diff --git a/ptx/src/pass/insert_post_saturation.rs b/ptx/src/pass/insert_post_saturation.rs
index 525ae15..620e46b 100644
--- a/ptx/src/pass/insert_post_saturation.rs
+++ b/ptx/src/pass/insert_post_saturation.rs
@@ -164,7 +164,6 @@ fn run_instruction<'input>(
         | ast::Instruction::Or { .. }
         | ast::Instruction::Popc { .. }
         | ast::Instruction::Prmt { .. }
-        | ast::Instruction::PrmtSlow { .. }
         | ast::Instruction::Rcp { .. }
         | ast::Instruction::Rem { .. }
         | ast::Instruction::Ret { .. }
diff --git a/ptx/src/pass/instruction_mode_to_global_mode/mod.rs b/ptx/src/pass/instruction_mode_to_global_mode/mod.rs
index d365e29..12851a6 100644
--- a/ptx/src/pass/instruction_mode_to_global_mode/mod.rs
+++ b/ptx/src/pass/instruction_mode_to_global_mode/mod.rs
@@ -1818,7 +1818,6 @@ fn get_modes<T: ast::Operand>(inst: &ast::Instruction<T>) -> InstructionModes {
         | ast::Instruction::Mov { .. }
         | ast::Instruction::Ld { .. }
         | ast::Instruction::St { .. }
-        | ast::Instruction::PrmtSlow { .. }
         | ast::Instruction::Prmt { .. }
         | ast::Instruction::Activemask { .. }
         | ast::Instruction::Membar { .. }
diff --git a/ptx/src/pass/llvm/emit.rs b/ptx/src/pass/llvm/emit.rs
index d9ec666..e449b01 100644
--- a/ptx/src/pass/llvm/emit.rs
+++ b/ptx/src/pass/llvm/emit.rs
@@ -511,7 +511,6 @@ impl<'a> MethodEmitContext<'a> {
             ast::Instruction::Xor { data, arguments } => self.emit_xor(data, arguments),
             ast::Instruction::Rem { data, arguments } => self.emit_rem(data, arguments),
             ast::Instruction::BarWarp { .. } => self.emit_bar_warp(),
-            ast::Instruction::Prmt { data, arguments } => self.emit_prmt(data, arguments),
             ast::Instruction::Membar { data } => self.emit_membar(data),
             ast::Instruction::Trap {} => self.emit_trap(),
             ast::Instruction::Tanh { data, arguments } => self.emit_tanh(data, arguments),
@@ -532,7 +531,7 @@ impl<'a> MethodEmitContext<'a> {
             | ast::Instruction::ReduxSync { .. }
             | ast::Instruction::LdMatrix { .. }
             | ast::Instruction::Mma { .. }
-            | ast::Instruction::PrmtSlow { .. } => return Err(error_unreachable()),
+            | ast::Instruction::Prmt { .. } => return Err(error_unreachable()),
         }
     }
 
@@ -2446,48 +2445,6 @@ impl<'a> MethodEmitContext<'a> {
         Ok(())
     }
 
-    fn emit_prmt(
-        &mut self,
-        control: u16,
-        arguments: ptx_parser::PrmtArgs<SpirvWord>,
-    ) -> Result<(), TranslateError> {
-        let components = [
-            (control >> 0) & 0b1111,
-            (control >> 4) & 0b1111,
-            (control >> 8) & 0b1111,
-            (control >> 12) & 0b1111,
-        ];
-        if components.iter().any(|&c| c > 7) {
-            return Err(error_todo());
-        }
-        let u32_type = get_scalar_type(self.context, ast::ScalarType::U32);
-        let v4u8_type = get_type(self.context, &ast::Type::Vector(4, ast::ScalarType::U8))?;
-        let mut components = [
-            unsafe { LLVMConstInt(u32_type, components[0] as _, 0) },
-            unsafe { LLVMConstInt(u32_type, components[1] as _, 0) },
-            unsafe { LLVMConstInt(u32_type, components[2] as _, 0) },
-            unsafe { LLVMConstInt(u32_type, components[3] as _, 0) },
-        ];
-        let components_indices =
-            unsafe { LLVMConstVector(components.as_mut_ptr(), components.len() as u32) };
-        let src1 = self.resolver.value(arguments.src1)?;
-        let src1_vector =
-            unsafe { LLVMBuildBitCast(self.builder, src1, v4u8_type, LLVM_UNNAMED.as_ptr()) };
-        let src2 = self.resolver.value(arguments.src2)?;
-        let src2_vector =
-            unsafe { LLVMBuildBitCast(self.builder, src2, v4u8_type, LLVM_UNNAMED.as_ptr()) };
-        self.resolver.with_result(arguments.dst, |dst| unsafe {
-            LLVMBuildShuffleVector(
-                self.builder,
-                src1_vector,
-                src2_vector,
-                components_indices,
-                dst,
-            )
-        });
-        Ok(())
-    }
-
     fn emit_abs(
         &mut self,
         data: ast::TypeFtz,
diff --git a/ptx/src/pass/replace_instructions_with_functions.rs b/ptx/src/pass/replace_instructions_with_functions.rs
index 2304538..a0198d1 100644
--- a/ptx/src/pass/replace_instructions_with_functions.rs
+++ b/ptx/src/pass/replace_instructions_with_functions.rs
@@ -519,7 +519,7 @@ fn run_instruction<'input>(
                 i,
             )?
         }
-        i @ ptx_parser::Instruction::PrmtSlow { .. } => {
+        i @ ptx_parser::Instruction::Prmt { .. } => {
             to_call(resolver, fn_declarations, "prmt_b32".into(), i)?
         }
         i => i,
diff --git a/ptx/src/test/ll/prmt.ll b/ptx/src/test/ll/prmt.ll
index 7753f5c..933229d 100644
--- a/ptx/src/test/ll/prmt.ll
+++ b/ptx/src/test/ll/prmt.ll
@@ -1,38 +1,39 @@
-define amdgpu_kernel void @prmt(ptr addrspace(4) byref(i64) %"36", ptr addrspace(4) byref(i64) %"37") #0 {
-  %"38" = alloca i64, align 8, addrspace(5)
+declare hidden i32 @__zluda_ptx_impl_prmt_b32(i32, i32, i32) #0
+
+define amdgpu_kernel void @prmt(ptr addrspace(4) byref(i64) %"37", ptr addrspace(4) byref(i64) %"38") #1 {
   %"39" = alloca i64, align 8, addrspace(5)
-  %"40" = alloca i32, align 4, addrspace(5)
+  %"40" = alloca i64, align 8, addrspace(5)
   %"41" = alloca i32, align 4, addrspace(5)
+  %"42" = alloca i32, align 4, addrspace(5)
   br label %1
 
