+#[cfg(all(riscvf, not(riscvd)))]
+riscv_rt_macros::loop_global_asm!("    fmv.w.x f{}, x0", 32);
+
+// SET UP INTERRUPTS, RESTORE a0..a2, AND JUMP TO MAIN RUST FUNCTION
+cfg_global_asm!(
+    "call _setup_interrupts",
+    #[cfg(riscv32)]
+    "lw a0, 4 * 0(sp)
+    lw a1, 4 * 1(sp)
+    lw a2, 4 * 2(sp)
+    addi sp, sp, 4 * 3",
+    #[cfg(riscv64)]
+    "ld a0, 8 * 0(sp)
+    ld a1, 8 * 1(sp)
+    ld a2, 8 * 2(sp)
+    addi sp, sp, 8 * 3",
+    "jal zero, main
     .cfi_endproc",
 );
 
diff --git a/riscv-rt/src/lib.rs b/riscv-rt/src/lib.rs
index a7a900eb..242eb664 100644
--- a/riscv-rt/src/lib.rs
+++ b/riscv-rt/src/lib.rs
@@ -404,179 +404,18 @@
 #[cfg(riscv)]
 mod asm;
 
-use core::sync::atomic::{compiler_fence, Ordering};
-
 #[cfg(feature = "s-mode")]
 use riscv::register::{scause as xcause, stvec as xtvec, stvec::TrapMode as xTrapMode};
 
 #[cfg(not(feature = "s-mode"))]
 use riscv::register::{mcause as xcause, mtvec as xtvec, mtvec::TrapMode as xTrapMode};
 
-#[cfg(all(not(feature = "single-hart"), not(feature = "s-mode")))]
-use riscv::register::mhartid;
-
-#[cfg(all(feature = "s-mode", any(riscvf, riscvd)))]
-use riscv::register::sstatus as xstatus;
-
-#[cfg(all(not(feature = "s-mode"), any(riscvf, riscvd)))]
-use riscv::register::mstatus as xstatus;
-
 pub use riscv_rt_macros::{entry, pre_init};
 
 #[export_name = "error: riscv-rt appears more than once in the dependency graph"]
 #[doc(hidden)]
 pub static __ONCE__: () = ();
 
-/// Rust entry point (_start_rust)
-///
-/// Zeros bss section, initializes data section and calls main. This function never returns.
-///
-/// # Safety
-///
-/// This function must be called only from assembly `_start` function.
-/// Do **NOT** call this function directly.
-#[link_section = ".init.rust"]
-#[export_name = "_start_rust"]
-pub unsafe extern "C" fn start_rust(a0: usize, a1: usize, a2: usize) -> ! {
-    #[rustfmt::skip]
-    extern "Rust" {
-        // This symbol will be provided by the user via `#[entry]`
-        fn main(a0: usize, a1: usize, a2: usize) -> !;
-
-        // This symbol will be provided by the user via `#[pre_init]`
-        fn __pre_init();
-
-        fn _setup_interrupts();
-
-        fn _mp_hook(hartid: usize) -> bool;
-    }
-
-    #[cfg(not(feature = "single-hart"))]
-    let run_init = {
-        // sbi passes hartid as first parameter (a0)
-        #[cfg(feature = "s-mode")]
-        let hartid = a0;
-        #[cfg(not(feature = "s-mode"))]
-        let hartid = mhartid::read();
-
-        _mp_hook(hartid)
-    };
-    #[cfg(feature = "single-hart")]
-    let run_init = true;
-
-    if run_init {
-        __pre_init();
-
-        // Initialize RAM
-        // 1. Copy over .data from flash to RAM
-        // 2. Zero out .bss
-
-        #[cfg(target_arch = "riscv32")]
-        core::arch::asm!(
-            "
-                // Copy over .data
-                la      {start},_sdata
-                la      {end},_edata
-                la      {input},_sidata
-
-            	bgeu    {start},{end},2f
-            1:
-            	lw      {a},0({input})
-            	addi    {input},{input},4
-            	sw      {a},0({start})
-            	addi    {start},{start},4
-            	bltu    {start},{end},1b
-
-            2:
-                li      {a},0
-                li      {input},0
-
-                // Zero out .bss
-            	la      {start},_sbss
-            	la      {end},_ebss
-
-            	bgeu    {start},{end},3f
-            2:
-            	sw      zero,0({start})
-            	addi    {start},{start},4
-            	bltu    {start},{end},2b
-
-            3:
-                li      {start},0
-                li      {end},0
-            ",
-            start = out(reg) _,
-            end = out(reg) _,
-            input = out(reg) _,
-            a = out(reg) _,
-        );
-
-        #[cfg(target_arch = "riscv64")]
-        core::arch::asm!(
-            "
-                // Copy over .data
-                la      {start},_sdata
-                la      {end},_edata
-                la      {input},_sidata
-
-                bgeu    {start},{end},2f
-
-            1: // .data Main Loop
-            	ld      {a},0({input})
-            	addi    {input},{input},8
-            	sd      {a},0({start})
-            	addi    {start},{start},8
-            	bltu    {start},{end},1b
-
-            2: // .data zero registers
-                li      {a},0
-                li      {input},0
-
-            	la      {start},_sbss
-            	la      {end},_ebss
-
-                bgeu    {start},{end},4f
-
-            3: // .bss main loop
-            	sd      zero,0({start})
-            	addi    {start},{start},8
-            	bltu    {start},{end},3b
-
-            4: // .bss zero registers
-                //    Zero out used registers
-                li      {start},0
-                li      {end},0
-        ",
-            start = out(reg) _,
-            end = out(reg) _,
-            input = out(reg) _,
-            a = out(reg) _,
-        );
-
-        compiler_fence(Ordering::SeqCst);
-    }
-
-    #[cfg(any(riscvf, riscvd))]
-    {
-        xstatus::set_fs(xstatus::FS::Initial); // Enable fpu in xstatus
-        core::arch::asm!("fscsr x0"); // Zero out fcsr register csrrw x0, fcsr, x0
-
-        // Zero out floating point registers
-        #[cfg(all(target_arch = "riscv32", riscvd))]
-        riscv_rt_macros::loop_asm!("fcvt.d.w f{}, x0", 32);
-
-        #[cfg(all(target_arch = "riscv64", riscvd))]
-        riscv_rt_macros::loop_asm!("fmv.d.x f{}, x0", 32);
-
-        #[cfg(not(riscvd))]
-        riscv_rt_macros::loop_asm!("fmv.w.x f{}, x0", 32);
-    }
-
-    _setup_interrupts();
-
-    main(a0, a1, a2);
-}
-
 /// Registers saved in trap handler
 #[allow(missing_docs)]
 #[repr(C)]