 1:                                                ; preds = %0
-  br label %"35"
+  br label %"36"
 
-"35":                                             ; preds = %1
-  %"42" = load i64, ptr addrspace(4) %"36", align 8
-  store i64 %"42", ptr addrspace(5) %"38", align 8
+"36":                                             ; preds = %1
   %"43" = load i64, ptr addrspace(4) %"37", align 8
   store i64 %"43", ptr addrspace(5) %"39", align 8
-  %"45" = load i64, ptr addrspace(5) %"38", align 8
-  %"53" = inttoptr i64 %"45" to ptr
-  %"44" = load i32, ptr %"53", align 4
-  store i32 %"44", ptr addrspace(5) %"40", align 4
-  %"46" = load i64, ptr addrspace(5) %"38", align 8
+  %"44" = load i64, ptr addrspace(4) %"38", align 8
+  store i64 %"44", ptr addrspace(5) %"40", align 8
+  %"46" = load i64, ptr addrspace(5) %"39", align 8
   %"54" = inttoptr i64 %"46" to ptr
-  %"34" = getelementptr inbounds i8, ptr %"54", i64 4
-  %"47" = load i32, ptr %"34", align 4
-  store i32 %"47", ptr addrspace(5) %"41", align 4
-  %"49" = load i32, ptr addrspace(5) %"40", align 4
+  %"45" = load i32, ptr %"54", align 4
+  store i32 %"45", ptr addrspace(5) %"41", align 4
+  %"47" = load i64, ptr addrspace(5) %"39", align 8
+  %"55" = inttoptr i64 %"47" to ptr
+  %"34" = getelementptr inbounds i8, ptr %"55", i64 4
+  %"48" = load i32, ptr %"34", align 4
+  store i32 %"48", ptr addrspace(5) %"42", align 4
   %"50" = load i32, ptr addrspace(5) %"41", align 4
-  %2 = bitcast i32 %"49" to <4 x i8>
-  %3 = bitcast i32 %"50" to <4 x i8>
-  %"55" = shufflevector <4 x i8> %2, <4 x i8> %3, <4 x i32> <i32 4, i32 0, i32 6, i32 7>
-  store <4 x i8> %"55", ptr addrspace(5) %"41", align 4
-  %"51" = load i64, ptr addrspace(5) %"39", align 8
-  %"52" = load i32, ptr addrspace(5) %"41", align 4
-  %"58" = inttoptr i64 %"51" to ptr
-  store i32 %"52", ptr %"58", align 4
+  %"51" = load i32, ptr addrspace(5) %"42", align 4
+  %"56" = call i32 @__zluda_ptx_impl_prmt_b32(i32 %"50", i32 %"51", i32 30212)
+  store i32 %"56", ptr addrspace(5) %"42", align 4
+  %"52" = load i64, ptr addrspace(5) %"40", align 8
+  %"53" = load i32, ptr addrspace(5) %"42", align 4
+  %"59" = inttoptr i64 %"52" to ptr
+  store i32 %"53", ptr %"59", align 4
   ret void
 }
 
-attributes #0 = { "amdgpu-ieee"="false" "amdgpu-unsafe-fp-atomics"="true" "denormal-fp-math"="preserve-sign" "denormal-fp-math-f32"="preserve-sign" "no-trapping-math"="true" "uniform-work-group-size"="true" }
\ No newline at end of file
+attributes #0 = { "amdgpu-ieee"="false" "amdgpu-unsafe-fp-atomics"="true" "denormal-fp-math"="dynamic" "denormal-fp-math-f32"="dynamic" "no-trapping-math"="true" "uniform-work-group-size"="true" }
+attributes #1 = { "amdgpu-ieee"="false" "amdgpu-unsafe-fp-atomics"="true" "denormal-fp-math"="preserve-sign" "denormal-fp-math-f32"="preserve-sign" "no-trapping-math"="true" "uniform-work-group-size"="true" }
\ No newline at end of file
diff --git a/ptx/src/test/ll/prmt_slow.ll b/ptx/src/test/ll/prmt_slow.ll
index 3943afa..f178332 100644
--- a/ptx/src/test/ll/prmt_slow.ll
+++ b/ptx/src/test/ll/prmt_slow.ll
@@ -33,14 +33,14 @@ define amdgpu_kernel void @prmt_slow(ptr addrspace(4) byref(i64) %"39", ptr addr
   %"55" = load i32, ptr addrspace(5) %"43", align 4
   %"56" = load i32, ptr addrspace(5) %"44", align 4
   %"57" = load i32, ptr addrspace(5) %"45", align 4
-  %"54" = call i32 @__zluda_ptx_impl_prmt_b32(i32 %"55", i32 %"56", i32 %"57")
-  store i32 %"54", ptr addrspace(5) %"44", align 4
+  %"63" = call i32 @__zluda_ptx_impl_prmt_b32(i32 %"55", i32 %"56", i32 %"57")
+  store i32 %"63", ptr addrspace(5) %"44", align 4
   %"58" = load i64, ptr addrspace(5) %"42", align 8
   %"59" = load i32, ptr addrspace(5) %"44", align 4
-  %"63" = inttoptr i64 %"58" to ptr
-  store i32 %"59", ptr %"63", align 4
+  %"67" = inttoptr i64 %"58" to ptr
+  store i32 %"59", ptr %"67", align 4
   ret void
 }
 
 attributes #0 = { "amdgpu-ieee"="false" "amdgpu-unsafe-fp-atomics"="true" "denormal-fp-math"="dynamic" "denormal-fp-math-f32"="dynamic" "no-trapping-math"="true" "uniform-work-group-size"="true" }
-attributes #1 = { "amdgpu-ieee"="false" "amdgpu-unsafe-fp-atomics"="true" "denormal-fp-math"="preserve-sign" "denormal-fp-math-f32"="preserve-sign" "no-trapping-math"="true" "uniform-work-group-size"="true" }
+attributes #1 = { "amdgpu-ieee"="false" "amdgpu-unsafe-fp-atomics"="true" "denormal-fp-math"="preserve-sign" "denormal-fp-math-f32"="preserve-sign" "no-trapping-math"="true" "uniform-work-group-size"="true" }
\ No newline at end of file
diff --git a/ptx_parser/src/ast.rs b/ptx_parser/src/ast.rs
index 1bc622c..9ab1037 100644
--- a/ptx_parser/src/ast.rs
+++ b/ptx_parser/src/ast.rs
@@ -432,15 +432,6 @@ ptx_parser_macros::generate_instruction_type!(
         },
         Prmt {
             type: Type::Scalar(ScalarType::B32),
-            data: u16,
-            arguments<T>: {
-                dst: T,
-                src1: T,
-                src2: T
-            }
-        },
-        PrmtSlow {
-            type: Type::Scalar(ScalarType::U32),
             arguments<T>: {
                 dst: T,
                 src1: T,
diff --git a/ptx_parser/src/lib.rs b/ptx_parser/src/lib.rs
index a4f9080..ceffc65 100644
--- a/ptx_parser/src/lib.rs
+++ b/ptx_parser/src/lib.rs
@@ -3659,17 +3659,9 @@ derive_parser!(
     // prmt.b32{.mode}  d, a, b, c;
     // .mode = { .f4e, .b4e, .rc8, .ecl, .ecr, .rc16 };
     prmt.b32  d, a, b, c => {
-        match c {
-            ast::ParsedOperand::Imm(ImmediateValue::S64(control)) => ast::Instruction::Prmt {
-                data: control as u16,
-                arguments: PrmtArgs {
-                    dst: d, src1: a, src2: b
-                }
-            },
-            _ => ast::Instruction::PrmtSlow {
-                arguments: PrmtSlowArgs {
-                    dst: d, src1: a, src2: b, src3: c
-                }
+        ast::Instruction::Prmt {
+            arguments: PrmtArgs {
+                dst: d, src1: a, src2: b, src3: c
             }
         }
     }